DeepSense: A Unified Deep Learning Framework for Time-Series Mobile Sensing Data Processing. Yao, S., Hu, S., Zhao, Y., Zhang, A., & Abdelzaher, T. Paper doi abstract bibtex Mobile sensing and computing applications usually require time-series inputs from sensors, such as accelerometers, gyroscopes, and magnetometers. Some applications, such as tracking, can use sensed acceleration and rate of rotation to calculate displacement based on physical system models. Other applications, such as ac-tivity recognition, extract manually designed features from sensor inputs for classification. Such applications face two challenges. On one hand, on-device sensor measurements are noisy. For many mobile applications, it is hard to find a distribution that exactly describes the noise in practice. Unfortunately, calculating target quantities based on physical system and noise models is only as accurate as the noise assumptions. Similarly, in classification ap-plications, although manually designed features have proven to be effective, it is not always straightforward to find the most robust features to accommodate diverse sensor noise patterns and het-erogeneous user behaviors. To this end, we propose DeepSense, a deep learning framework that directly addresses the aforemen-tioned noise and feature customization challenges in a unified man-ner. DeepSense integrates convolutional and recurrent neural net-works to exploit local interactions among similar mobile sensors, merge local interactions of different sensory modalities into global interactions, and extract temporal relationships to model signal dy-namics. DeepSense thus provides a general signal estimation and classification framework that accommodates a wide range of ap-plications. We demonstrate the effectiveness of DeepSense using three representative and challenging tasks: car tracking with mo-tion sensors, heterogeneous human activity recognition, and user identification with biometric motion analysis. DeepSense signifi-cantly outperforms the state-of-the-art methods for all three tasks. In addition, we show that DeepSense is feasible to implement on smartphones and embedded devices thanks to its moderate energy consumption and low latency.
@article{yaoDeepSenseUnifiedDeep2017,
title = {{{DeepSense}}: A {{Unified Deep Learning Framework}} for {{Time}}-{{Series Mobile Sensing Data Processing}}},
url = {http://dx.doi.org/10.1145/3038912.3052577},
doi = {10.1145/3038912.3052577},
abstract = {Mobile sensing and computing applications usually require time-series inputs from sensors, such as accelerometers, gyroscopes, and magnetometers. Some applications, such as tracking, can use sensed acceleration and rate of rotation to calculate displacement based on physical system models. Other applications, such as ac-tivity recognition, extract manually designed features from sensor inputs for classification. Such applications face two challenges. On one hand, on-device sensor measurements are noisy. For many mobile applications, it is hard to find a distribution that exactly describes the noise in practice. Unfortunately, calculating target quantities based on physical system and noise models is only as accurate as the noise assumptions. Similarly, in classification ap-plications, although manually designed features have proven to be effective, it is not always straightforward to find the most robust features to accommodate diverse sensor noise patterns and het-erogeneous user behaviors. To this end, we propose DeepSense, a deep learning framework that directly addresses the aforemen-tioned noise and feature customization challenges in a unified man-ner. DeepSense integrates convolutional and recurrent neural net-works to exploit local interactions among similar mobile sensors, merge local interactions of different sensory modalities into global interactions, and extract temporal relationships to model signal dy-namics. DeepSense thus provides a general signal estimation and classification framework that accommodates a wide range of ap-plications. We demonstrate the effectiveness of DeepSense using three representative and challenging tasks: car tracking with mo-tion sensors, heterogeneous human activity recognition, and user identification with biometric motion analysis. DeepSense signifi-cantly outperforms the state-of-the-art methods for all three tasks. In addition, we show that DeepSense is feasible to implement on smartphones and embedded devices thanks to its moderate energy consumption and low latency.},
date = {2017},
pages = {17--4},
keywords = {Activity Recognition,Deep Learning,Internet of Things,Mobile Computing,Mobile Sensing,Tracking,User Identification c 2017},
author = {Yao, Shuochao and Hu, Shaohan and Zhao, Yiran and Zhang, Aston and Abdelzaher, Tarek},
file = {/home/dimitri/Nextcloud/Zotero/storage/YXTTIG45/Yao et al. - 2016 - DeepSense A Unified Deep Learning Framework for Time-Series Mobile Sensing Data Processing.pdf}
}
Downloads: 0
{"_id":"9mG288gTBoSCgiHBH","bibbaseid":"yao-hu-zhao-zhang-abdelzaher-deepsenseaunifieddeeplearningframeworkfortimeseriesmobilesensingdataprocessing","authorIDs":[],"author_short":["Yao, S.","Hu, S.","Zhao, Y.","Zhang, A.","Abdelzaher, T."],"bibdata":{"bibtype":"article","type":"article","title":"DeepSense: A Unified Deep Learning Framework for Time-Series Mobile Sensing Data Processing","url":"http://dx.doi.org/10.1145/3038912.3052577","doi":"10.1145/3038912.3052577","abstract":"Mobile sensing and computing applications usually require time-series inputs from sensors, such as accelerometers, gyroscopes, and magnetometers. Some applications, such as tracking, can use sensed acceleration and rate of rotation to calculate displacement based on physical system models. Other applications, such as ac-tivity recognition, extract manually designed features from sensor inputs for classification. Such applications face two challenges. On one hand, on-device sensor measurements are noisy. For many mobile applications, it is hard to find a distribution that exactly describes the noise in practice. Unfortunately, calculating target quantities based on physical system and noise models is only as accurate as the noise assumptions. Similarly, in classification ap-plications, although manually designed features have proven to be effective, it is not always straightforward to find the most robust features to accommodate diverse sensor noise patterns and het-erogeneous user behaviors. To this end, we propose DeepSense, a deep learning framework that directly addresses the aforemen-tioned noise and feature customization challenges in a unified man-ner. DeepSense integrates convolutional and recurrent neural net-works to exploit local interactions among similar mobile sensors, merge local interactions of different sensory modalities into global interactions, and extract temporal relationships to model signal dy-namics. DeepSense thus provides a general signal estimation and classification framework that accommodates a wide range of ap-plications. We demonstrate the effectiveness of DeepSense using three representative and challenging tasks: car tracking with mo-tion sensors, heterogeneous human activity recognition, and user identification with biometric motion analysis. DeepSense signifi-cantly outperforms the state-of-the-art methods for all three tasks. In addition, we show that DeepSense is feasible to implement on smartphones and embedded devices thanks to its moderate energy consumption and low latency.","date":"2017","pages":"17–4","keywords":"Activity Recognition,Deep Learning,Internet of Things,Mobile Computing,Mobile Sensing,Tracking,User Identification c 2017","author":[{"propositions":[],"lastnames":["Yao"],"firstnames":["Shuochao"],"suffixes":[]},{"propositions":[],"lastnames":["Hu"],"firstnames":["Shaohan"],"suffixes":[]},{"propositions":[],"lastnames":["Zhao"],"firstnames":["Yiran"],"suffixes":[]},{"propositions":[],"lastnames":["Zhang"],"firstnames":["Aston"],"suffixes":[]},{"propositions":[],"lastnames":["Abdelzaher"],"firstnames":["Tarek"],"suffixes":[]}],"file":"/home/dimitri/Nextcloud/Zotero/storage/YXTTIG45/Yao et al. - 2016 - DeepSense A Unified Deep Learning Framework for Time-Series Mobile Sensing Data Processing.pdf","bibtex":"@article{yaoDeepSenseUnifiedDeep2017,\n title = {{{DeepSense}}: A {{Unified Deep Learning Framework}} for {{Time}}-{{Series Mobile Sensing Data Processing}}},\n url = {http://dx.doi.org/10.1145/3038912.3052577},\n doi = {10.1145/3038912.3052577},\n abstract = {Mobile sensing and computing applications usually require time-series inputs from sensors, such as accelerometers, gyroscopes, and magnetometers. Some applications, such as tracking, can use sensed acceleration and rate of rotation to calculate displacement based on physical system models. Other applications, such as ac-tivity recognition, extract manually designed features from sensor inputs for classification. Such applications face two challenges. On one hand, on-device sensor measurements are noisy. For many mobile applications, it is hard to find a distribution that exactly describes the noise in practice. Unfortunately, calculating target quantities based on physical system and noise models is only as accurate as the noise assumptions. Similarly, in classification ap-plications, although manually designed features have proven to be effective, it is not always straightforward to find the most robust features to accommodate diverse sensor noise patterns and het-erogeneous user behaviors. To this end, we propose DeepSense, a deep learning framework that directly addresses the aforemen-tioned noise and feature customization challenges in a unified man-ner. DeepSense integrates convolutional and recurrent neural net-works to exploit local interactions among similar mobile sensors, merge local interactions of different sensory modalities into global interactions, and extract temporal relationships to model signal dy-namics. DeepSense thus provides a general signal estimation and classification framework that accommodates a wide range of ap-plications. We demonstrate the effectiveness of DeepSense using three representative and challenging tasks: car tracking with mo-tion sensors, heterogeneous human activity recognition, and user identification with biometric motion analysis. DeepSense signifi-cantly outperforms the state-of-the-art methods for all three tasks. In addition, we show that DeepSense is feasible to implement on smartphones and embedded devices thanks to its moderate energy consumption and low latency.},\n date = {2017},\n pages = {17--4},\n keywords = {Activity Recognition,Deep Learning,Internet of Things,Mobile Computing,Mobile Sensing,Tracking,User Identification c 2017},\n author = {Yao, Shuochao and Hu, Shaohan and Zhao, Yiran and Zhang, Aston and Abdelzaher, Tarek},\n file = {/home/dimitri/Nextcloud/Zotero/storage/YXTTIG45/Yao et al. - 2016 - DeepSense A Unified Deep Learning Framework for Time-Series Mobile Sensing Data Processing.pdf}\n}\n\n","author_short":["Yao, S.","Hu, S.","Zhao, Y.","Zhang, A.","Abdelzaher, T."],"key":"yaoDeepSenseUnifiedDeep2017","id":"yaoDeepSenseUnifiedDeep2017","bibbaseid":"yao-hu-zhao-zhang-abdelzaher-deepsenseaunifieddeeplearningframeworkfortimeseriesmobilesensingdataprocessing","role":"author","urls":{"Paper":"http://dx.doi.org/10.1145/3038912.3052577"},"keyword":["Activity Recognition","Deep Learning","Internet of Things","Mobile Computing","Mobile Sensing","Tracking","User Identification c 2017"],"downloads":0},"bibtype":"article","biburl":"https://raw.githubusercontent.com/dlozeve/newblog/master/bib/all.bib","creationDate":"2020-01-08T20:39:38.929Z","downloads":0,"keywords":["activity recognition","deep learning","internet of things","mobile computing","mobile sensing","tracking","user identification c 2017"],"search_terms":["deepsense","unified","deep","learning","framework","time","series","mobile","sensing","data","processing","yao","hu","zhao","zhang","abdelzaher"],"title":"DeepSense: A Unified Deep Learning Framework for Time-Series Mobile Sensing Data Processing","year":null,"dataSources":["3XqdvqRE7zuX4cm8m"]}