@inproceedings{
 title = {AM-MobileNet1D: A Portable Model for Speaker Recognition},
 type = {inproceedings},
 year = {2020},
 keywords = {AM-SincNet,MobileNet,Portable Deep Learning,SincNet,Speaker Identification},
 id = {9b41e4e6-a3df-3257-a8f3-f16d8ef648cc},
 created = {2020-10-30T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2020-11-02T18:24:48.502Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2020 IEEE. Speaker Recognition and Speaker Identification are challenging tasks with essential applications such as automation, authentication, and security. Deep learning approaches like SincNet and AM-SincNet presented great results on these tasks. The promising performance took these models to real-world applications that becoming fundamentally end-user driven and mostly mobile. The mobile computation requires applications with reduced storage size, non-processing and memory intensive and efficient energy-consuming. The deep learning approaches, in contrast, usually are energy expensive, demanding storage, processing power, and memory. To address this demand, we propose a portable model called Additive Margin MobileNet1D (AM-MobileNet1D) to Speaker Identification on mobile devices. We evaluated the proposed approach on TIMIT and MIT datasets obtaining equivalent or better performances concerning the baseline methods. Additionally, the proposed model takes only 11.6 megabytes on disk storage against 91.2 from SincNet and AM-SincNet architectures, making the model seven times faster, with eight times fewer parameters.},
 bibtype = {inproceedings},
 author = {Chagas Nunes, J.A. and Macêdo, D. and Zanchettin, C.},
 doi = {10.1109/IJCNN48605.2020.9207519},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

© 2020 IEEE. Speaker Recognition and Speaker Identification are challenging tasks with essential applications such as automation, authentication, and security. Deep learning approaches like SincNet and AM-SincNet presented great results on these tasks. The promising performance took these models to real-world applications that becoming fundamentally end-user driven and mostly mobile. The mobile computation requires applications with reduced storage size, non-processing and memory intensive and efficient energy-consuming. The deep learning approaches, in contrast, usually are energy expensive, demanding storage, processing power, and memory. To address this demand, we propose a portable model called Additive Margin MobileNet1D (AM-MobileNet1D) to Speaker Identification on mobile devices. We evaluated the proposed approach on TIMIT and MIT datasets obtaining equivalent or better performances concerning the baseline methods. Additionally, the proposed model takes only 11.6 megabytes on disk storage against 91.2 from SincNet and AM-SincNet architectures, making the model seven times faster, with eight times fewer parameters.

@inproceedings{
 title = {A Fast Fully Octave Convolutional Neural Network for Document Image Segmentation},
 type = {inproceedings},
 year = {2020},
 keywords = {Document Segmentation,FOCN,FOCNN,Fully Octave Convolutional Network,Octave Convolution,U-Net},
 id = {b85ad3ca-556a-32e7-8492-8f652f859e94},
 created = {2020-10-30T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2020-11-02T18:51:41.452Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2020 IEEE. The Know Your Customer (KYC) and Anti Money Laundering (AML) are worldwide practices to online customer identification based on personal identification documents, similarity and liveness checking, and proof of address. To answer the basic regulation question: are you whom you say you are? The customer needs to upload valid identification documents (ID). This task imposes some computational challenges since these documents are diverse, may present different and complex backgrounds, some occlusion, partial rotation, poor quality, or damage. Advanced text and document segmentation algorithms were used to process the ID images. In this context, we investigated a method based on U-Net to detect the document edges and text regions in ID images. Besides the promising results on image segmentation, the U-Net based approach is computationally expensive for a real application, since the image segmentation is a customer device task. We propose a model optimization based on Octave Convolutions to qualify the method to situations where storage, processing, and time resources are limited, such as in mobile and robotic applications. We conducted the evaluation experiments in two new datasets CDPhotoDataset and DTDDataset, which are composed of real ID images of Brazilian documents. Our results showed that the proposed models are efficient to document segmentation tasks and portable.},
 bibtype = {inproceedings},
 author = {Das Neves, R.B. and Felipe Vercosa, L. and MacEdo, D. and Dantas Bezerra, B.L. and Zanchettin, C.},
 doi = {10.1109/IJCNN48605.2020.9206711},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

© 2020 IEEE. The Know Your Customer (KYC) and Anti Money Laundering (AML) are worldwide practices to online customer identification based on personal identification documents, similarity and liveness checking, and proof of address. To answer the basic regulation question: are you whom you say you are? The customer needs to upload valid identification documents (ID). This task imposes some computational challenges since these documents are diverse, may present different and complex backgrounds, some occlusion, partial rotation, poor quality, or damage. Advanced text and document segmentation algorithms were used to process the ID images. In this context, we investigated a method based on U-Net to detect the document edges and text regions in ID images. Besides the promising results on image segmentation, the U-Net based approach is computationally expensive for a real application, since the image segmentation is a customer device task. We propose a model optimization based on Octave Convolutions to qualify the method to situations where storage, processing, and time resources are limited, such as in mobile and robotic applications. We conducted the evaluation experiments in two new datasets CDPhotoDataset and DTDDataset, which are composed of real ID images of Brazilian documents. Our results showed that the proposed models are efficient to document segmentation tasks and portable.

@inproceedings{
 title = {Distantly-Supervised Neural Relation Extraction with Side Information using BERT},
 type = {inproceedings},
 year = {2020},
 keywords = {BERT,Distantly-supervised,RESIDE,Relation Extraction},
 id = {29d5ff4c-0948-3180-9ca9-5d5779d7f770},
 created = {2020-10-30T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2020-11-04T01:21:23.459Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2020 IEEE. Relation extraction (RE) consists in categorizing the relationship between entities in a sentence. A recent paradigm to develop relation extractors is Distant Supervision (DS), which allows the automatic creation of new datasets by taking an alignment between a text corpus and a Knowledge Base (KB). KBs can sometimes also provide additional information to the RE task. One of the methods that adopt this strategy is the RESIDE model, which proposes a distantly-supervised neural relation extraction using side information from KBs. Considering that this method outperformed state-of-the-art baselines, in this paper, we propose a related approach to RESIDE also using additional side information, but simplifying the sentence encoding with BERT embeddings. Through experiments, we show the effectiveness of the proposed method in Google Distant Supervision and Riedel datasets concerning the BGWA and RESIDE baseline methods. Although Area Under the Curve is decreased because of unbalanced datasets, P@N results have shown that the use of BERT as sentence encoding allows superior performance to baseline methods.},
 bibtype = {inproceedings},
 author = {Moreira, J. and Oliveira, C. and MacEdo, D. and Zanchettin, C. and Barbosa, L.},
 doi = {10.1109/IJCNN48605.2020.9206648},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

© 2020 IEEE. Relation extraction (RE) consists in categorizing the relationship between entities in a sentence. A recent paradigm to develop relation extractors is Distant Supervision (DS), which allows the automatic creation of new datasets by taking an alignment between a text corpus and a Knowledge Base (KB). KBs can sometimes also provide additional information to the RE task. One of the methods that adopt this strategy is the RESIDE model, which proposes a distantly-supervised neural relation extraction using side information from KBs. Considering that this method outperformed state-of-the-art baselines, in this paper, we propose a related approach to RESIDE also using additional side information, but simplifying the sentence encoding with BERT embeddings. Through experiments, we show the effectiveness of the proposed method in Google Distant Supervision and Riedel datasets concerning the BGWA and RESIDE baseline methods. Although Area Under the Curve is decreased because of unbalanced datasets, P@N results have shown that the use of BERT as sentence encoding allows superior performance to baseline methods.

@inproceedings{
 title = {Squeezed Deep 6DoF Object Detection using Knowledge Distillation},
 type = {inproceedings},
 year = {2020},
 keywords = {6DoF,Knowledge Distillation,Object Detection,Squeezed Network},
 id = {cc5de410-b20f-3bd8-b834-eafd90948eb0},
 created = {2020-10-30T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2020-11-04T07:55:15.937Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2020 IEEE. The detection of objects considering a 6DoF pose is a common requirement to build virtual and augmented reality applications. It is usually a complex task which requires real-time processing and high precision results for adequate user experience. Recently, different deep learning techniques have been proposed to detect objects in 6DoF in RGB images. However, they rely on high complexity networks, requiring a computational power that prevents them from working on mobile devices. In this paper, we propose an approach to reduce the complexity of 6DoF detection networks while maintaining accuracy. We used Knowledge Distillation to teach portables Convolutional Neural Networks (CNN) to learn from a real-time 6DoF detection CNN. The proposed method allows real-time applications using only RGB images while decreasing the hardware requirements. We used the LINEMOD dataset to evaluate the proposed method, and the experimental results show that the proposed method reduces the memory requirement by almost 99% in comparison to the original architecture with the cost of reducing half the accuracy in one of the metrics. Code is available at https://github.com/heitorcfelix/singleshot6Dpose.},
 bibtype = {inproceedings},
 author = {Feliz, H. and Rodrigues, W.M. and Macedo, D. and Simoes, F. and Oliveira, A.L.I. and Teichrieb, V. and Zanchettin, C.},
 doi = {10.1109/IJCNN48605.2020.9207459},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

© 2020 IEEE. The detection of objects considering a 6DoF pose is a common requirement to build virtual and augmented reality applications. It is usually a complex task which requires real-time processing and high precision results for adequate user experience. Recently, different deep learning techniques have been proposed to detect objects in 6DoF in RGB images. However, they rely on high complexity networks, requiring a computational power that prevents them from working on mobile devices. In this paper, we propose an approach to reduce the complexity of 6DoF detection networks while maintaining accuracy. We used Knowledge Distillation to teach portables Convolutional Neural Networks (CNN) to learn from a real-time 6DoF detection CNN. The proposed method allows real-time applications using only RGB images while decreasing the hardware requirements. We used the LINEMOD dataset to evaluate the proposed method, and the experimental results show that the proposed method reduces the memory requirement by almost 99% in comparison to the original architecture with the cost of reducing half the accuracy in one of the metrics. Code is available at https://github.com/heitorcfelix/singleshot6Dpose.

@inproceedings{
 title = {KutralNet: A Portable Deep Learning Model for Fire Recognition},
 type = {inproceedings},
 year = {2020},
 keywords = {deep learning,fire recognition,portable models},
 id = {0cbf3327-68bb-3de0-a2f1-eebec883a99e},
 created = {2020-10-30T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2020-11-04T08:26:12.683Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2020 IEEE. Most of the automatic fire alarm systems detect the fire presence through sensors like thermal, smoke, or flame. One of the new approaches to the problem is the use of images to perform the detection. The image approach is promising since it does not need specific sensors and can be easily embedded in different devices. However, besides the high performance, the computational cost of the used deep learning methods is a challenge to their deployment in portable devices. In this work, we propose a new deep learning architecture that requires fewer floating-point operations (flops) for fire recognition. Additionally, we propose a portable approach for fire recognition and the use of modern techniques such as inverted residual block, convolutions like depth-wise, and octave, to reduce the model's computational cost. The experiments show that our model keeps high accuracy while substantially reducing the number of parameters and flops. One of our models presents 71% fewer parameters than FireNet, while still presenting competitive accuracy and AUROC performance. The proposed methods are evaluated on FireNet and FiSmo datasets. The obtained results are promising for the implementation of the model in a mobile device, considering the reduced number of flops and parameters acquired.},
 bibtype = {inproceedings},
 author = {Ayala, A. and Fernandes, B. and Cruz, F. and MacEdo, D. and Oliveira, A.L.I. and Zanchettin, C.},
 doi = {10.1109/IJCNN48605.2020.9207202},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

© 2020 IEEE. Most of the automatic fire alarm systems detect the fire presence through sensors like thermal, smoke, or flame. One of the new approaches to the problem is the use of images to perform the detection. The image approach is promising since it does not need specific sensors and can be easily embedded in different devices. However, besides the high performance, the computational cost of the used deep learning methods is a challenge to their deployment in portable devices. In this work, we propose a new deep learning architecture that requires fewer floating-point operations (flops) for fire recognition. Additionally, we propose a portable approach for fire recognition and the use of modern techniques such as inverted residual block, convolutions like depth-wise, and octave, to reduce the model's computational cost. The experiments show that our model keeps high accuracy while substantially reducing the number of parameters and flops. One of our models presents 71% fewer parameters than FireNet, while still presenting competitive accuracy and AUROC performance. The proposed methods are evaluated on FireNet and FiSmo datasets. The obtained results are promising for the implementation of the model in a mobile device, considering the reduced number of flops and parameters acquired.

@article{
 title = {Enhancing batch normalized convolutional networks using displaced rectifier linear units: A systematic comparative study},
 type = {article},
 year = {2019},
 keywords = {Activation function,Batch normalization,Comparative study,Convolutional Neural Networks,DReLU,Deep learning},
 volume = {124},
 id = {b0ea06ba-8502-3e85-a7d5-dca22c78ab5f},
 created = {2019-02-14T18:02:01.289Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.289Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2019 Elsevier Ltd A substantial number of expert and intelligent systems rely on deep learning methods to solve problems in areas such as economics, physics, and medicine. Improving the accuracy of the activation functions used by such methods can directly and positively impact the overall performance and quality of the mentioned systems at no cost whatsoever. In this sense, enhancing the design of such theoretical fundamental blocks is of great significance as it immediately impacts a broad range of current and future real-world deep learning based applications. Therefore, in this paper, we turn our attention to the interworking between the activation functions and the batch normalization, which is practically a mandatory technique to train deep networks currently. We propose the activation function Displaced Rectifier Linear Unit (DReLU) by conjecturing that extending the identity function of ReLU to the third quadrant enhances compatibility with batch normalization. Moreover, we used statistical tests to compare the impact of using distinct activation functions (ReLU, LReLU, PReLU, ELU, and DReLU) on the learning speed and test accuracy performance of standardized VGG and Residual Networks state-of-the-art models. These Convolutional Neural Networks were trained on CIFAR-100 and CIFAR-10, the most commonly used deep learning computer vision datasets. The results showed DReLU speeded up learning in all models and datasets. Besides, statistical significant performance assessments (p < 0.05) showed DReLU enhanced the test accuracy presented by ReLU in all scenarios. Furthermore, DReLU showed better test accuracy than any other tested activation function in all experiments with one exception, in which case it presented the second best performance. Therefore, this work demonstrates that it is possible to increase performance replacing ReLU by an enhanced activation function.},
 bibtype = {article},
 author = {Macêdo, D. and Zanchettin, C. and Oliveira, A.L.I. and Ludermir, T.},
 doi = {10.1016/j.eswa.2019.01.066},
 journal = {Expert Systems with Applications}
}

© 2019 Elsevier Ltd A substantial number of expert and intelligent systems rely on deep learning methods to solve problems in areas such as economics, physics, and medicine. Improving the accuracy of the activation functions used by such methods can directly and positively impact the overall performance and quality of the mentioned systems at no cost whatsoever. In this sense, enhancing the design of such theoretical fundamental blocks is of great significance as it immediately impacts a broad range of current and future real-world deep learning based applications. Therefore, in this paper, we turn our attention to the interworking between the activation functions and the batch normalization, which is practically a mandatory technique to train deep networks currently. We propose the activation function Displaced Rectifier Linear Unit (DReLU) by conjecturing that extending the identity function of ReLU to the third quadrant enhances compatibility with batch normalization. Moreover, we used statistical tests to compare the impact of using distinct activation functions (ReLU, LReLU, PReLU, ELU, and DReLU) on the learning speed and test accuracy performance of standardized VGG and Residual Networks state-of-the-art models. These Convolutional Neural Networks were trained on CIFAR-100 and CIFAR-10, the most commonly used deep learning computer vision datasets. The results showed DReLU speeded up learning in all models and datasets. Besides, statistical significant performance assessments (p < 0.05) showed DReLU enhanced the test accuracy presented by ReLU in all scenarios. Furthermore, DReLU showed better test accuracy than any other tested activation function in all experiments with one exception, in which case it presented the second best performance. Therefore, this work demonstrates that it is possible to increase performance replacing ReLU by an enhanced activation function.

@misc{
 title = {Squeezed very deep convolutional neural networks for text classification},
 type = {misc},
 year = {2019},
 source = {arXiv},
 id = {9f590495-c742-3a73-b4f8-bde5b6900ffe},
 created = {2020-10-27T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2020-10-29T03:44:48.806Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Copyright © 2019, arXiv, All rights reserved. Most of the research in convolutional neural networks has focused on increasing network depth to improve accuracy, resulting in a massive number of parameters which restricts the trained network to platforms with memory and processing constraints. We propose to modify the structure of the Very Deep Convolutional Neural Networks (VDCNN) model to fit mobile platforms constraints and keep performance. In this paper, we evaluate the impact of Temporal Depthwise Separable Convolutions and Global Average Pooling in the network parameters, storage size, and latency. The squeezed model (SVDCNN) is between 10x and 20x smaller, depending on the network depth, maintaining a maximum size of 6MB. Regarding accuracy, the network experiences a loss between 0.4% and 1.3% and obtains lower latencies compared to the baseline model.},
 bibtype = {misc},
 author = {Duque, A.B. and Santos, L.L.J. and Macêdo, D. and Zanchettin, C.}
}

Copyright © 2019, arXiv, All rights reserved. Most of the research in convolutional neural networks has focused on increasing network depth to improve accuracy, resulting in a massive number of parameters which restricts the trained network to platforms with memory and processing constraints. We propose to modify the structure of the Very Deep Convolutional Neural Networks (VDCNN) model to fit mobile platforms constraints and keep performance. In this paper, we evaluate the impact of Temporal Depthwise Separable Convolutions and Global Average Pooling in the network parameters, storage size, and latency. The squeezed model (SVDCNN) is between 10x and 20x smaller, depending on the network depth, maintaining a maximum size of 6MB. Regarding accuracy, the network experiences a loss between 0.4% and 1.3% and obtains lower latencies compared to the baseline model.

@misc{
 title = {Additive margin sincnet for speaker recognition},
 type = {misc},
 year = {2019},
 source = {arXiv},
 id = {99f2c571-4bed-3665-a801-844209da1526},
 created = {2020-10-27T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2020-10-29T06:45:17.259Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Copyright © 2019, arXiv, All rights reserved. Speaker Recognition is a challenging task with essential applications such as authentication, automation, and security. The SincNet is a new deep learning based model which has produced promising results to tackle the mentioned task. To train deep learning systems, the loss function is essential to the network performance. The Softmax loss function is a widely used function in deep learning methods, but it is not the best choice for all kind of problems. For distance-based problems, one new Softmax based loss function called Additive Margin Softmax (AM-Softmax) is proving to be a better choice than the traditional Softmax. The AM-Softmax introduces a margin of separation between the classes that forces the samples from the same class to be closer to each other and also maximizes the distance between classes. In this paper, we propose a new approach for speaker recognition systems called AM-SincNet, which is based on the SincNet but uses an improved AM-Softmax layer. The proposed method is evaluated in the TIMIT dataset and obtained an improvement of approximately 40% in the Frame Error Rate compared to SincNet.},
 bibtype = {misc},
 author = {Nunes, J.A.C. and Macêdo, D. and Zanchettin, C.}
}

Copyright © 2019, arXiv, All rights reserved. Speaker Recognition is a challenging task with essential applications such as authentication, automation, and security. The SincNet is a new deep learning based model which has produced promising results to tackle the mentioned task. To train deep learning systems, the loss function is essential to the network performance. The Softmax loss function is a widely used function in deep learning methods, but it is not the best choice for all kind of problems. For distance-based problems, one new Softmax based loss function called Additive Margin Softmax (AM-Softmax) is proving to be a better choice than the traditional Softmax. The AM-Softmax introduces a margin of separation between the classes that forces the samples from the same class to be closer to each other and also maximizes the distance between classes. In this paper, we propose a new approach for speaker recognition systems called AM-SincNet, which is based on the SincNet but uses an improved AM-Softmax layer. The proposed method is evaluated in the TIMIT dataset and obtained an improvement of approximately 40% in the Frame Error Rate compared to SincNet.

@misc{
 title = {Heartbeat anomaly detection using adversarial oversampling},
 type = {misc},
 year = {2019},
 source = {arXiv},
 id = {18b7c3c8-cde3-3c01-aef5-106cc0f71408},
 created = {2020-10-27T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2020-10-30T04:47:14.586Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Copyright © 2019, arXiv, All rights reserved. Cardiovascular diseases are one of the most common causes of death in the world. Prevention, knowledge of previous cases in the family, and early detection is the best strategy to reduce this fact. Different machine learning approaches to automatic diagnostic are being proposed to this task. As in most health problems, the imbalance between examples and classes is predominant in this problem and affects the performance of the automated solution. In this paper, we address the classification of heartbeats images in different cardiovascular diseases. We propose a two-dimensional Convolutional Neural Network for classification after using a InfoGAN architecture for generating synthetic images to unbalanced classes. We call this proposal Adversarial Oversampling and compare it with the classical oversampling methods as SMOTE, ADASYN, and RandomOversampling. The results show that the proposed approach improves the classifier performance for the minority classes without harming the performance in the balanced classes.},
 bibtype = {misc},
 author = {Lima, J.L.P. and Macêdo, D. and Zanchettin, C.}
}

Copyright © 2019, arXiv, All rights reserved. Cardiovascular diseases are one of the most common causes of death in the world. Prevention, knowledge of previous cases in the family, and early detection is the best strategy to reduce this fact. Different machine learning approaches to automatic diagnostic are being proposed to this task. As in most health problems, the imbalance between examples and classes is predominant in this problem and affects the performance of the automated solution. In this paper, we address the classification of heartbeats images in different cardiovascular diseases. We propose a two-dimensional Convolutional Neural Network for classification after using a InfoGAN architecture for generating synthetic images to unbalanced classes. We call this proposal Adversarial Oversampling and compare it with the classical oversampling methods as SMOTE, ADASYN, and RandomOversampling. The results show that the proposed approach improves the classifier performance for the minority classes without harming the performance in the balanced classes.

@misc{
 title = {Spatial-temporal graph convolutional networks for sign language recognition},
 type = {misc},
 year = {2019},
 source = {arXiv},
 id = {efd92981-1bb3-3cf5-90c5-dc8265034331},
 created = {2020-10-27T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2020-10-31T06:51:24.165Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Copyright © 2019, arXiv, All rights reserved. The recognition of sign language is a challenging task with an important role in society to facilitate the communication of deaf persons. We propose a new approach of Spatial-Temporal Graph Convolutional Network to sign language recognition based on the human skeletal movements. The method uses graphs to capture the signs dynamics in two dimensions, spatial and temporal, considering the complex aspects of the language. Additionally, we present a new dataset of human skeletons for sign language based on ASLLVD to contribute to future related studies.},
 bibtype = {misc},
 author = {De Amorim, C.C. and Macêdo, D. and Zanchettin, C.}
}

Copyright © 2019, arXiv, All rights reserved. The recognition of sign language is a challenging task with an important role in society to facilitate the communication of deaf persons. We propose a new approach of Spatial-Temporal Graph Convolutional Network to sign language recognition based on the human skeletal movements. The method uses graphs to capture the signs dynamics in two dimensions, spatial and temporal, considering the complex aspects of the language. Additionally, we present a new dataset of human skeletons for sign language based on ASLLVD to contribute to future related studies.

@misc{
 title = {Improving deep image clustering with spatial transformer layers},
 type = {misc},
 year = {2019},
 source = {arXiv},
 keywords = {Adaptive Clustering,Deep Neural Networks,Image Clustering,Spatial Transformer Networks,Visual Attention},
 id = {482d1a61-81cb-366c-9867-009002e9784c},
 created = {2020-10-28T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2020-10-31T22:01:52.735Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Copyright © 2019, arXiv, All rights reserved. Image clustering is an important but challenging task in machine learning. As in most image processing areas, the latest improvements came from models based on the deep learning approach. However, classical deep learning methods have problems to deal with spatial image transformations like scale and rotation. In this paper, we propose the use of visual attention techniques to reduce this problem in image clustering methods. We evaluate the combination of a deep image clustering model called Deep Adaptive Clustering (DAC) with the Spatial Transformer Networks (STN). The proposed model is evaluated in the datasets MNIST and FashionMNIST and outperformed the baseline model.},
 bibtype = {misc},
 author = {Souza, T.V.M. and Zanchettin, C.}
}

Copyright © 2019, arXiv, All rights reserved. Image clustering is an important but challenging task in machine learning. As in most image processing areas, the latest improvements came from models based on the deep learning approach. However, classical deep learning methods have problems to deal with spatial image transformations like scale and rotation. In this paper, we propose the use of visual attention techniques to reduce this problem in image clustering methods. We evaluate the combination of a deep image clustering model called Deep Adaptive Clustering (DAC) with the Spatial Transformer Networks (STN). The proposed model is evaluated in the datasets MNIST and FashionMNIST and outperformed the baseline model.

@misc{
 title = {Distinction maximization loss: Fast, scalable, turnkey, and native neural networks out-of-distribution detection simply by replacing the softmax loss},
 type = {misc},
 year = {2019},
 source = {arXiv},
 id = {48d442f2-6f41-314b-b755-a88d57c0e391},
 created = {2020-11-03T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2020-11-04T11:54:36.797Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Copyright © 2019, arXiv, All rights reserved. Recently, many methods to reduce neural networks uncertainty have been proposed. However, most of the techniques used in these solutions usually present severe drawbacks. In this paper, we argue that neural networks low outof- distribution detection performance is mainly due to the SoftMax loss anisotropy. Therefore, we built an isotropic loss to reduce neural networks uncertainty in a fast, scalable, turnkey, and native approach. Our experiments show that replacing SoftMax with the proposed loss does not affect classification accuracy. Moreover, our proposal overcomes ODIN typically by a large margin while producing usually competitive results against a state-of-the-art Mahalanobis method despite avoiding their limitations. Hence, neural networks uncertainty may be significantly reduced by a simple loss change without relying on special procedures such as data augmentation, adversarial training/validation, ensembles, or additional classification/regression models.},
 bibtype = {misc},
 author = {MacÊdo, D. and Ren, T.I. and Zanchettin, C. and Oliveira, A.L.I. and Tapp, A. and Ludermir, T.}
}

Copyright © 2019, arXiv, All rights reserved. Recently, many methods to reduce neural networks uncertainty have been proposed. However, most of the techniques used in these solutions usually present severe drawbacks. In this paper, we argue that neural networks low outof- distribution detection performance is mainly due to the SoftMax loss anisotropy. Therefore, we built an isotropic loss to reduce neural networks uncertainty in a fast, scalable, turnkey, and native approach. Our experiments show that replacing SoftMax with the proposed loss does not affect classification accuracy. Moreover, our proposal overcomes ODIN typically by a large margin while producing usually competitive results against a state-of-the-art Mahalanobis method despite avoiding their limitations. Hence, neural networks uncertainty may be significantly reduced by a simple loss change without relying on special procedures such as data augmentation, adversarial training/validation, ensembles, or additional classification/regression models.

@book{
 title = {Active Image Data Augmentation},
 type = {book},
 year = {2019},
 source = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
 keywords = {Data augmentation,Interpretability,Robustness},
 volume = {11734 LNAI},
 id = {9b95d783-baea-32e5-a5e3-037495a35e09},
 created = {2019-10-12T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2021-01-13T11:30:28.662Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2019, Springer Nature Switzerland AG. Deep neural networks models have achieved state-of-the-art results in a great number of different tasks in different domains (e.g., natural language processing and computer vision). However, the notions of robustness, causality, and fairness are not measured in traditional evaluated settings. In this work, we proposed an active data augmentation method to improve the model robustness to new data. We use the Vanilla Backpropagation to visualize what the trained model consider important in the input information. Based on that information, we augment the training dataset with new data to refine the model training. The objective is to make the model robust and effective for important input information. We evaluated our approach in a Spinal Cord Gray Matter Segmentation task and verified improvement in robustness while keeping the model competitive in the traditional metrics. Besides, we achieve the state-of-the-art results on that task using a U-Net based model.},
 bibtype = {book},
 author = {Santos, F.A.O. and Zanchettin, C. and Matos, L.N. and Novais, P.},
 doi = {10.1007/978-3-030-29859-3_27}
}

© 2019, Springer Nature Switzerland AG. Deep neural networks models have achieved state-of-the-art results in a great number of different tasks in different domains (e.g., natural language processing and computer vision). However, the notions of robustness, causality, and fairness are not measured in traditional evaluated settings. In this work, we proposed an active data augmentation method to improve the model robustness to new data. We use the Vanilla Backpropagation to visualize what the trained model consider important in the input information. Based on that information, we augment the training dataset with new data to refine the model training. The objective is to make the model robust and effective for important input information. We evaluated our approach in a Spinal Cord Gray Matter Segmentation task and verified improvement in robustness while keeping the model competitive in the traditional metrics. Besides, we achieve the state-of-the-art results on that task using a U-Net based model.

@book{
 title = {Dynamic Centroid Insertion and Adjustment for Data Sets with Multiple Imbalanced Classes},
 type = {book},
 year = {2019},
 source = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
 keywords = {Imbalanced domains,Multiclass,Prototype Generation},
 volume = {11728 LNCS},
 id = {0830d9fa-cab1-3d56-897f-3c0551fe1e92},
 created = {2019-10-11T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2021-01-13T16:04:59.804Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2019, Springer Nature Switzerland AG. The imbalance problem is receiving an increasing attention in the literature. Studies on binary cases are recurrent but limited when considering the multiple classes approach. Solutions to imbalance domains may be divided into two groups, data level approaches, and algorithmic approaches. The first approach is more common and focuses on changing the training data aiming to balance the data set, oversampling the smallest classes, undersampling the biggest ones or using a combination of both. Instance reduction is another approach to the problem. It tries to find the best-reduced set of instances that represent the original training set. In this work, we propose a new Prototype Generation method called DCIA. It dynamically inserts new prototypes for each class and then adjusts their positions with a search algorithm. The set of generated prototypes may be used to train any classifier. Experiments showed its potentiality by enabling an 1NN classifier to perform sometimes as well or even better than some ensemble classifiers created for different multiclass imbalanced domains.},
 bibtype = {book},
 author = {Silva, E.J.R. and Zanchettin, C.},
 doi = {10.1007/978-3-030-30484-3_60}
}

© 2019, Springer Nature Switzerland AG. The imbalance problem is receiving an increasing attention in the literature. Studies on binary cases are recurrent but limited when considering the multiple classes approach. Solutions to imbalance domains may be divided into two groups, data level approaches, and algorithmic approaches. The first approach is more common and focuses on changing the training data aiming to balance the data set, oversampling the smallest classes, undersampling the biggest ones or using a combination of both. Instance reduction is another approach to the problem. It tries to find the best-reduced set of instances that represent the original training set. In this work, we propose a new Prototype Generation method called DCIA. It dynamically inserts new prototypes for each class and then adjusts their positions with a search algorithm. The set of generated prototypes may be used to train any classifier. Experiments showed its potentiality by enabling an 1NN classifier to perform sometimes as well or even better than some ensemble classifiers created for different multiclass imbalanced domains.

@book{
 title = {Spatial-Temporal Graph Convolutional Networks for Sign Language Recognition},
 type = {book},
 year = {2019},
 source = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
 keywords = {Convolutional Neural Network,Sign language,Spatial Temporal Graph},
 volume = {11731 LNCS},
 id = {f0498f0c-a302-3d57-b314-ff26a3941e60},
 created = {2019-10-14T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2021-01-14T07:56:00.742Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© Springer Nature Switzerland AG 2019. The recognition of sign language is a challenging task with an important role in society to facilitate the communication of deaf persons. We propose a new approach of Spatial-Temporal Graph Convolutional Network for sign language recognition based on the human skeletal movements. The method uses graphs to capture the dynamics of the signs in two dimensions, spatial and temporal, considering the complex aspects of the language. Additionally, we present a new dataset of human skeletons for sign language based on ASLLVD to contribute to future related studies.},
 bibtype = {book},
 author = {de Amorim, C.C. and Macêdo, D. and Zanchettin, C.},
 doi = {10.1007/978-3-030-30493-5_59}
}

© Springer Nature Switzerland AG 2019. The recognition of sign language is a challenging task with an important role in society to facilitate the communication of deaf persons. We propose a new approach of Spatial-Temporal Graph Convolutional Network for sign language recognition based on the human skeletal movements. The method uses graphs to capture the dynamics of the signs in two dimensions, spatial and temporal, considering the complex aspects of the language. Additionally, we present a new dataset of human skeletons for sign language based on ASLLVD to contribute to future related studies.

@book{
 title = {Hierarchical Attentional Hybrid Neural Networks for Document Classification},
 type = {book},
 year = {2019},
 source = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
 keywords = {Attention mechanisms,Convolutional neural networks,Document classification,Text classification},
 volume = {11731 LNCS},
 id = {b59a5dad-d056-36b9-8718-818f5f26b1f5},
 created = {2019-10-14T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2021-01-14T12:48:24.575Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© Springer Nature Switzerland AG 2019. Document classification is a challenging task with important applications. The deep learning approaches to the problem have gained much attention recently. Despite the progress, the proposed models do not incorporate the knowledge of the document structure in the architecture efficiently and not take into account the contexting importance of words and sentences. In this paper, we propose a new approach based on a combination of convolutional neural networks, gated recurrent units, and attention mechanisms for document classification tasks. We use of convolution layers varying window sizes to extract more meaningful, generalizable and abstract features by the hierarchical representation. The proposed method in improves the results of the current attention-based approaches for document classification.},
 bibtype = {book},
 author = {Abreu, J. and Fred, L. and Macêdo, D. and Zanchettin, C.},
 doi = {10.1007/978-3-030-30493-5_39}
}

© Springer Nature Switzerland AG 2019. Document classification is a challenging task with important applications. The deep learning approaches to the problem have gained much attention recently. Despite the progress, the proposed models do not incorporate the knowledge of the document structure in the architecture efficiently and not take into account the contexting importance of words and sentences. In this paper, we propose a new approach based on a combination of convolutional neural networks, gated recurrent units, and attention mechanisms for document classification tasks. We use of convolution layers varying window sizes to extract more meaningful, generalizable and abstract features by the hierarchical representation. The proposed method in improves the results of the current attention-based approaches for document classification.

@inproceedings{
 title = {On the Influence of the Color Model for Image Boundary Detection Algorithms based on Convolutional Neural Networks},
 type = {inproceedings},
 year = {2019},
 keywords = {Boundary detection,CNN,Color models},
 volume = {2019-July},
 id = {d893554d-9c05-34af-9a77-009249ab277e},
 created = {2019-10-20T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2021-01-15T17:56:55.710Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2019 IEEE. Image analysis and understanding are challenging tasks, usually having segmentation as a major step. Boundary detection is a type of segmentation which aims to highlight the boundaries of the objects in a scene. Models based on Convolutional Neural Networks (CNN) have presented promising results for boundary detection, where the input usually is the entire image or some patches, often described in the RGB color model. In this paper, we provide a qualitative analysis of boundary detection algorithms based on CNN but considering images in different color models. We have used the color models RGB, Lab, Luv, dRdGdB, YO1O2 and HSV for this analysis. The Holistically-Nested Edge Detection (HED) and Convolutional Encoder Decoder Network (CEDN) are the CNN's chosen due to their high performance. The benchmark BSDS is the boundary detection evaluator. Experiments show that the results of the edge detection process tend to be similar when training the CNN with weights randomly initialized, regardless of the color model used. For the HED architecture, the use of Lab and Luv color models has resulted in a significant improvement to the case of transfer learning and fine-tuning of weights.},
 bibtype = {inproceedings},
 author = {Dos Santos, T.J. and Mello, C.A.B. and Zanchettin, C. and De Souza, T.V.M.},
 doi = {10.1109/IJCNN.2019.8851701},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

© 2019 IEEE. Image analysis and understanding are challenging tasks, usually having segmentation as a major step. Boundary detection is a type of segmentation which aims to highlight the boundaries of the objects in a scene. Models based on Convolutional Neural Networks (CNN) have presented promising results for boundary detection, where the input usually is the entire image or some patches, often described in the RGB color model. In this paper, we provide a qualitative analysis of boundary detection algorithms based on CNN but considering images in different color models. We have used the color models RGB, Lab, Luv, dRdGdB, YO1O2 and HSV for this analysis. The Holistically-Nested Edge Detection (HED) and Convolutional Encoder Decoder Network (CEDN) are the CNN's chosen due to their high performance. The benchmark BSDS is the boundary detection evaluator. Experiments show that the results of the edge detection process tend to be similar when training the CNN with weights randomly initialized, regardless of the color model used. For the HED architecture, the use of Lab and Luv color models has resulted in a significant improvement to the case of transfer learning and fine-tuning of weights.

@inproceedings{
 title = {Towards Optimizing Convolutional Neural Networks for Robotic Surgery Skill Evaluation},
 type = {inproceedings},
 year = {2019},
 volume = {2019-July},
 id = {dc7bbf0c-2033-37de-9b8e-de99a24531fc},
 created = {2019-10-20T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2021-01-15T18:04:48.886Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2019 IEEE. In medicine courses, improve the skills of surgery students is an essential part of the program. For training the surgeon residents the institutions normally using a standard checklist to evaluate the student evolution. However, the checklist evaluation is susceptible to evaluator bias, inter-evaluator variability, besides being time-consuming. The automation of this process is an important evolution in medical training. An alternative to the instructor checklist is capturing and evaluation of kinematic data regarding the surgical motion. We propose a novel CNN architecture for automated robot-assisted skill assessment. We explore the use of the SELU activation function and a global mixed pooling approach based on the average and max-pooling layers. Finally, we examine two types of convolutional layers: real-value and quaternion-valued. The results suggest that our model presents a higher average accuracy across the three surgical subtasks of the JIGSAWS dataset.},
 bibtype = {inproceedings},
 author = {Castro, D. and Pereira, D. and Zanchettin, C. and MacEdo, D. and Bezerra, B.L.D.},
 doi = {10.1109/IJCNN.2019.8852341},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

© 2019 IEEE. In medicine courses, improve the skills of surgery students is an essential part of the program. For training the surgeon residents the institutions normally using a standard checklist to evaluate the student evolution. However, the checklist evaluation is susceptible to evaluator bias, inter-evaluator variability, besides being time-consuming. The automation of this process is an important evolution in medical training. An alternative to the instructor checklist is capturing and evaluation of kinematic data regarding the surgical motion. We propose a novel CNN architecture for automated robot-assisted skill assessment. We explore the use of the SELU activation function and a global mixed pooling approach based on the average and max-pooling layers. Finally, we examine two types of convolutional layers: real-value and quaternion-valued. The results suggest that our model presents a higher average accuracy across the three surgical subtasks of the JIGSAWS dataset.

@inproceedings{
 title = {Improving Universal Language Model Fine-Tuning using Attention Mechanism},
 type = {inproceedings},
 year = {2019},
 volume = {2019-July},
 id = {7915d8a1-6b7d-3f5d-aca5-37bafcdfa86c},
 created = {2019-10-20T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2021-01-15T18:17:59.305Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2019 IEEE. Inductive transfer learning is widespread in computer vision applications. However, in natural language processing (NLP) applications is still an under-explored area. The most common transfer learning method in NLP is the use of pre-trained word embeddings. The Universal Language Model Fine-Tuning (ULMFiT) is a recent approach which proposes to train a language model and transfer its knowledge to a final classifier. During the classification step, ULMFiT uses a max and average pooling layer to select the useful information of an embedding sequence. We propose to replace max and average pooling layers with a soft attention mechanism. The goal is to learn the most important information of the embedding sequence rather than assuming that they are max and average values. We evaluate the proposed approach in six datasets and achieve the best performance in all of them against literature approaches.},
 bibtype = {inproceedings},
 author = {Santos, F.A.O. and Ponce-Guevara, K.L. and MacEdo, D. and Zanchettin, C.},
 doi = {10.1109/IJCNN.2019.8852398},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

© 2019 IEEE. Inductive transfer learning is widespread in computer vision applications. However, in natural language processing (NLP) applications is still an under-explored area. The most common transfer learning method in NLP is the use of pre-trained word embeddings. The Universal Language Model Fine-Tuning (ULMFiT) is a recent approach which proposes to train a language model and transfer its knowledge to a final classifier. During the classification step, ULMFiT uses a max and average pooling layer to select the useful information of an embedding sequence. We propose to replace max and average pooling layers with a soft attention mechanism. The goal is to learn the most important information of the embedding sequence rather than assuming that they are max and average values. We evaluate the proposed approach in six datasets and achieve the best performance in all of them against literature approaches.

@inproceedings{
 title = {Additive Margin SincNet for Speaker Recognition},
 type = {inproceedings},
 year = {2019},
 volume = {2019-July},
 id = {6874f57c-dc8d-349d-a8ac-fde95019664f},
 created = {2019-10-20T23:59:00.000Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2021-01-15T18:24:37.787Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2019 IEEE. Speaker Recognition is a challenging task with essential applications such as authentication, automation, and security. The SincNet is a new deep learning based model which has produced promising results to tackle the mentioned task. To train deep learning systems, the loss function is essential to the network performance. The Softmax loss function is a widely used function in deep learning methods, but it is not the best choice for all kind of problems. For distance-based problems, one new Softmax based loss function called Additive Margin Softmax (AM-Softmax) is proving to be a better choice than the traditional Softmax. The AM-Softmax introduces a margin of separation between the classes that forces the samples from the same class to be closer to each other and also maximizes the distance between classes. In this paper, we propose a new approach for speaker recognition systems called AM-SincNet, which is based on the SincNet but uses an improved AM-Softmax layer. The proposed method is evaluated in the TIMIT dataset and obtained an improvement of approximately 40% in the Frame Error Rate when compared to SincNet.},
 bibtype = {inproceedings},
 author = {Chagas Nunes, J.A. and MacEdo, D. and Zanchettin, C.},
 doi = {10.1109/IJCNN.2019.8852112},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

© 2019 IEEE. Speaker Recognition is a challenging task with essential applications such as authentication, automation, and security. The SincNet is a new deep learning based model which has produced promising results to tackle the mentioned task. To train deep learning systems, the loss function is essential to the network performance. The Softmax loss function is a widely used function in deep learning methods, but it is not the best choice for all kind of problems. For distance-based problems, one new Softmax based loss function called Additive Margin Softmax (AM-Softmax) is proving to be a better choice than the traditional Softmax. The AM-Softmax introduces a margin of separation between the classes that forces the samples from the same class to be closer to each other and also maximizes the distance between classes. In this paper, we propose a new approach for speaker recognition systems called AM-SincNet, which is based on the SincNet but uses an improved AM-Softmax layer. The proposed method is evaluated in the TIMIT dataset and obtained an improvement of approximately 40% in the Frame Error Rate when compared to SincNet.

@article{
 title = {Gesture recognition: A review focusing on sign language in a mobile context},
 type = {article},
 year = {2018},
 keywords = {Gesture recognition,Mobile devices,Sign language},
 volume = {103},
 id = {dbe4f932-7484-3a20-86d7-b0e34ceb33c8},
 created = {2019-02-14T18:02:00.301Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.301Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2018 Elsevier Ltd Sign languages, which consist of a combination of hand movements and facial expressions, are used by deaf persons around the world to communicate. However, hearing persons rarely know sign languages, creating barriers to inclusion. The increasing progress of mobile technology, along with new forms of user interaction, opens up possibilities for overcoming such barriers, particularly through the use of gesture recognition through smartphones. This Literature Review discusses works from 2009 to 2017 that present solutions for gesture recognition in a mobile context as well as facial recognition in sign languages. Among a diversity of hardware and techniques, sensor-based gloves were the most used special hardware, along with brute force comparison to classify gestures. Works that did not adopt special hardware mostly used skin color for feature extraction in gesture recognition. Classification algorithms included: Support Vector Machines, Hierarchical Temporal Memory and Feedforward backpropagation neural network, among others. Recognition of static gestures typically achieved results higher than 80%. Fewer papers recognized dynamic gestures, obtaining results above 90%. However, most experiments were performed under controlled environments, with specific lighting conditions, and were only using a small set of gestures. In addition, the majority of works dealt with a simple background and used special hardware (which is often cumbersome for the user) to facilitate feature extraction. Facial expression recognition achieved high classification results using Random-Forest and Multi-layer Perceptron. Despite the progress being made with the increasing interest in gesture recognition, there are still important gaps to be addressed in the context of sign languages. Besides improving usability and efficacy of the solutions, recognition of facial expression and of both static and dynamic gestures in complex backgrounds must be considered.},
 bibtype = {article},
 author = {Hirafuji Neiva, D. and Zanchettin, C.},
 doi = {10.1016/j.eswa.2018.01.051},
 journal = {Expert Systems with Applications}
}

© 2018 Elsevier Ltd Sign languages, which consist of a combination of hand movements and facial expressions, are used by deaf persons around the world to communicate. However, hearing persons rarely know sign languages, creating barriers to inclusion. The increasing progress of mobile technology, along with new forms of user interaction, opens up possibilities for overcoming such barriers, particularly through the use of gesture recognition through smartphones. This Literature Review discusses works from 2009 to 2017 that present solutions for gesture recognition in a mobile context as well as facial recognition in sign languages. Among a diversity of hardware and techniques, sensor-based gloves were the most used special hardware, along with brute force comparison to classify gestures. Works that did not adopt special hardware mostly used skin color for feature extraction in gesture recognition. Classification algorithms included: Support Vector Machines, Hierarchical Temporal Memory and Feedforward backpropagation neural network, among others. Recognition of static gestures typically achieved results higher than 80%. Fewer papers recognized dynamic gestures, obtaining results above 90%. However, most experiments were performed under controlled environments, with specific lighting conditions, and were only using a small set of gestures. In addition, the majority of works dealt with a simple background and used special hardware (which is often cumbersome for the user) to facilitate feature extraction. Facial expression recognition achieved high classification results using Random-Forest and Multi-layer Perceptron. Despite the progress being made with the increasing interest in gesture recognition, there are still important gaps to be addressed in the context of sign languages. Besides improving usability and efficacy of the solutions, recognition of facial expression and of both static and dynamic gestures in complex backgrounds must be considered.

@inproceedings{
 title = {Reducing SqueezeNet Storage Size with Depthwise Separable Convolutions},
 type = {inproceedings},
 year = {2018},
 volume = {2018-July},
 id = {7cc54a14-0680-387c-908c-f547044fd5b8},
 created = {2019-02-14T18:02:01.676Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.676Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2018 IEEE. Current research in the field of convolutional neu-ral networks usually focuses on improving network accuracy, regardless of the network size and inference time. In this paper, we investigate the effects of storage space reduction in SqueezeNet as it relates to inference time when processing single test samples. In order to reduce the storage space, we suggest adjusting SqueezeNet's Fire Modules to include Depthwise Separable Convolutions (DSC). The resulting network, referred to as SqueezeNet-DSC, is compared to different convolutional neural networks such as MobileNet, AlexNet, VGG19, and the original SqueezeNet itself. When analyzing the models, we consider accuracy, the number of parameters, parameter storage size and processing time of a single test sample on CIFAR-10 and CIFAR-100 databases. The SqueezeNet-DSC exhibited a considerable size reduction (37% the size of SqueezeNet), while experiencing a loss in network accuracy of 1,07% in CIFAR-10 and 3,06% in top 1 CIFAR-100.},
 bibtype = {inproceedings},
 author = {Santos, A.G. and De Souza, C.O. and Zanchettin, C. and Macedo, D. and Oliveira, A.L.I. and Ludermir, T.},
 doi = {10.1109/IJCNN.2018.8489442},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

© 2018 IEEE. Current research in the field of convolutional neu-ral networks usually focuses on improving network accuracy, regardless of the network size and inference time. In this paper, we investigate the effects of storage space reduction in SqueezeNet as it relates to inference time when processing single test samples. In order to reduce the storage space, we suggest adjusting SqueezeNet's Fire Modules to include Depthwise Separable Convolutions (DSC). The resulting network, referred to as SqueezeNet-DSC, is compared to different convolutional neural networks such as MobileNet, AlexNet, VGG19, and the original SqueezeNet itself. When analyzing the models, we consider accuracy, the number of parameters, parameter storage size and processing time of a single test sample on CIFAR-10 and CIFAR-100 databases. The SqueezeNet-DSC exhibited a considerable size reduction (37% the size of SqueezeNet), while experiencing a loss in network accuracy of 1,07% in CIFAR-10 and 3,06% in top 1 CIFAR-100.

@inproceedings{
 title = {SegNetRes-CRF: A Deep Convolutional Encoder-Decoder Architecture for Semantic Image Segmentation},
 type = {inproceedings},
 year = {2018},
 keywords = {CRF,Encoder-decoder architectures,Image segmentation,Residual networks,SegNet,U-Net},
 volume = {2018-July},
 id = {8f4e3e07-0c35-3589-a26c-60f718ae8cf7},
 created = {2019-02-14T18:02:01.791Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.791Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2018 IEEE. Semantic segmentation is an essential task in computer vision that aims to label each image pixel. Several of the actual best approaches in this context are based on deep neural networks. For example, SegNet is a deep encoderdecoder architecture approach whose results were disruptive because it is fast and performs well. However, this architecture fails to finedelineating the edges between the objects of interest in the image. We propose some modifications in the SegNet-Basic architecture by using a post-processing segmentation layer (using Conditional Random Fields) and by transferring high resolution features combined to the decoder network. The proposed method was evaluated in the dataset CamVid. Moreover, it was compared with important variants of SegNet and showed to be able to improve the overall accuracy of SegNet-Basic by up to 17.5%.},
 bibtype = {inproceedings},
 author = {De Oliveira Junior, L.A. and Medeiros, H.R. and Macedo, D. and Zanchettin, C. and Oliveira, A.L.I. and Ludermir, T.},
 doi = {10.1109/IJCNN.2018.8489376},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

© 2018 IEEE. Semantic segmentation is an essential task in computer vision that aims to label each image pixel. Several of the actual best approaches in this context are based on deep neural networks. For example, SegNet is a deep encoderdecoder architecture approach whose results were disruptive because it is fast and performs well. However, this architecture fails to finedelineating the edges between the objects of interest in the image. We propose some modifications in the SegNet-Basic architecture by using a post-processing segmentation layer (using Conditional Random Fields) and by transferring high resolution features combined to the decoder network. The proposed method was evaluated in the dataset CamVid. Moreover, it was compared with important variants of SegNet and showed to be able to improve the overall accuracy of SegNet-Basic by up to 17.5%.

@inproceedings{
 title = {A Dynamic Gesture Recognition System to Translate between Sign Languages in Complex Backgrounds},
 type = {inproceedings},
 year = {2017},
 keywords = {Computer vision,ELM,Gesture recognition,Sign language,Translation between gestures},
 id = {29abf880-c0d3-3c65-a4f1-d90f6bb503fb},
 created = {2019-02-14T18:02:00.241Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.241Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2016 IEEE. Sign languages, used by people with hearing impairments to communicate, are not universal. Just like any other language, every country has its own, which makes it difficult to establish communication, as usually people only know their native sign language. This paper presents a novel integrated system to minimize this barrier by combining a web application that uses computer vision techniques and Extreme Learning Machines with a mobile application, to translate between sign languages.},
 bibtype = {inproceedings},
 author = {Neiva, D.H. and Zanchettin, C.},
 doi = {10.1109/BRACIS.2016.082},
 booktitle = {Proceedings - 2016 5th Brazilian Conference on Intelligent Systems, BRACIS 2016}
}

© 2016 IEEE. Sign languages, used by people with hearing impairments to communicate, are not universal. Just like any other language, every country has its own, which makes it difficult to establish communication, as usually people only know their native sign language. This paper presents a novel integrated system to minimize this barrier by combining a web application that uses computer vision techniques and Extreme Learning Machines with a mobile application, to translate between sign languages.

@inproceedings{
 title = {On Validation Setup for Multiclass Imbalanced Data Sets},
 type = {inproceedings},
 year = {2017},
 id = {93c32dde-9f96-3045-a61d-c99fdeaf09a0},
 created = {2019-02-14T18:02:00.249Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.249Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2016 IEEE. The validation of experiments is commonly evaluated with Cross-Validation methods. In the literature the 10-fold, followed by bootstrap, are the most indicated methods. However there lacks a study of a proper validation procedure for imbalanced data sets, specially for the rare class case. In this work the most used validation methods were tested in ten imbalanced data sets, with a generic and an ad hoc classifiers. Analyses showed that 10-fold, followed by hold-out, are the indicated methods when using a generic classifier. For the ad hoc classifier the 10-fold, followed by bootstrap, are the indicated ones. In the case of rare classes in a data set, the most indicated method is the repeated hold-out.},
 bibtype = {inproceedings},
 author = {Silva, E.J.R. and Zanchettin, C.},
 doi = {10.1109/BRACIS.2016.090},
 booktitle = {Proceedings - 2016 5th Brazilian Conference on Intelligent Systems, BRACIS 2016}
}

© 2016 IEEE. The validation of experiments is commonly evaluated with Cross-Validation methods. In the literature the 10-fold, followed by bootstrap, are the most indicated methods. However there lacks a study of a proper validation procedure for imbalanced data sets, specially for the rare class case. In this work the most used validation methods were tested in ten imbalanced data sets, with a generic and an ad hoc classifiers. Analyses showed that 10-fold, followed by hold-out, are the indicated methods when using a generic classifier. For the ad hoc classifier the 10-fold, followed by bootstrap, are the indicated ones. In the case of rare classes in a data set, the most indicated method is the repeated hold-out.

@inproceedings{
 title = {A Voronoi Diagram Based Classifier for Multiclass Imbalanced Data Sets},
 type = {inproceedings},
 year = {2017},
 id = {36b1c66a-c924-3564-b52a-a1b0805498c4},
 created = {2019-02-14T18:02:00.283Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.283Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2016 IEEE. The imbalance problem is receiving an increasing attention in the literature. Studies in binary cases are recurrent, however there still are several real world problems with more than two classes. The known solutions for binary datasets may not be applicable in this case. Some efforts are being applied in decomposition techniques which transforms a multiclass problem into some binary problems. However it is also possible to face a multiclass problem with an ad hoc approach, i.e., a classifier able to handle all classes at once. In this work a method able to handle several classes is proposed. This new method is based on the Voronoi diagram. We try to dynamically divide the feature space into several regions, each one assigned to a different class. It is expected for the method to be able to construct a complex classification model. However, as it is in its beginning, some tests need to be performed in order to evaluate its feasibility. Experiments with some classical classifiers confirm its feasibility, and comparisons with ad hoc methods found in literature show its potentiality.},
 bibtype = {inproceedings},
 author = {Silva, E.J.R. and Zanchettin, C.},
 doi = {10.1109/BRACIS.2016.030},
 booktitle = {Proceedings - 2016 5th Brazilian Conference on Intelligent Systems, BRACIS 2016}
}

© 2016 IEEE. The imbalance problem is receiving an increasing attention in the literature. Studies in binary cases are recurrent, however there still are several real world problems with more than two classes. The known solutions for binary datasets may not be applicable in this case. Some efforts are being applied in decomposition techniques which transforms a multiclass problem into some binary problems. However it is also possible to face a multiclass problem with an ad hoc approach, i.e., a classifier able to handle all classes at once. In this work a method able to handle several classes is proposed. This new method is based on the Voronoi diagram. We try to dynamically divide the feature space into several regions, each one assigned to a different class. It is expected for the method to be able to construct a complex classification model. However, as it is in its beginning, some tests need to be performed in order to evaluate its feasibility. Experiments with some classical classifiers confirm its feasibility, and comparisons with ad hoc methods found in literature show its potentiality.

@book{
 title = {Preface},
 type = {book},
 year = {2017},
 source = {Handwriting: Recognition, Development and Analysis},
 id = {4a2c6b88-51f3-36f7-a6cd-dbae7ae31532},
 created = {2019-02-14T18:02:00.940Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.940Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 bibtype = {book},
 author = {Bezerra, B.L.D. and Zanchettin, C. and Toselli, A.H. and Pirlo, G.}
}

@inproceedings{
 title = {Building Ensembles with Classifier Selection Using Self-Organizing Maps},
 type = {inproceedings},
 year = {2017},
 keywords = {Artificial Neural Networks,Ensemble of classifiers,Self-Organizing Maps},
 id = {368076f0-5bf4-349b-be9d-9db288af47cb},
 created = {2019-02-14T18:02:01.091Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.091Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2016 IEEE. Improving the performance of supervised classification methods is a subject of many literature works. An efficient strategy is the adoption of an ensemble of classifiers to divide the classification problem. In ensembles with classifier selection, there is no fusion of the classifiers decisions. A particular classifier is selected according to the input data instead. In this paper, well-known clustering methods based on self-organizing structures are used to implement ensembles with classifier selection. The self-organizing structures are used to detect the topological structure of data and help to divide the problem into smaller and easier sub-problems to solve. Experiments with different datasets show that the use of clustering methods to perform the classifier selection can contribute to split the problem and improve the classification accuracy compared to some traditional strategies. Additionally, the results encourage the development of more research to find out other ways to split problems using data clustering techniques.},
 bibtype = {inproceedings},
 author = {Almeida, L.M. and Zanchettin, C. and Leite, H.P.B.},
 doi = {10.1109/BRACIS.2016.087},
 booktitle = {Proceedings - 2016 5th Brazilian Conference on Intelligent Systems, BRACIS 2016}
}

© 2016 IEEE. Improving the performance of supervised classification methods is a subject of many literature works. An efficient strategy is the adoption of an ensemble of classifiers to divide the classification problem. In ensembles with classifier selection, there is no fusion of the classifiers decisions. A particular classifier is selected according to the input data instead. In this paper, well-known clustering methods based on self-organizing structures are used to implement ensembles with classifier selection. The self-organizing structures are used to detect the topological structure of data and help to divide the problem into smaller and easier sub-problems to solve. Experiments with different datasets show that the use of clustering methods to perform the classifier selection can contribute to split the problem and improve the classification accuracy compared to some traditional strategies. Additionally, the results encourage the development of more research to find out other ways to split problems using data clustering techniques.

@book{
 title = {The impact of dataset complexity on transfer learning over convolutional neural networks},
 type = {book},
 year = {2017},
 source = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
 keywords = {Convolution neural networks,Dataset complexity,Transfer learning},
 volume = {10614 LNCS},
 id = {3322aac2-0ffe-37a7-bf81-6d5c19f4d1c2},
 created = {2019-02-14T18:02:01.350Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.350Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© Springer International Publishing AG 2017. This paper makes use of diverse domains’ datasets to analyze the impact of image complexity and diversity on the task of transfer learning in deep neural networks. As the availability of labels and quality instances for several domains are still scarce, it is imperative to use the knowledge acquired from similar problems to improve classifier performance by transferring the learned parameters. We performed a statistical analysis through several experiments in which the convolutional neural networks (LeNet-5, AlexNet, VGG-11 and VGG-16) were trained and transferred to different target tasks layer by layer. We show that when working with complex low-quality images and small datasets, fine-tuning the transferred features learned from a low complexity source dataset gives the best results.},
 bibtype = {book},
 author = {Wanderley, M.D.S. and E Bueno, L.A. and Zanchettin, C. and Oliveira, A.L.I.},
 doi = {10.1007/978-3-319-68612-7_66}
}

© Springer International Publishing AG 2017. This paper makes use of diverse domains’ datasets to analyze the impact of image complexity and diversity on the task of transfer learning in deep neural networks. As the availability of labels and quality instances for several domains are still scarce, it is imperative to use the knowledge acquired from similar problems to improve classifier performance by transferring the learned parameters. We performed a statistical analysis through several experiments in which the convolutional neural networks (LeNet-5, AlexNet, VGG-11 and VGG-16) were trained and transferred to different target tasks layer by layer. We show that when working with complex low-quality images and small datasets, fine-tuning the transferred features learned from a low complexity source dataset gives the best results.

@book{
 title = {Handwriting: Recognition, development and analysis},
 type = {book},
 year = {2017},
 source = {Handwriting: Recognition, Development and Analysis},
 id = {f234e7b5-0d59-37e7-a87c-7a04b374db5a},
 created = {2019-02-14T18:02:01.509Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.509Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2017 by Nova Science Publishers, Inc. All rights reserved. This book has the primary goal of presenting and discussing some recent advances and ongoing developments in the Handwritten Text Recognition (HTR) field, resulting from works done on different HTR-related topics for the achievement of more accurate and efficient recognition systems. Nowadays, there is an enormous worldwide interest in HTR systems, which is mostly driven by the emergence of new portable devices incorporating handwriting recognition functions. Others interests are the biometric identification systems employing handwritten signatures, as well as the requirements from cultural heritage institutions like historical archives and libraries in order to preserve their large collections of historical (handwritten) documents. The book is organized into two sections: the first one is mainly devoted to describing the current state-of-the-art applications in HTR and the last advances in some of the steps involved in HTR workflow (that is, preprocessing, feature extraction, recognition engines, etc.), whereas the second focuses more on some relevant HTR-related applications. In more depth, the first part offers an overview of the current state-of-the-art applications of HTR technology and introduces the new challenges and research opportunities in the field. Besides, it provides a general discussion of currently ongoing approaches towards solving the underlying search problems on the basis of existing methods for HTR in terms of both accuracy and efficiency. In particular, there are chapters especially focused on image thresholding and enhancement, text image preprocessing techniques for historical handwritten documents and feature extraction methods for HTR. Likewise, in line with the breakout success of Deep Neural Networks (DNNs) in the field, a whole chapter is devoted to describing the designing of HTR systems based on DNNs. Finally, a chapter listing the most used benchmarking datasets for HTR is also included, providing detailed information about which types of HTR systems (on/offline) and features are commonly considered for each of them. In the second part, several systems - also developed on the basis of the fundamental concepts and general approaches outlined in the first part - are described for several HTR-related applications. Presented in the corresponding chapters, these applications cover a wide spectrum of scenarios: mathematical formulae recognition, scripting language recognition, multimodal handwriting-speech recognition, hardware design for online HTR, student performance evaluation through handwriting analysis, performance evaluation methods, keyword spotting, and handwritten signature verification systems. Last but not least, it is important to remark that to a large extent, this book is the result of works carried out by several researchers in the Handwritten Text Recognition field. Therefore, it owes credit to these researchers that have directly contributed to their ideas, discussions and technical collaborations, and in general who, in one manner or another, have made it possible.},
 bibtype = {book},
 author = {Bezerra, B.L.D. and Zanchettin, C. and Toselli, A.H. and Pirlo, G.}
}

© 2017 by Nova Science Publishers, Inc. All rights reserved. This book has the primary goal of presenting and discussing some recent advances and ongoing developments in the Handwritten Text Recognition (HTR) field, resulting from works done on different HTR-related topics for the achievement of more accurate and efficient recognition systems. Nowadays, there is an enormous worldwide interest in HTR systems, which is mostly driven by the emergence of new portable devices incorporating handwriting recognition functions. Others interests are the biometric identification systems employing handwritten signatures, as well as the requirements from cultural heritage institutions like historical archives and libraries in order to preserve their large collections of historical (handwritten) documents. The book is organized into two sections: the first one is mainly devoted to describing the current state-of-the-art applications in HTR and the last advances in some of the steps involved in HTR workflow (that is, preprocessing, feature extraction, recognition engines, etc.), whereas the second focuses more on some relevant HTR-related applications. In more depth, the first part offers an overview of the current state-of-the-art applications of HTR technology and introduces the new challenges and research opportunities in the field. Besides, it provides a general discussion of currently ongoing approaches towards solving the underlying search problems on the basis of existing methods for HTR in terms of both accuracy and efficiency. In particular, there are chapters especially focused on image thresholding and enhancement, text image preprocessing techniques for historical handwritten documents and feature extraction methods for HTR. Likewise, in line with the breakout success of Deep Neural Networks (DNNs) in the field, a whole chapter is devoted to describing the designing of HTR systems based on DNNs. Finally, a chapter listing the most used benchmarking datasets for HTR is also included, providing detailed information about which types of HTR systems (on/offline) and features are commonly considered for each of them. In the second part, several systems - also developed on the basis of the fundamental concepts and general approaches outlined in the first part - are described for several HTR-related applications. Presented in the corresponding chapters, these applications cover a wide spectrum of scenarios: mathematical formulae recognition, scripting language recognition, multimodal handwriting-speech recognition, hardware design for online HTR, student performance evaluation through handwriting analysis, performance evaluation methods, keyword spotting, and handwritten signature verification systems. Last but not least, it is important to remark that to a large extent, this book is the result of works carried out by several researchers in the Handwritten Text Recognition field. Therefore, it owes credit to these researchers that have directly contributed to their ideas, discussions and technical collaborations, and in general who, in one manner or another, have made it possible.

@book{
 title = {Handwriting recognition: Overview, challenges and future trends},
 type = {book},
 year = {2017},
 source = {Handwriting: Recognition, Development and Analysis},
 id = {515d9f25-db30-38a8-a8e1-1630588bddde},
 created = {2019-02-14T18:02:01.586Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.586Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 bibtype = {book},
 author = {Lacerda, E.B. and de Souza, T.V.M. and Zanchettin, C. and Rabelo, J.C.B. and Coutinho, L.D.}
}

@article{
 title = {QRNN: Q-generalized random neural network},
 type = {article},
 year = {2017},
 keywords = {Activation functions,Random neural networks (RNNs),Tsallis statistics,q-Gaussian},
 volume = {28},
 id = {f8b4221c-acf1-305e-bfdf-e383a134975e},
 created = {2019-02-14T18:02:01.633Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.633Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2012 IEEE. Artificial neural networks (ANNs) are widely used in applications with complex decision boundaries. A large number of activation functions have been proposed in the literature to achieve better representations of the observed data. However, only a few works employ Tsallis statistics, which has successfully been applied to various other fields. This paper presents a random neural network (RNN) with q-Gaussian activation functions [ q -generalized RNN (QRNN)] based on Tsallis statistics. The proposed method employs an additional parameter q (called the entropic index) which reflects the degree of nonextensivity. This approach has the flexibility to model complex decision boundaries of different shapes by varying the entropic index. We conduct numerical experiments to analyze the efficiency of QRNN compared with RNNs and several other classical methods. Statistical tests (Wilcoxon and Friedman) are used to validate our results and show that the QRNN performs significantly better than RNNs with different activation functions. In addition, we find that QRNN outperforms many of the compared classical methods, with the exception of support vector machines, in which case it still exhibits a substantial advantage in terms of implementation simplicity and speed.},
 bibtype = {article},
 author = {Stosic, D. and Stosic, D. and Zanchettin, C. and Ludermir, T. and Stosic, B.},
 doi = {10.1109/TNNLS.2015.2513365},
 journal = {IEEE Transactions on Neural Networks and Learning Systems},
 number = {2}
}

© 2012 IEEE. Artificial neural networks (ANNs) are widely used in applications with complex decision boundaries. A large number of activation functions have been proposed in the literature to achieve better representations of the observed data. However, only a few works employ Tsallis statistics, which has successfully been applied to various other fields. This paper presents a random neural network (RNN) with q-Gaussian activation functions [ q -generalized RNN (QRNN)] based on Tsallis statistics. The proposed method employs an additional parameter q (called the entropic index) which reflects the degree of nonextensivity. This approach has the flexibility to model complex decision boundaries of different shapes by varying the entropic index. We conduct numerical experiments to analyze the efficiency of QRNN compared with RNNs and several other classical methods. Statistical tests (Wilcoxon and Friedman) are used to validate our results and show that the QRNN performs significantly better than RNNs with different activation functions. In addition, we find that QRNN outperforms many of the compared classical methods, with the exception of support vector machines, in which case it still exhibits a substantial advantage in terms of implementation simplicity and speed.

@book{
 title = {Handwritten and printed image datasets: A review and proposals for automatic building},
 type = {book},
 year = {2017},
 source = {Handwriting: Recognition, Development and Analysis},
 id = {6c05373a-f723-321d-bba9-6d9c4a6c830e},
 created = {2019-02-14T18:02:01.725Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.725Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 bibtype = {book},
 author = {Ferreira, G.V. and Gouveia, F.M. and Bezerra, B.L.D. and Muller, E. and Zanchettin, C. and Toselli, A.}
}

@inproceedings{
 title = {On the Existence of a Threshold in Class Imbalance Problems},
 type = {inproceedings},
 year = {2016},
 keywords = {Class imbalance problem,overlapping borders,resampling},
 id = {d7e78d90-5fa1-375d-bf40-9a20ad9b6537},
 created = {2019-02-14T18:02:00.042Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.042Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2015 IEEE. One common approach to class imbalance problem is the resampling of data. However this strategy has some drawbacks, e.g., Unnecessary noise or the possibility of throwing out useful information. These inconveniences may be avoided or minimized by using a class proportion threshold allowing to identify when the imbalance data represent a problem to the classifier performance. In this paper we investigate the existence of this threshold and evaluate the performance of different classifiers in imbalanced problems. Results showed that for classifiers sensible to imbalanced data this threshold exists.},
 bibtype = {inproceedings},
 author = {Silva, E.J.R. and Zanchettin, C.},
 doi = {10.1109/SMC.2015.474},
 booktitle = {Proceedings - 2015 IEEE International Conference on Systems, Man, and Cybernetics, SMC 2015}
}

© 2015 IEEE. One common approach to class imbalance problem is the resampling of data. However this strategy has some drawbacks, e.g., Unnecessary noise or the possibility of throwing out useful information. These inconveniences may be avoided or minimized by using a class proportion threshold allowing to identify when the imbalance data represent a problem to the classifier performance. In this paper we investigate the existence of this threshold and evaluate the performance of different classifiers in imbalanced problems. Results showed that for classifiers sensible to imbalanced data this threshold exists.

@inproceedings{
 title = {Objective Video Quality Assessment Based on Neural Networks},
 type = {inproceedings},
 year = {2016},
 keywords = {Video quality assessment,neural networks,quality of experience},
 volume = {96},
 id = {9cfefe46-717c-3156-8508-fca8f32cfbb1},
 created = {2019-02-14T18:02:01.196Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.196Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2016 The Authors. Published by Elsevier B.V. Image/Video Quality Assessment (IQA/VQA) plays a significant role in image and video processing, as it can directly predict the impact of distortions on the video in the quality of experience (QoE) of the user. For this propose, in this paper, it is presented a new method for objective video quality assessment using an artificial neural network to predict the subjective evaluation of the video as if it were observed by a human user. The network was trained using degradation indicators extracted from the VQEG Phase I video database, which describe the level of distortion suffered by the original video under spatial and temporal scopes. The proposed method obtained an excellent correlation with the subjective scores over this same database.},
 bibtype = {inproceedings},
 author = {Menor, D.P.A. and Mello, C.A.B. and Zanchettin, C.},
 doi = {10.1016/j.procs.2016.08.202},
 booktitle = {Procedia Computer Science}
}

© 2016 The Authors. Published by Elsevier B.V. Image/Video Quality Assessment (IQA/VQA) plays a significant role in image and video processing, as it can directly predict the impact of distortions on the video in the quality of experience (QoE) of the user. For this propose, in this paper, it is presented a new method for objective video quality assessment using an artificial neural network to predict the subjective evaluation of the video as if it were observed by a human user. The network was trained using degradation indicators extracted from the VQEG Phase I video database, which describe the level of distortion suffered by the original video under spatial and temporal scopes. The proposed method obtained an excellent correlation with the subjective scores over this same database.

@inproceedings{
 title = {Extreme Learning Machine for Real Time Recognition of Brazilian Sign Language},
 type = {inproceedings},
 year = {2016},
 keywords = {Extreme Learning Machine,Real-Time Application,Sign Language Recognizer},
 id = {d73f6417-b6f3-3dc0-921c-cc697f08cf73},
 created = {2019-02-14T18:02:01.670Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.670Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2015 IEEE. The quantity of computing application that interacts with users through gesture or body motion has been growing. Among these applications is the sign language recognizer used to help hearing impaired people. This work proposes an architecture able to recognize Brazilian sign language (LIBRAS) in an embedded platform. The system focuses on a simple feature from 'finger spelling expressions' represented by a series of hands gestural images, and uses the Extreme Learning Machine network to classify them. The proposed structure uses camera images only and does not need any gloves or sensors. The obtained results are 5 times faster and 16 times better than classical approaches.},
 bibtype = {inproceedings},
 author = {Neto, F.M.D.P. and Cambuim, L.F. and Macieira, R.M. and Ludermir, T.B. and Zanchettin, C. and Barros, E.N.},
 doi = {10.1109/SMC.2015.259},
 booktitle = {Proceedings - 2015 IEEE International Conference on Systems, Man, and Cybernetics, SMC 2015}
}

© 2015 IEEE. The quantity of computing application that interacts with users through gesture or body motion has been growing. Among these applications is the sign language recognizer used to help hearing impaired people. This work proposes an architecture able to recognize Brazilian sign language (LIBRAS) in an embedded platform. The system focuses on a simple feature from 'finger spelling expressions' represented by a series of hands gestural images, and uses the Extreme Learning Machine network to classify them. The proposed structure uses camera images only and does not need any gloves or sensors. The obtained results are 5 times faster and 16 times better than classical approaches.

@article{
 title = {An efficient static gesture recognizer embedded system based on ELM pattern recognition algorithm},
 type = {article},
 year = {2016},
 keywords = {Computer vision,Embedded systems,FPGA,Neural network},
 volume = {68},
 id = {648988d1-d72c-3327-825d-8a0c1feba880},
 created = {2019-02-14T18:02:01.719Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.719Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2016 Elsevier B.V. All rights reserved. All rights reserved. Millions of people throughout the world describe themselves as being deaf. Some of them suffer from severe hearing loss and consequently use an alternative manner with which to communicate with society by means of either written or visual language. There are several sign languages capable of dealing with such a need. Nonetheless, a communication gap still exists even when using such languages, since only a small fraction of the population is able to use them. Over the last few years, due to the increasing need for universal accessibility when using computational resources, gesture recognition has been widely researched. Thus, in an attempt to reduce this communication gap, our approach proposes a computational solution in order to translate static gesture symbols into text symbols, through computer vision, without the use of hand sensors or gloves. In order to guarantee the highest quality, with emphasis on the reliability of the system and real-time translation, we have developed an approach based on the Extreme Learning Machine (ELM) pattern recognition algorithms fully implemented in hardware, and have assessed it to measure these two metrics. Hardware components were designed in order to perform the best image processing and pattern recognition tasks used within the project. As a case study, and so as to validate the technique, a recognition system for the Brazilian Sign Language (LIBRAS) was implemented. Besides ensuring that this approach could be used for any static hand gesture symbol recognition, our main goal was to guarantee fast, reliable gesture recognition for communication between humans. Experimental results have demonstrated that the system is able to recognize LIBRAS symbols with an accuracy of 97%, a response time of 6.5ms per letter recognition, and using only 43% (about 64,851 logic elements) of the FPGA area.},
 bibtype = {article},
 author = {Cambuim, L.F.S. and Macieira, R.M. and Neto, F.M.P. and Barros, E. and Ludermir, T.B. and Zanchettin, C.},
 doi = {10.1016/j.sysarc.2016.06.002},
 journal = {Journal of Systems Architecture}
}

© 2016 Elsevier B.V. All rights reserved. All rights reserved. Millions of people throughout the world describe themselves as being deaf. Some of them suffer from severe hearing loss and consequently use an alternative manner with which to communicate with society by means of either written or visual language. There are several sign languages capable of dealing with such a need. Nonetheless, a communication gap still exists even when using such languages, since only a small fraction of the population is able to use them. Over the last few years, due to the increasing need for universal accessibility when using computational resources, gesture recognition has been widely researched. Thus, in an attempt to reduce this communication gap, our approach proposes a computational solution in order to translate static gesture symbols into text symbols, through computer vision, without the use of hand sensors or gloves. In order to guarantee the highest quality, with emphasis on the reliability of the system and real-time translation, we have developed an approach based on the Extreme Learning Machine (ELM) pattern recognition algorithms fully implemented in hardware, and have assessed it to measure these two metrics. Hardware components were designed in order to perform the best image processing and pattern recognition tasks used within the project. As a case study, and so as to validate the technique, a recognition system for the Brazilian Sign Language (LIBRAS) was implemented. Besides ensuring that this approach could be used for any static hand gesture symbol recognition, our main goal was to guarantee fast, reliable gesture recognition for communication between humans. Experimental results have demonstrated that the system is able to recognize LIBRAS symbols with an accuracy of 97%, a response time of 6.5ms per letter recognition, and using only 43% (about 64,851 logic elements) of the FPGA area.

@article{
 title = {An intelligent monitoring system for natural gas odorization},
 type = {article},
 year = {2015},
 volume = {15},
 id = {33b7eaa1-2130-3851-93e5-213ce8800f62},
 created = {2019-02-14T18:02:01.054Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.054Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2001-2012 IEEE. In this paper, we present the design of an intelligent monitoring system consisting of physical sensors and intelligent software for the automatic identification of the concentration of natural gas odorants in the environment. An optical-based sensor array was proposed comprising the hardware module. The software module employs wavelets filters and artificial neural networks to recognize the concentration of odorant in a natural gas sample. The objective is to help the natural gas odorization process by means of end point monitoring through the recognizing of the odorant concentration. The recognizing process uses a benchmark index, which measures the degrees of human perception of gas in the environment. In this way, the proposed system tries to mimic the human perception of a natural gas leak and helps one to indicate if more or less amount of odorant should be added into the gas pipeline. Experiments were conducted comparing the performance of the system with human performance, which is normally used to deal with this problem. The proposed system demonstrated promising results and improvements are presented.},
 bibtype = {article},
 author = {Zanchettin, C. and Almeida, L.M. and De Menezes, F.D.},
 doi = {10.1109/JSEN.2014.2345476},
 journal = {IEEE Sensors Journal},
 number = {1}
}

© 2001-2012 IEEE. In this paper, we present the design of an intelligent monitoring system consisting of physical sensors and intelligent software for the automatic identification of the concentration of natural gas odorants in the environment. An optical-based sensor array was proposed comprising the hardware module. The software module employs wavelets filters and artificial neural networks to recognize the concentration of odorant in a natural gas sample. The objective is to help the natural gas odorization process by means of end point monitoring through the recognizing of the odorant concentration. The recognizing process uses a benchmark index, which measures the degrees of human perception of gas in the environment. In this way, the proposed system tries to mimic the human perception of a natural gas leak and helps one to indicate if more or less amount of odorant should be added into the gas pipeline. Experiments were conducted comparing the performance of the system with human performance, which is normally used to deal with this problem. The proposed system demonstrated promising results and improvements are presented.

@inproceedings{
 title = {Face recognition based on global and local features},
 type = {inproceedings},
 year = {2014},
 keywords = {2D-DCT,Artificial neural nerworks,Gabor filters,Wavelets},
 id = {4c9a3a83-cd29-3244-b38a-966860683ced},
 created = {2019-02-14T18:02:00.041Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.041Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper presents an evaluation of different methods considering the usually problems in face recognition. We consider variations in illumination, facial expression and facial details to propose a new method combining global and local face image features. This approach combines PCA, 2D-DCT and Gabor Wavelet Transform to obtain the global and local features representation. The Nearest Neighbor using the Euclidean distance performs the classification. The experiments were performed in the classical ORL and Yale face recognition databases. The proposed approach presented interesting results in comparison with the literature methods. Copyright 2014 ACM.},
 bibtype = {inproceedings},
 author = {Zanchettin, C.},
 doi = {10.1145/2554850.2555078},
 booktitle = {Proceedings of the ACM Symposium on Applied Computing}
}

This paper presents an evaluation of different methods considering the usually problems in face recognition. We consider variations in illumination, facial expression and facial details to propose a new method combining global and local face image features. This approach combines PCA, 2D-DCT and Gabor Wavelet Transform to obtain the global and local features representation. The Nearest Neighbor using the Euclidean distance performs the classification. The experiments were performed in the classical ORL and Yale face recognition databases. The proposed approach presented interesting results in comparison with the literature methods. Copyright 2014 ACM.

@article{
 title = {Advances in intelligent systems},
 type = {article},
 year = {2014},
 volume = {127},
 id = {f5bb2eb3-37ca-37ba-8af3-cd11b3fb9c00},
 created = {2019-02-14T18:02:00.925Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.925Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 bibtype = {article},
 author = {Ludermir, T.B. and Zanchettin, C. and Carolina Lorena, A.},
 doi = {10.1016/j.neucom.2013.07.040},
 journal = {Neurocomputing}
}

@inproceedings{
 title = {Handwriting recognition system for mobile accessibility to the visually impaired people},
 type = {inproceedings},
 year = {2014},
 volume = {2014-Janua},
 issue = {January},
 id = {0b9eef90-82bb-3a9e-9fca-4a327a5d3909},
 created = {2019-02-14T18:02:01.156Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.156Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2014 IEEE. This paper proposes the combination of preprocessing and handwriting recognition approaches aiming to develop a flexible and assistive mobile tool to help visually impaired people to understand and interact with handwriting text. The proposed system is described and its performance is evaluated in the IAM Handwriting Database. The system presented promising results and may aid visually impaired to overcome an everyday accessibility barrier as handwriting recognition.},
 bibtype = {inproceedings},
 author = {Gouveia, F.M. and Bezerra, B.L.D. and Zanchettin, C. and Meneses, J.R.J.},
 doi = {10.1109/SMC.2014.6974543},
 booktitle = {Conference Proceedings - IEEE International Conference on Systems, Man and Cybernetics}
}

© 2014 IEEE. This paper proposes the combination of preprocessing and handwriting recognition approaches aiming to develop a flexible and assistive mobile tool to help visually impaired people to understand and interact with handwriting text. The proposed system is described and its performance is evaluated in the IAM Handwriting Database. The system presented promising results and may aid visually impaired to overcome an everyday accessibility barrier as handwriting recognition.

@inproceedings{
 title = {An adaptive thresholding algorithm based on edge detection and morphological operations for document images},
 type = {inproceedings},
 year = {2013},
 keywords = {binarization,handwritten document,segmentation,thresholding},
 id = {5463a4cc-ffd0-3637-8477-754531cbf5f9},
 created = {2019-02-14T18:02:01.037Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.037Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper presents a new algorithm to threshold document images. The proposed algorithm deal with complex background images, illumination and aspect variants, back-to-front interference, variation of brightness and different positioned shadows. The algorithm have two phases. The first one uses edge detection and morphological operations to identify the text on the image. The second phase uses the positions of the text to define the threshold value in an adaptive process. Our approach presents promising results in images with complex background released from the Document Image Binarization Contest (DIBCO) when compared with other literature and competition thresholding algorithms. © 2013 ACM.},
 bibtype = {inproceedings},
 author = {Neves, R.F.D.P. and Zanchettin, C. and Mello, C.A.B.},
 doi = {10.1145/2494266.2494301},
 booktitle = {DocEng 2013 - Proceedings of the 2013 ACM Symposium on Document Engineering}
}

This paper presents a new algorithm to threshold document images. The proposed algorithm deal with complex background images, illumination and aspect variants, back-to-front interference, variation of brightness and different positioned shadows. The algorithm have two phases. The first one uses edge detection and morphological operations to identify the text on the image. The second phase uses the positions of the text to define the threshold value in an adaptive process. Our approach presents promising results in images with complex background released from the Document Image Binarization Contest (DIBCO) when compared with other literature and competition thresholding algorithms. © 2013 ACM.

@inproceedings{
 title = {Metaclasses and zoning for handwritten document recognition},
 type = {inproceedings},
 year = {2013},
 id = {4cbfcb2c-fcab-3eb9-8cd5-80d84343533e},
 created = {2019-02-14T18:02:01.563Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.563Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This work presents a complete method for improving the handwritten document recognition. In this task some characters are confused with others because of their visual/structural similarity. A SOM and TreeSOM neural network were used to sort different characters in metaclasses. In each metaclass a zoning approach was applied trying to get particular features to improve the character classification. The experiments with this new approach were performed in the NIST database with the classic MLP and a fast neural network RBF-DDA. © 2013 IEEE.},
 bibtype = {inproceedings},
 author = {Macário, V. and Silva, G.F.P. and Souza, M.R.P. and Zanchettin, C. and Cavalcanti, G.D.C.},
 doi = {10.1109/IJCNN.2013.6707056},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

This work presents a complete method for improving the handwritten document recognition. In this task some characters are confused with others because of their visual/structural similarity. A SOM and TreeSOM neural network were used to sort different characters in metaclasses. In each metaclass a zoning approach was applied trying to get particular features to improve the character classification. The experiments with this new approach were performed in the NIST database with the classic MLP and a fast neural network RBF-DDA. © 2013 IEEE.

@inproceedings{
 title = {Offline handwritten signature verification through network radial basis functions optimized by Differential Evolution},
 type = {inproceedings},
 year = {2012},
 keywords = {Differential Evolution,Handwritten Signature,Off-line Verification,Radial Basis Function},
 id = {16bad803-6544-3f90-99e6-5a3e358e4a07},
 created = {2019-02-14T18:02:00.156Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.156Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The handwritten signature is present in all important documents. In law, if the signature on a document is false, this document is also considered a fraud. This paper uses a neural network of radial basis function optimized by Differential Evolution Algorithm with features that best discriminates between a genuine signature of a simulated forgery. The experiments with this promising technique were made with a GPDS-300gray images base and the results subjected to statistical tests with the performance of technical literature. © 2012 IEEE.},
 bibtype = {inproceedings},
 author = {De Medeiros Nápoles, S.H.L. and Zanchettin, C.},
 doi = {10.1109/IJCNN.2012.6252720},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

The handwritten signature is present in all important documents. In law, if the signature on a document is false, this document is also considered a fraud. This paper uses a neural network of radial basis function optimized by Differential Evolution Algorithm with features that best discriminates between a genuine signature of a simulated forgery. The experiments with this promising technique were made with a GPDS-300gray images base and the results subjected to statistical tests with the performance of technical literature. © 2012 IEEE.

@inproceedings{
 title = {A hybrid RNN model for cursive offline handwriting recognition},
 type = {inproceedings},
 year = {2012},
 id = {515359b7-aff5-3def-84d8-10c89c7f67b0},
 created = {2019-02-14T18:02:00.970Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.970Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper presents an approach to handwriting character recognition using recurrent neural networks. The method Multi-dimensional Recurrent Neural Network is evaluated against the classical techniques. To improve the model performance we propose the use of specialized Support Vector Machine combined with the original MDRNN in cases of confusion letters to avoid misclassifications. The performance of the method is verified in the C-Cube database and compared with different classifiers. The hierarchical combination presented promising results. © 2012 IEEE.},
 bibtype = {inproceedings},
 author = {Bezerra, B.L.D. and Zanchettin, C. and Braga De Andrade, V.},
 doi = {10.1109/SBRN.2012.41},
 booktitle = {Proceedings - Brazilian Symposium on Neural Networks, SBRN}
}

This paper presents an approach to handwriting character recognition using recurrent neural networks. The method Multi-dimensional Recurrent Neural Network is evaluated against the classical techniques. To improve the model performance we propose the use of specialized Support Vector Machine combined with the original MDRNN in cases of confusion letters to avoid misclassifications. The performance of the method is verified in the C-Cube database and compared with different classifiers. The hierarchical combination presented promising results. © 2012 IEEE.

@book{
 title = {An efficient way of combining SVMs for handwritten digit recognition},
 type = {book},
 year = {2012},
 source = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
 keywords = {handwriting digit classifier,pattern recognition,support vector machine},
 volume = {7553 LNCS},
 issue = {PART 2},
 id = {c0dc3f0e-e52a-340a-9509-7619234c23d6},
 created = {2019-02-14T18:02:00.995Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.995Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper presents a method of combining SVMs (support vector machines) for multiclass problems that ensures a high recognition rate and a short processing time when compared to other classifiers. This hierarchical SVM combination considers the high recognition rate and short processing time as evaluation criteria. The used case study was the handwritten digit recognition problem with promising results. © 2012 Springer-Verlag.},
 bibtype = {book},
 author = {Neves, R.F.P. and Zanchettin, C. and Filho, A.N.G.L.},
 doi = {10.1007/978-3-642-33266-1_29}
}

This paper presents a method of combining SVMs (support vector machines) for multiclass problems that ensures a high recognition rate and a short processing time when compared to other classifiers. This hierarchical SVM combination considers the high recognition rate and short processing time as evaluation criteria. The used case study was the handwritten digit recognition problem with promising results. © 2012 Springer-Verlag.

@book{
 title = {A MDRNN-SVM hybrid model for cursive offline handwriting recognition},
 type = {book},
 year = {2012},
 source = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
 volume = {7553 LNCS},
 issue = {PART 2},
 id = {6b68800f-c7a4-3d31-9a6c-7432acc50567},
 created = {2019-02-14T18:02:01.291Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.291Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper presents a recurrent neural networks applied to handwriting character recognition. The method Multi-dimensional Recurrent Neural Network is evaluated against classical techniques. To improve the model performance we propose the use of specialized Support Vector Machine combined whit the original Multi-dimensional Recurrent Neural Network in cases of confusion letters. The experiments were performed in the C-Cube database and compared with different classifiers. The hierarchical combination presented promising results. © 2012 Springer-Verlag.},
 bibtype = {book},
 author = {Bezerra, B.L.D. and Zanchettin, C. and De Andrade, V.B.},
 doi = {10.1007/978-3-642-33266-1_31}
}

This paper presents a recurrent neural networks applied to handwriting character recognition. The method Multi-dimensional Recurrent Neural Network is evaluated against classical techniques. To improve the model performance we propose the use of specialized Support Vector Machine combined whit the original Multi-dimensional Recurrent Neural Network in cases of confusion letters. The experiments were performed in the C-Cube database and compared with different classifiers. The hierarchical combination presented promising results. © 2012 Springer-Verlag.

@inproceedings{
 title = {A KNN-SVM hybrid model for cursive handwriting recognition},
 type = {inproceedings},
 year = {2012},
 id = {65ff4511-3677-37d6-9c6a-2e7686462624},
 created = {2019-02-14T18:02:01.407Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.407Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper presents a hybrid KNN-SVM method for cursive character recognition. Specialized Support Vector Machines (SVMs) are introduced to significantly improve the performance of KNN in handwrite recognition. This hybrid approach is based on the observation that when using KNN in the task of handwritten characters recognition, the correct class is almost always one of the two nearest neighbors of the KNN. Specialized local SVMs are introduced to detect the correct class among these two different classification hypotheses. The hybrid KNN-SVM recognizer showed significant improvement in terms of recognition rate compared with MLP, KNN and a hybrid MLP-SVM approach for a task of character recognition. © 2012 IEEE.},
 bibtype = {inproceedings},
 author = {Zanchettin, C. and Bezerra, B.L.D. and Azevedo, W.W.},
 doi = {10.1109/IJCNN.2012.6252719},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

This paper presents a hybrid KNN-SVM method for cursive character recognition. Specialized Support Vector Machines (SVMs) are introduced to significantly improve the performance of KNN in handwrite recognition. This hybrid approach is based on the observation that when using KNN in the task of handwritten characters recognition, the correct class is almost always one of the two nearest neighbors of the KNN. Specialized local SVMs are introduced to detect the correct class among these two different classification hypotheses. The hybrid KNN-SVM recognizer showed significant improvement in terms of recognition rate compared with MLP, KNN and a hybrid MLP-SVM approach for a task of character recognition. © 2012 IEEE.

@inproceedings{
 title = {Odor recognition systems for natural gas odorization monitoring},
 type = {inproceedings},
 year = {2012},
 id = {41379446-f0d3-38ea-a5dd-ba55129e23a3},
 created = {2019-02-14T18:02:01.618Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.618Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper presents a system consisting of physical sensors and intelligent software for the automatic identification of the concentration of natural gas odorants and details the development of the sensor and pattern recognition systems. The sensor system uses spectroscopic technology and the pattern recognition system uses wavelet and artificial neural network technology. The aim is to determine the concentration of a natural gas odorant in the environment and associate this concentration with the benchmark index, which measures the degree of human perception to the presence of gas in the environment. Experiments were conducted comparing the performance of the system with human performance, which is normally used to deal with this problem. The proposed system demonstrated promising results. © 2012 IEEE.},
 bibtype = {inproceedings},
 author = {Zanchettin, C. and Almeida, L.M. and Menezes, F.D. and Ludermir, T.B. and Azevedo, W.M.},
 doi = {10.1109/IJCNN.2012.6252601},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

This paper presents a system consisting of physical sensors and intelligent software for the automatic identification of the concentration of natural gas odorants and details the development of the sensor and pattern recognition systems. The sensor system uses spectroscopic technology and the pattern recognition system uses wavelet and artificial neural network technology. The aim is to determine the concentration of a natural gas odorant in the environment and associate this concentration with the benchmark index, which measures the degree of human perception to the presence of gas in the environment. Experiments were conducted comparing the performance of the system with human performance, which is normally used to deal with this problem. The proposed system demonstrated promising results. © 2012 IEEE.

@inproceedings{
 title = {A multi-layer perceptron approach to threshold documents with complex background},
 type = {inproceedings},
 year = {2011},
 keywords = {Document Image Processing,Image Processing,Neural Networks},
 id = {7e1571de-abca-37b3-9790-21fcc2d5e68b},
 created = {2019-02-14T18:02:00.888Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.888Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper describes a thresholding method based on a Multi-Layer Perceptron approach for documents with complex backgrounds. The study case is focused on two regions of Brazilian bank checks: the courtesy amount and the character magnetic code. Those images have complex backgrounds with different patterns, which is a problem for an automatic recognition system. The new approach is based on a connectionistic approach to find the best threshold value. The proposed method is compared to ten thresholding algorithms (classical and specific for bank checks) in three different real bank checks databases, according to different evaluation metrics (recognition rate, peak signal-to-noise ratio, mean square error, precision, recall, accuracy, specificity, negative rate metric, misclassification penalty metric and f-measure). Based on the results, we may conclude the proposed method is more robust to variations in the image acquiring process, which influences the contrast, bright, hue and amount of noise verified in the image. © 2011 IEEE.},
 bibtype = {inproceedings},
 author = {Rabelo, J.C.B. and Zanchettin, C. and Mello, C.A.B. and Bezerra, B.L.D.},
 doi = {10.1109/ICSMC.2011.6084056},
 booktitle = {Conference Proceedings - IEEE International Conference on Systems, Man and Cybernetics}
}

This paper describes a thresholding method based on a Multi-Layer Perceptron approach for documents with complex backgrounds. The study case is focused on two regions of Brazilian bank checks: the courtesy amount and the character magnetic code. Those images have complex backgrounds with different patterns, which is a problem for an automatic recognition system. The new approach is based on a connectionistic approach to find the best threshold value. The proposed method is compared to ten thresholding algorithms (classical and specific for bank checks) in three different real bank checks databases, according to different evaluation metrics (recognition rate, peak signal-to-noise ratio, mean square error, precision, recall, accuracy, specificity, negative rate metric, misclassification penalty metric and f-measure). Based on the results, we may conclude the proposed method is more robust to variations in the image acquiring process, which influences the contrast, bright, hue and amount of noise verified in the image. © 2011 IEEE.

@article{
 title = {Hybrid training method for MLP: Optimization of architecture and training},
 type = {article},
 year = {2011},
 keywords = {Genetic algorithms (GAs),multilayer perceptron (MLP),optimization,simulating annealing,tabu search (TS)},
 volume = {41},
 id = {9ed6cd4c-a72f-3d6e-a486-a4124513e4c2},
 created = {2019-02-14T18:02:01.013Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.013Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The performance of an artificial neural network (ANN) depends upon the selection of proper connection weights, network architecture, and cost function during network training. This paper presents a hybrid approach (GaTSa) to optimize the performance of the ANN in terms of architecture and weights. GaTSa is an extension of a previous method (TSa) proposed by the authors. GaTSa is based on the integration of the heuristic simulated annealing (SA), tabu search (TS), genetic algorithms (GA), and backpropagation, whereas TSa does not use GA. The main advantages of GaTSa are the following: a constructive process to add new nodes in the architecture based on GA, the ability to escape from local minima with uphill moves (SA feature), and faster convergence by the evaluation of a set of solutions (TS feature). The performance of GaTSa is investigated through an empirical evaluation of 11 public-domain data sets using different cost functions in the simultaneous optimization of the multilayer perceptron ANN architecture and weights. Experiments demonstrated that GaTSa can also be used for relevant feature selection. GaTSa presented statistically relevant results in comparison with other global and local optimization techniques. © 2011 IEEE.},
 bibtype = {article},
 author = {Zanchettin, C. and Ludermir, T.B. and Almeida, L.M.},
 doi = {10.1109/TSMCB.2011.2107035},
 journal = {IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics},
 number = {4}
}

The performance of an artificial neural network (ANN) depends upon the selection of proper connection weights, network architecture, and cost function during network training. This paper presents a hybrid approach (GaTSa) to optimize the performance of the ANN in terms of architecture and weights. GaTSa is an extension of a previous method (TSa) proposed by the authors. GaTSa is based on the integration of the heuristic simulated annealing (SA), tabu search (TS), genetic algorithms (GA), and backpropagation, whereas TSa does not use GA. The main advantages of GaTSa are the following: a constructive process to add new nodes in the architecture based on GA, the ability to escape from local minima with uphill moves (SA feature), and faster convergence by the evaluation of a set of solutions (TS feature). The performance of GaTSa is investigated through an empirical evaluation of 11 public-domain data sets using different cost functions in the simultaneous optimization of the multilayer perceptron ANN architecture and weights. Experiments demonstrated that GaTSa can also be used for relevant feature selection. GaTSa presented statistically relevant results in comparison with other global and local optimization techniques. © 2011 IEEE.

@inproceedings{
 title = {A SVM based off-line handwritten digit recognizer},
 type = {inproceedings},
 year = {2011},
 keywords = {Handwritten Digit Recognizer,MLP,OCR,SVM},
 id = {c6224484-3c93-3eb0-b612-6691a5efba8c},
 created = {2019-02-14T18:02:01.244Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.244Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper presents an efficient method for handwritten digit recognition. The proposed method makes use of Support Vector Machines (SVM), benefitting from its generalization power. The method presents improved recognition rates when compared to Multi-Layer Perceptron (MLP) classifiers, other SVM classifiers and hybrid classifiers. Experiments and comparisons were done using a digit set extracted from the NIST SD19 digit database. The proposed SVM method achieved higher recognition rates and it outperformed other methods. It is also shown that although using solely SVMs for the task, the new method does not suffer when considering processing time. © 2011 IEEE.},
 bibtype = {inproceedings},
 author = {Neves, R.F.P. and Lopes Filho, A.N.G. and Mello, C.A.B. and Zanchettin, C.},
 doi = {10.1109/ICSMC.2011.6083734},
 booktitle = {Conference Proceedings - IEEE International Conference on Systems, Man and Cybernetics}
}

This paper presents an efficient method for handwritten digit recognition. The proposed method makes use of Support Vector Machines (SVM), benefitting from its generalization power. The method presents improved recognition rates when compared to Multi-Layer Perceptron (MLP) classifiers, other SVM classifiers and hybrid classifiers. Experiments and comparisons were done using a digit set extracted from the NIST SD19 digit database. The proposed SVM method achieved higher recognition rates and it outperformed other methods. It is also shown that although using solely SVMs for the task, the new method does not suffer when considering processing time. © 2011 IEEE.

@inproceedings{
 title = {Evolving clonal adaptive resonance theory based on ECOS theory},
 type = {inproceedings},
 year = {2011},
 id = {32efd16c-33b0-366f-8fc5-32091da3ef00},
 created = {2019-02-14T18:02:01.346Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.346Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The present work describes an evolution of the hybrid immune approach called Clonart (Clonal Adaptive Resonance Theory) using ECOS (Evolving Connectionist Systems) architectures. Some improvements were developed to allow the control of the growth of the clusters. Clonart's architecture is an Evolutionary Algorithm biologically inspired on the use of the Clonal Selection Principle. Therefore, a technique inspired on ART 1 network was combined to store the best antibodies. However, these strategies may create a lot of clusters due to the ART behavior. For that reason, techniques of insertion, aggregation and pruning inspired on ECOS operation were used to control the amount of clusters in Clonart. In this way, old and unnecessary clusters may confuse the Clonart and increase the learning error rate. This behavior was especially important, because many problems need constant retraining. The effectiveness of this approach was evaluated using ten databases from UCI Machine Learning Repository. © 2011 IEEE.},
 bibtype = {inproceedings},
 author = {Alexandrino, J. and Zanchettin, C. and Carvalho Filho, E.},
 doi = {10.1109/IJCNN.2011.6033477},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

The present work describes an evolution of the hybrid immune approach called Clonart (Clonal Adaptive Resonance Theory) using ECOS (Evolving Connectionist Systems) architectures. Some improvements were developed to allow the control of the growth of the clusters. Clonart's architecture is an Evolutionary Algorithm biologically inspired on the use of the Clonal Selection Principle. Therefore, a technique inspired on ART 1 network was combined to store the best antibodies. However, these strategies may create a lot of clusters due to the ART behavior. For that reason, techniques of insertion, aggregation and pruning inspired on ECOS operation were used to control the amount of clusters in Clonart. In this way, old and unnecessary clusters may confuse the Clonart and increase the learning error rate. This behavior was especially important, because many problems need constant retraining. The effectiveness of this approach was evaluated using ten databases from UCI Machine Learning Repository. © 2011 IEEE.

@article{
 title = {Design of experiments in neuro-fuzzy systems},
 type = {article},
 year = {2010},
 keywords = {Adaptive neuro fuzzy inference system,Design of experiments,Evolving fuzzy neural networks,Neuro fuzzy systems},
 volume = {9},
 id = {a1d49810-fe5d-32cf-9cdf-a1f72218e7d0},
 created = {2019-02-14T18:02:01.139Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.139Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Interest in hybrid methods that combine artificial neural networks and fuzzy inference systems has grown in recent years. These systems are robust solutions that search for representations of domain knowledge, reasoning on uncertainty, automatic learning and adaptation. However, the design and definition of the parameter effectiveness of such systems is still a hard task. In the present work, we perform a statistical analysis to verify interactions and interrelations between parameters in the design of neuro-fuzzy systems. The analysis is carried out using a powerful statistical tool, namely, Design of Experiments (DOE), in two neuro-fuzzy models Adaptive Neuro Fuzzy Inference System (ANFIS) and Evolving Fuzzy Neural Networks (EFuNN). The results show that, for ANFIS, input MFs number and output MFs shape are usually the factors with the largest influence on the system's RMSE. For EFFuNN, the MF shape and the interaction between MF shape and number usually have the largest effect size. © 2010 Imperial College Press.},
 bibtype = {article},
 author = {Zanchettin, C. and Minku, L.L. and Ludermir, T.B.},
 doi = {10.1142/S1469026810002823},
 journal = {International Journal of Computational Intelligence and Applications},
 number = {2}
}

Interest in hybrid methods that combine artificial neural networks and fuzzy inference systems has grown in recent years. These systems are robust solutions that search for representations of domain knowledge, reasoning on uncertainty, automatic learning and adaptation. However, the design and definition of the parameter effectiveness of such systems is still a hard task. In the present work, we perform a statistical analysis to verify interactions and interrelations between parameters in the design of neuro-fuzzy systems. The analysis is carried out using a powerful statistical tool, namely, Design of Experiments (DOE), in two neuro-fuzzy models Adaptive Neuro Fuzzy Inference System (ANFIS) and Evolving Fuzzy Neural Networks (EFuNN). The results show that, for ANFIS, input MFs number and output MFs shape are usually the factors with the largest influence on the system's RMSE. For EFFuNN, the MF shape and the interaction between MF shape and number usually have the largest effect size. © 2010 Imperial College Press.

@inproceedings{
 title = {Hybrid optimization technique for artificial neural networks design},
 type = {inproceedings},
 year = {2009},
 keywords = {Artificial neural networks,Experimental design,Global optimization,Relevant feature selection},
 volume = {AIDSS},
 id = {dc0c906d-2a03-38e2-8725-885d2bac908f},
 created = {2019-02-14T18:02:00.092Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.092Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {In this paper a global and local optimization method is presented. This method is based on the integration of the heuristic Simulated Annealing, Tabu Search, Genetic Algorithms and Backpropagation. The performance of the method is investigated in the optimization of Multi-layer Perceptron artificial neural network architecture and weights. The heuristics perform the search in a constructive way and based on the pruning of irrelevant connections among the network nodes. Experiments demonstrated that the method can also be used for relevant feature selection. Experiments are performed with four classification and one prediction datasets.},
 bibtype = {inproceedings},
 author = {Zanchettin, C. and Ludermir, T.B.},
 booktitle = {ICEIS 2009 - 11th International Conference on Enterprise Information Systems, Proceedings}
}

In this paper a global and local optimization method is presented. This method is based on the integration of the heuristic Simulated Annealing, Tabu Search, Genetic Algorithms and Backpropagation. The performance of the method is investigated in the optimization of Multi-layer Perceptron artificial neural network architecture and weights. The heuristics perform the search in a constructive way and based on the pruning of irrelevant connections among the network nodes. Experiments demonstrated that the method can also be used for relevant feature selection. Experiments are performed with four classification and one prediction datasets.

@inproceedings{
 title = {Hybrid intelligent immune system using Radial Basis Function applied to Time Series Analysis},
 type = {inproceedings},
 year = {2009},
 id = {d2259f02-6332-3dc4-8930-bd80ff6e31ca},
 created = {2019-02-14T18:02:01.463Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.463Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The present work proposes an integration of Clonal Adaptive Resonance Theory framework (Clonart) with Radial Basis Function (RBF) called ClonalRBF. This framework was already used in a handwritten digit classification problem, a forecasting for the Brazilian Energy Distribution System and now a Time Series Analysis in Gas Furnace and Mackey-Glass databases. In Clonart, the population memory was organized using an ART 1 network and in the new approach it was organized using a RBF network. This framework has biologically inspired characteristics like the grouping of similar antibodies and memory antibodies. It was studied to allow the evolution of the artificial immune system. The focus of this study was to evaluate the ClonalRBF and to compare with Clonart using these two databases. © 2009 IEEE.},
 bibtype = {inproceedings},
 author = {Alexandrino, J.L. and Zanchettin, C. and Filho, E.C.B.C.},
 doi = {10.1109/IJCNN.2009.5178944},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

The present work proposes an integration of Clonal Adaptive Resonance Theory framework (Clonart) with Radial Basis Function (RBF) called ClonalRBF. This framework was already used in a handwritten digit classification problem, a forecasting for the Brazilian Energy Distribution System and now a Time Series Analysis in Gas Furnace and Mackey-Glass databases. In Clonart, the population memory was organized using an ART 1 network and in the new approach it was organized using a RBF network. This framework has biologically inspired characteristics like the grouping of similar antibodies and memory antibodies. It was studied to allow the evolution of the artificial immune system. The focus of this study was to evaluate the ClonalRBF and to compare with Clonart using these two databases. © 2009 IEEE.

@inproceedings{
 title = {Feature subset selection in a methodology for training and improving artificial neural network weights and connections},
 type = {inproceedings},
 year = {2008},
 id = {0df7d1ea-4d8c-3c25-ace6-ca142227866c},
 created = {2019-02-14T18:02:00.345Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.345Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper investigates the problem of feature subset selection as part of a methodology that integrates heuristic tabu search, simulated annealing, genetic algorithms and backpropagation. This technique combines both global and local search strategies for the simultaneous optimization of the number of connections and connection values of Multi-Layer Perceptron neural networks. We compare the performance of the proposed method for feature subset selection to five classical feature selection methods in three different classification problems. © 2008 IEEE.},
 bibtype = {inproceedings},
 author = {Zanchettin, C. and Ludermir, T.B.},
 doi = {10.1109/IJCNN.2008.4634065},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

This paper investigates the problem of feature subset selection as part of a methodology that integrates heuristic tabu search, simulated annealing, genetic algorithms and backpropagation. This technique combines both global and local search strategies for the simultaneous optimization of the number of connections and connection values of Multi-Layer Perceptron neural networks. We compare the performance of the proposed method for feature subset selection to five classical feature selection methods in three different classification problems. © 2008 IEEE.

@inproceedings{
 title = {A hybrid intelligent system clonart for short and mid-term forecasting for the brazilian energy distribution system},
 type = {inproceedings},
 year = {2008},
 id = {8fd4ac6d-baf3-39ee-97cf-7c914fbbe6d2},
 created = {2019-02-14T18:02:01.111Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.111Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The present work describes an application of Clonart (Clonal Adaptive Resonance Theory) for forecasting of amount of precipitation for the Brazilian Energy Distribution System. The effectiveness of the Brazilian electricity system directly depends on the difference between hydroelectric energy production and consumer use. Production depends upon the volume of water stored in the reservoirs. A forecasting system for the amount of rainfall throughout the year contributes significantly to the analysis. The plasticity of the Clonart ensures that a new piece of knowledge does not overshadow previous knowledge. This is especially important for forecast problems because this type of problem needs constants training. © 2008 IEEE.},
 bibtype = {inproceedings},
 author = {Alexandrino, J.L. and Zanchettin, C. and De Barros Carvalho Filho, E.C.},
 doi = {10.1109/IJCNN.2008.4634295},
 booktitle = {Proceedings of the International Joint Conference on Neural Networks}
}

The present work describes an application of Clonart (Clonal Adaptive Resonance Theory) for forecasting of amount of precipitation for the Brazilian Energy Distribution System. The effectiveness of the Brazilian electricity system directly depends on the difference between hydroelectric energy production and consumer use. Production depends upon the volume of water stored in the reservoirs. A forecasting system for the amount of rainfall throughout the year contributes significantly to the analysis. The plasticity of the Clonart ensures that a new piece of knowledge does not overshadow previous knowledge. This is especially important for forecast problems because this type of problem needs constants training. © 2008 IEEE.

@article{
 title = {Wavelet filter for noise reduction and signal compression in an artificial nose},
 type = {article},
 year = {2007},
 keywords = {Artificial neural networks,Artificial noses,Wavelet transform},
 volume = {7},
 id = {1c3627ef-faea-3a4d-80f8-5a5b9852c795},
 created = {2019-02-14T18:02:00.144Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.144Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This work presents results of the use of a wavelet filter for noise reduction and data compression of signals generated by artificial nose sensors. To verify the performance of the wavelet analysis in the treatment of odor patterns, we compare two widely used artificial nose classifiers, multi-layer perceptron neural network and time delay neural network in the analysis of signals generated by eight conducting polymer sensors exposed to gases derived from the petroliferous industry. © 2005 Elsevier B.V. All rights reserved.},
 bibtype = {article},
 author = {Zanchettin, C. and Ludermir, T.B.},
 doi = {10.1016/j.asoc.2005.06.004},
 journal = {Applied Soft Computing Journal},
 number = {1}
}

This work presents results of the use of a wavelet filter for noise reduction and data compression of signals generated by artificial nose sensors. To verify the performance of the wavelet analysis in the treatment of odor patterns, we compare two widely used artificial nose classifiers, multi-layer perceptron neural network and time delay neural network in the analysis of signals generated by eight conducting polymer sensors exposed to gases derived from the petroliferous industry. © 2005 Elsevier B.V. All rights reserved.

@inproceedings{
 title = {Comparison of the effectiveness of different cost functions in global optimization techniques},
 type = {inproceedings},
 year = {2007},
 id = {c5215474-d39e-3e35-a7b0-201ddb73f289},
 created = {2019-02-14T18:02:00.462Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.462Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {It is present herein an evaluation of the effect of different cost functions on a methodology that integrates heuristic tabu search, simulated annealing, genetic algorithms and backpropagation. We investigated five cost function approaches: the average method, weighted average, weight-decay, multi-objective optimization, combined multi-objective and weight-decay. The weight-decay approach presented promising results in the optimization process. The experiments were performed in four classifications and one prediction problem. ©2007 IEEE.},
 bibtype = {inproceedings},
 author = {Zanchettin, C. and Ludermir, T.B.},
 doi = {10.1109/IJCNN.2007.4371385},
 booktitle = {IEEE International Conference on Neural Networks - Conference Proceedings}
}

It is present herein an evaluation of the effect of different cost functions on a methodology that integrates heuristic tabu search, simulated annealing, genetic algorithms and backpropagation. We investigated five cost function approaches: the average method, weighted average, weight-decay, multi-objective optimization, combined multi-objective and weight-decay. The weight-decay approach presented promising results in the optimization process. The experiments were performed in four classifications and one prediction problem. ©2007 IEEE.

@article{
 title = {Artificial immune system with art memory hibridization},
 type = {article},
 year = {2007},
 keywords = {Artificial immune systems,Hybrid intelligent systems and adaptive resonance},
 volume = {17},
 id = {8f3f0bea-c44f-39c8-b8ba-cc3af3ae59ec},
 created = {2019-02-14T18:02:01.241Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.241Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The present work proposes the architecture Clonart (Clonal Adaptive Resonance Theory), a Hybrid Model that employs techniques like intelligent operators, clonal selection principle, local search, memory antibodies and ART clusterization, in order to increase the performance of the algorithm. The approach uses a mechanism similar to the ART 1 network for storing a population of memory antibodies that will be responsible for the acquired knowledge of the algorithm. This characteristic allows the algorithm a self-organization of the antibodies in accordance with the complexity of the database. © ICS AS CR 2007.},
 bibtype = {article},
 author = {Alexandrino, J.L. and Zanchettin, C. and Carvalho Filho, E.C.D.B.},
 journal = {Neural Network World},
 number = {6}
}

The present work proposes the architecture Clonart (Clonal Adaptive Resonance Theory), a Hybrid Model that employs techniques like intelligent operators, clonal selection principle, local search, memory antibodies and ART clusterization, in order to increase the performance of the algorithm. The approach uses a mechanism similar to the ART 1 network for storing a population of memory antibodies that will be responsible for the acquired knowledge of the algorithm. This characteristic allows the algorithm a self-organization of the antibodies in accordance with the complexity of the database. © ICS AS CR 2007.

@inproceedings{
 title = {Artificial immune system with ART memory hibridization},
 type = {inproceedings},
 year = {2007},
 id = {bdce2b3d-cb0f-39fa-830c-0dcf88b81d05},
 created = {2019-02-14T18:02:01.466Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.466Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {The present work proposes the architecture Clonart (Clonal Adaptive Resonance Theory) that employs many different techniques like intelligent operators, clonal selection principle, local search, memory antibodies and ART clusterization in order to increase the performance of the algorithm. The approach uses a mechanism similar to the ART 1 network for storing a population of memory antibodies that will be responsible for the acquired knowledge of the algorithm. This characteristic allows the algorithm a self-organization of the antibodies in accordance with the complexity of the database used. © 2007 IEEE.},
 bibtype = {inproceedings},
 author = {Alexandrino, J.L. and Zanchettin, C. and De Barros Carvalho Filho, E.C.},
 doi = {10.1109/ICHIS.2007.4344028},
 booktitle = {Proceedings - 7th International Conference on Hybrid Intelligent Systems, HIS 2007}
}

The present work proposes the architecture Clonart (Clonal Adaptive Resonance Theory) that employs many different techniques like intelligent operators, clonal selection principle, local search, memory antibodies and ART clusterization in order to increase the performance of the algorithm. The approach uses a mechanism similar to the ART 1 network for storing a population of memory antibodies that will be responsible for the acquired knowledge of the algorithm. This characteristic allows the algorithm a self-organization of the antibodies in accordance with the complexity of the database used. © 2007 IEEE.

@inproceedings{
 title = {An efficient thresholding algorithm for Brazilian bank checks},
 type = {inproceedings},
 year = {2007},
 volume = {1},
 id = {3cd44268-354a-34ab-94f2-0e31ba79acf8},
 created = {2019-02-14T18:02:01.508Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.508Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {It is present herein an algorithm for thresholding images of bank checks. These images have complex background elements. Some of these patterns make very hard to distinguish between the text and the texture pattern defined by the bank. For the binarizing process, an adaptive global thresholding algorithm is proposed based on ROC curves and it is compared to several well-known algorithms. The images generated by the new algorithm achieved a hit rate of 97% for recognition of the CMC7 code. © 2007 IEEE.},
 bibtype = {inproceedings},
 author = {Mello, C.A.B. and Bezerra, B.L.D. and Zanchettin, C. and Macário, V.},
 doi = {10.1109/ICDAR.2007.4378702},
 booktitle = {Proceedings of the International Conference on Document Analysis and Recognition, ICDAR}
}

It is present herein an algorithm for thresholding images of bank checks. These images have complex background elements. Some of these patterns make very hard to distinguish between the text and the texture pattern defined by the bank. For the binarizing process, an adaptive global thresholding algorithm is proposed based on ROC curves and it is compared to several well-known algorithms. The images generated by the new algorithm achieved a hit rate of 97% for recognition of the CMC7 code. © 2007 IEEE.

@inproceedings{
 title = {The influence of different cost functions in global optimization techniques},
 type = {inproceedings},
 year = {2006},
 id = {e7c3b559-6416-3d92-8cb4-2142bde2ca51},
 created = {2019-02-14T18:02:00.091Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.091Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This work presents an evaluation of the effect of different cost functions in a methodology that integrates heuristic tabu search, simulated annealing, genetic algorithms and backpropagation. We investigated four cost function approaches: average method, weight-decay, multi-objective optimization, combined multi-objective and weight-decay. The weight-decay approach presented promising results in the simultaneous optimization of artificial neural network architecture and weights. The experiments were performed in four classifications and one prediction problem. © 2006 IEEE.},
 bibtype = {inproceedings},
 author = {Zanchettin, C. and Ludermir, T.B.},
 doi = {10.1109/SBRN.2006.42},
 booktitle = {Proceedings of the Ninth Brazilian Symposium on Neural Networks, SBRN'06}
}

This work presents an evaluation of the effect of different cost functions in a methodology that integrates heuristic tabu search, simulated annealing, genetic algorithms and backpropagation. We investigated four cost function approaches: average method, weight-decay, multi-objective optimization, combined multi-objective and weight-decay. The weight-decay approach presented promising results in the simultaneous optimization of artificial neural network architecture and weights. The experiments were performed in four classifications and one prediction problem. © 2006 IEEE.

@inproceedings{
 title = {A methodology to train and improve artificial neural networks' weights and connections},
 type = {inproceedings},
 year = {2006},
 id = {88990269-dd3e-3464-988a-71d62b6b7ce7},
 created = {2019-02-14T18:02:00.386Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.386Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This work presents a new methodology that integrates the heuristics tabu search, simulated annealing, genetic algorithms and backpropagation in a prunning and constructive way. The approach obtained promising results in the simultaneous optimization of artificial neural network architecture and weights. The experiments were performed in four classification and one prediction problem. © 2006 IEEE.},
 bibtype = {inproceedings},
 author = {Zanchettin, C. and Ludermir, T.B.},
 booktitle = {IEEE International Conference on Neural Networks - Conference Proceedings}
}

This work presents a new methodology that integrates the heuristics tabu search, simulated annealing, genetic algorithms and backpropagation in a prunning and constructive way. The approach obtained promising results in the simultaneous optimization of artificial neural network architecture and weights. The experiments were performed in four classification and one prediction problem. © 2006 IEEE.

@article{
 title = {An optimization methodology for neural network weights and architectures},
 type = {article},
 year = {2006},
 keywords = {Multilayer perceptron (MLP),Optimization of weights and architectures,Simulating annealing,Tabu search},
 volume = {17},
 id = {3d673406-d70d-3e1d-ac59-d3023aa026ff},
 created = {2019-02-14T18:02:01.197Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.197Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper introduces a methodology for neural network global optimization. The aim is the simultaneous optimization of multilayer perceptron (MLP) network weights and architectures, in order to generate topologies with few connections and high classification performance for any data sets. The approach combines the advantages of simulated annealing, tabu search and the backpropagation training algorithm in order to generate an automatic process for producing networks with high classification performance and low complexity. Experimental results obtained with four classification problems and one prediction problem has shown to be better than those obtained by the most commonly used optimization techniques. © 2006 IEEE.},
 bibtype = {article},
 author = {Ludermir, T.B. and Yamazaki, A. and Zanchettin, C.},
 doi = {10.1109/TNN.2006.881047},
 journal = {IEEE Transactions on Neural Networks},
 number = {6}
}

This paper introduces a methodology for neural network global optimization. The aim is the simultaneous optimization of multilayer perceptron (MLP) network weights and architectures, in order to generate topologies with few connections and high classification performance for any data sets. The approach combines the advantages of simulated annealing, tabu search and the backpropagation training algorithm in order to generate an automatic process for producing networks with high classification performance and low complexity. Experimental results obtained with four classification problems and one prediction problem has shown to be better than those obtained by the most commonly used optimization techniques. © 2006 IEEE.

@inproceedings{
 title = {A heuristic binarization algorithm for documents with complex background},
 type = {inproceedings},
 year = {2006},
 keywords = {Automatic bank check processing,Binarization,Document image processing,Documents with complex background},
 id = {450fe7a2-e38b-3806-901e-56a13c7752b9},
 created = {2019-02-14T18:02:01.803Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.803Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper proposes a new method for binarization of digital documents. The proposed approach performs binarization by using a heuristic algorithm with two different thresholds and the combination of the thresholded images. The method is suitable for binarization of complex background document images. In experiments, it obtained better results than classical techniques in the binarization of real bank checks. ©2006 IEEE.},
 bibtype = {inproceedings},
 author = {Cavalcanti, G.D.C. and Silva, E.F.A. and Zanchettin, C. and Bezerral, B.L.D. and Dórial, R.C. and Rabelo, J.C.B.},
 doi = {10.1109/ICIP.2006.312475},
 booktitle = {Proceedings - International Conference on Image Processing, ICIP}
}

This paper proposes a new method for binarization of digital documents. The proposed approach performs binarization by using a heuristic algorithm with two different thresholds and the combination of the thresholded images. The method is suitable for binarization of complex background document images. In experiments, it obtained better results than classical techniques in the binarization of real bank checks. ©2006 IEEE.

@article{
 title = {Hybrid neural systems for pattern recognition in artificial noses},
 type = {article},
 year = {2005},
 keywords = {Artificial nose,Evolving fuzzy neural networks,Hybrid neural systems,Multi-layer perceptron,Time delay neural networks,Wavelet filter},
 volume = {15},
 id = {fcafda8e-da99-3a51-907f-08e4af37367c},
 created = {2019-02-14T18:02:00.205Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.205Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This work examines the use of Hybrid Intelligent Systems in the pattern recognition system of an artificial nose. The connectionist approaches Multi-Layer Perception and Time Delay Neural Networks, and the hybrid approaches Feature-Weighted Detector and Evolving Neural Fuzzy Networks were investigated. A Wavelet Filter is evaluated as a preprocessing method for odor signals. The signals generated by an artificial nose were composed by an array of conducting polymer sensors and exposed to two different odor databases. © World Scientific Publishing Company.},
 bibtype = {article},
 author = {Zanchettin, C. and Ludermir, T.B.},
 doi = {10.1142/S0129065705000141},
 journal = {International Journal of Neural Systems},
 number = {1-2}
}

This work examines the use of Hybrid Intelligent Systems in the pattern recognition system of an artificial nose. The connectionist approaches Multi-Layer Perception and Time Delay Neural Networks, and the hybrid approaches Feature-Weighted Detector and Evolving Neural Fuzzy Networks were investigated. A Wavelet Filter is evaluated as a preprocessing method for odor signals. The signals generated by an artificial nose were composed by an array of conducting polymer sensors and exposed to two different odor databases. © World Scientific Publishing Company.

@book{
 title = {Hybrid technique for artificial neural network architecture and weight optimization},
 type = {book},
 year = {2005},
 source = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
 volume = {3721 LNAI},
 id = {d91a0a47-daab-3ea0-9d81-b9b2c8b2b73d},
 created = {2019-02-14T18:02:00.323Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.323Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This work presents a technique that integrates the heuristics tabu search, simulated annealing, genetic algorithms and backpropagation. This approach obtained promising results in the simultaneous optimization of the artificial neural network architecture and weights. © Springer-Verlag Berlin Heidelberg 2005.},
 bibtype = {book},
 author = {Zanchettin, C. and Ludermir, T.B.},
 doi = {10.1007/11564126_76}
}

This work presents a technique that integrates the heuristics tabu search, simulated annealing, genetic algorithms and backpropagation. This approach obtained promising results in the simultaneous optimization of the artificial neural network architecture and weights. © Springer-Verlag Berlin Heidelberg 2005.

@article{
 title = {Hybrid neural systems for recognition of patterns in artificial noses},
 type = {article},
 year = {2005},
 keywords = {Artificial Neural Networks,Artificial Nose,Hybrid Intelligent Systems},
 volume = {16},
 id = {b6db2f93-29eb-3795-9277-a312c5af4452},
 created = {2019-02-14T18:02:00.882Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.882Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This work investigates the use of Hybrid Intelligent Systems in the pattern recognition system of an artificial nose. The connectionist approaches Multi-Layer Perceptron and Time Delay Neural Networks; and the hybrid approaches Feature-weighted Detector and Evolving Neural Fuzzy Networks were investigated. A wavelet filter as preprocessing method of odors signals is evaluated. The signals generated by an artificial nose, composed by an array of conducting polymer sensors, exposed to two different odor databases.},
 bibtype = {article},
 author = {Zanchettin, C. and Ludermir, T.B.},
 journal = {Controle y Automacao},
 number = {2}
}

This work investigates the use of Hybrid Intelligent Systems in the pattern recognition system of an artificial nose. The connectionist approaches Multi-Layer Perceptron and Time Delay Neural Networks; and the hybrid approaches Feature-weighted Detector and Evolving Neural Fuzzy Networks were investigated. A wavelet filter as preprocessing method of odors signals is evaluated. The signals generated by an artificial nose, composed by an array of conducting polymer sensors, exposed to two different odor databases.

@inproceedings{
 title = {Design of experiments in neuro-fuzzy systems},
 type = {inproceedings},
 year = {2005},
 volume = {2005},
 id = {ceffd219-ac84-332e-8c1b-4a420d724246},
 created = {2019-02-14T18:02:01.422Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:01.422Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Interest in hybrid methods that combine artificial neural networks and fuzzy inference systems has grown in the last few years. These systems are robust solutions that search for representation of domain knowledge, reasoning on uncertainty, automatic learning and adaptation. However, the design and the definition of parameters effectiveness of these systems is a hard task yet. In this work we perform a statistical analysis to verify the interactions and interrelations between parameters in the design of neuro-fuzzy systems. The analysis carries out using a powerful statistical tool, the Design Of Experiments (DOE) in two neuro-fuzzy models, Adaptive Neuro Fuzzy Inference System (ANFIS) and Evolving Fuzzy Neural Networks (EFuNN). © 2005 IEEE.},
 bibtype = {inproceedings},
 author = {Zanchettin, C. and Minku, F.L. and Ludermir, T.B.},
 doi = {10.1109/ICHIS.2005.34},
 booktitle = {Proceedings - HIS 2005: Fifth International Conference on Hybrid Intelligent Systems}
}

Interest in hybrid methods that combine artificial neural networks and fuzzy inference systems has grown in the last few years. These systems are robust solutions that search for representation of domain knowledge, reasoning on uncertainty, automatic learning and adaptation. However, the design and the definition of parameters effectiveness of these systems is a hard task yet. In this work we perform a statistical analysis to verify the interactions and interrelations between parameters in the design of neuro-fuzzy systems. The analysis carries out using a powerful statistical tool, the Design Of Experiments (DOE) in two neuro-fuzzy models, Adaptive Neuro Fuzzy Inference System (ANFIS) and Evolving Fuzzy Neural Networks (EFuNN). © 2005 IEEE.

@inproceedings{
 title = {Evolving Fuzzy Neural Networks applied to odor recognition in an artificial nose},
 type = {inproceedings},
 year = {2004},
 volume = {1},
 id = {8af5016f-81df-3109-9c00-d7945386abea},
 created = {2019-02-14T18:02:00.194Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.194Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {A pattern recognition system using Evolving Fuzzy Neural Networks for an artificial nose is presented. The artificial nose is composed of an adaptive and on-line learning method. For the classification of gases derived from the petroliferous industry, the method presented achieves better results (mean classification error of 0.88%) than those obtained by Time Delay Neural Networks (10.54%).},
 bibtype = {inproceedings},
 author = {Zanchettin, C. and Ludermir, T.B.},
 booktitle = {IEEE International Conference on Neural Networks - Conference Proceedings}
}

A pattern recognition system using Evolving Fuzzy Neural Networks for an artificial nose is presented. The artificial nose is composed of an adaptive and on-line learning method. For the classification of gases derived from the petroliferous industry, the method presented achieves better results (mean classification error of 0.88%) than those obtained by Time Delay Neural Networks (10.54%).

@book{
 title = {Evolving fuzzy neural networks applied to odor recognition},
 type = {book},
 year = {2004},
 source = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
 volume = {3316},
 id = {2b7fdb9d-4a74-32c5-9464-efeac3df1dae},
 created = {2019-02-14T18:02:00.426Z},
 file_attached = {false},
 profile_id = {74e7d4ea-3dac-3118-aab9-511a5b337e8f},
 last_modified = {2019-02-14T18:02:00.426Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {This paper presents the use of Evolving Fuzzy Neural Networks as pattern recognition system for odor recognition in an artificial nose. In the classification of gases derived from the petroliferous industry, the method presented achieves better results (mean classification error of 0.88%) than those obtained by Multi-Layer Perceptron (13.88%) and Time Delay Neural Networks (10.54%). © Springer-Verlag Berlin Heidelberg 2004.},
 bibtype = {book},
 author = {Zanchettin, C. and Ludermir, T.B.}
}

This paper presents the use of Evolving Fuzzy Neural Networks as pattern recognition system for odor recognition in an artificial nose. In the classification of gases derived from the petroliferous industry, the method presented achieves better results (mean classification error of 0.88%) than those obtained by Multi-Layer Perceptron (13.88%) and Time Delay Neural Networks (10.54%). © Springer-Verlag Berlin Heidelberg 2004.