@inproceedings{anonymous_improved_2023,
	title = {Improved {Test} {Input} {Prioritization} {Using} {Verification} {Monitors} with {False} {Prediction} {Cluster} {Centroids}},
	abstract = {Deep learning (DL) systems have been remarkably suc- cessful in various industries, but they could have critical misbehav- iors. To identify the weakness of a trained model and repair it through new data collection(s), one needs to figure out the corner cases of a trained model. Constructing new dataset(s) to re-trained a DL model requires extra budget and time. To identity corner cases more effec- tively, we propose a new method that uses a similarity for False Pre- diction Clusters (FPCs) in training data to prioritize error-revealing instances among unlabeled test datasets. We refer to the proposed method as DeepFPC. Different from existing methods that focus on classification tasks, we extend DeepFPC to be applicable to regres- sion tasks. Our numerical experiments show that proposed DeepFPC can achieve 1) higher detection rate of faults in classification task, 2) higher correlation to erroneous behavior in regression task, com- pared to the state-of-the-art method. Additional numerical experi- ments demonstrate that in active learning, proposed DeepFPC sig- nificantly improves test input prioritization over existing methods.},
	author = {Anonymous},
	year = {2023},
}

{"_id":"5NeDCJXLpxfHmksxe","bibbaseid":"anonymous-improvedtestinputprioritizationusingverificationmonitorswithfalsepredictionclustercentroids-2023","author_short":["Anonymous"],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","title":"Improved Test Input Prioritization Using Verification Monitors with False Prediction Cluster Centroids","abstract":"Deep learning (DL) systems have been remarkably suc- cessful in various industries, but they could have critical misbehav- iors. To identify the weakness of a trained model and repair it through new data collection(s), one needs to figure out the corner cases of a trained model. Constructing new dataset(s) to re-trained a DL model requires extra budget and time. To identity corner cases more effec- tively, we propose a new method that uses a similarity for False Pre- diction Clusters (FPCs) in training data to prioritize error-revealing instances among unlabeled test datasets. We refer to the proposed method as DeepFPC. Different from existing methods that focus on classification tasks, we extend DeepFPC to be applicable to regres- sion tasks. Our numerical experiments show that proposed DeepFPC can achieve 1) higher detection rate of faults in classification task, 2) higher correlation to erroneous behavior in regression task, com- pared to the state-of-the-art method. Additional numerical experi- ments demonstrate that in active learning, proposed DeepFPC sig- nificantly improves test input prioritization over existing methods.","author":[{"firstnames":[],"propositions":[],"lastnames":["Anonymous"],"suffixes":[]}],"year":"2023","bibtex":"@inproceedings{anonymous_improved_2023,\n\ttitle = {Improved {Test} {Input} {Prioritization} {Using} {Verification} {Monitors} with {False} {Prediction} {Cluster} {Centroids}},\n\tabstract = {Deep learning (DL) systems have been remarkably suc- cessful in various industries, but they could have critical misbehav- iors. To identify the weakness of a trained model and repair it through new data collection(s), one needs to figure out the corner cases of a trained model. Constructing new dataset(s) to re-trained a DL model requires extra budget and time. To identity corner cases more effec- tively, we propose a new method that uses a similarity for False Pre- diction Clusters (FPCs) in training data to prioritize error-revealing instances among unlabeled test datasets. We refer to the proposed method as DeepFPC. Different from existing methods that focus on classification tasks, we extend DeepFPC to be applicable to regres- sion tasks. Our numerical experiments show that proposed DeepFPC can achieve 1) higher detection rate of faults in classification task, 2) higher correlation to erroneous behavior in regression task, com- pared to the state-of-the-art method. Additional numerical experi- ments demonstrate that in active learning, proposed DeepFPC sig- nificantly improves test input prioritization over existing methods.},\n\tauthor = {Anonymous},\n\tyear = {2023},\n}\n\n\n\n\n\n\n\n","author_short":["Anonymous"],"key":"anonymous_improved_2023","id":"anonymous_improved_2023","bibbaseid":"anonymous-improvedtestinputprioritizationusingverificationmonitorswithfalsepredictionclustercentroids-2023","role":"author","urls":{},"metadata":{"authorlinks":{}},"downloads":0,"html":""},"bibtype":"inproceedings","biburl":"https://bibbase.org/zotero/matteocarde","dataSources":["BLJPWpPi3hneTGaPF"],"keywords":[],"search_terms":["improved","test","input","prioritization","using","verification","monitors","false","prediction","cluster","centroids","anonymous"],"title":"Improved Test Input Prioritization Using Verification Monitors with False Prediction Cluster Centroids","year":2023}