TSInsight: A Local-Global Attribution Framework for Interpretability in Time Series Data

With the rise in the employment of deep learning methods in safety-critical scenarios, interpretability is more essential than ever before. Although many different directions regarding interpretability have been explored for visual modalities, time series data has been neglected, with only a handful...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Shoaib Ahmed Siddiqui, Dominique Mercier, Andreas Dengel, Sheraz Ahmed
Formato:	article
Lenguaje:	EN
Publicado:	MDPI AG 2021
Materias:	interpretability time series analysis feature attribution deep learning auto-encoder feature importance Chemical technology TP1-1185
Acceso en línea:	https://doaj.org/article/881a494110f14e32aac9ec0578d41da6
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:881a494110f14e32aac9ec0578d41da6
record_format	dspace
spelling	oai:doaj.org-article:881a494110f14e32aac9ec0578d41da62021-11-11T19:18:33ZTSInsight: A Local-Global Attribution Framework for Interpretability in Time Series Data10.3390/s212173731424-8220https://doaj.org/article/881a494110f14e32aac9ec0578d41da62021-11-01T00:00:00Zhttps://www.mdpi.com/1424-8220/21/21/7373https://doaj.org/toc/1424-8220With the rise in the employment of deep learning methods in safety-critical scenarios, interpretability is more essential than ever before. Although many different directions regarding interpretability have been explored for visual modalities, time series data has been neglected, with only a handful of methods tested due to their poor intelligibility. We approach the problem of interpretability in a novel way by proposing TSInsight, where we attach an auto-encoder to the classifier with a sparsity-inducing norm on its output and fine-tune it based on the gradients from the classifier and a reconstruction penalty. TSInsight learns to preserve features that are important for prediction by the classifier and suppresses those that are irrelevant, i.e., serves as a feature attribution method to boost the interpretability. In contrast to most other attribution frameworks, TSInsight is capable of generating both instance-based and model-based explanations. We evaluated TSInsight along with nine other commonly used attribution methods on eight different time series datasets to validate its efficacy. The evaluation results show that TSInsight naturally achieves output space contraction; therefore, it is an effective tool for the interpretability of deep time series models.Shoaib Ahmed SiddiquiDominique MercierAndreas DengelSheraz AhmedMDPI AGarticleinterpretabilitytime series analysisfeature attributiondeep learningauto-encoderfeature importanceChemical technologyTP1-1185ENSensors, Vol 21, Iss 7373, p 7373 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	interpretability time series analysis feature attribution deep learning auto-encoder feature importance Chemical technology TP1-1185
spellingShingle	interpretability time series analysis feature attribution deep learning auto-encoder feature importance Chemical technology TP1-1185 Shoaib Ahmed Siddiqui Dominique Mercier Andreas Dengel Sheraz Ahmed TSInsight: A Local-Global Attribution Framework for Interpretability in Time Series Data
description	With the rise in the employment of deep learning methods in safety-critical scenarios, interpretability is more essential than ever before. Although many different directions regarding interpretability have been explored for visual modalities, time series data has been neglected, with only a handful of methods tested due to their poor intelligibility. We approach the problem of interpretability in a novel way by proposing TSInsight, where we attach an auto-encoder to the classifier with a sparsity-inducing norm on its output and fine-tune it based on the gradients from the classifier and a reconstruction penalty. TSInsight learns to preserve features that are important for prediction by the classifier and suppresses those that are irrelevant, i.e., serves as a feature attribution method to boost the interpretability. In contrast to most other attribution frameworks, TSInsight is capable of generating both instance-based and model-based explanations. We evaluated TSInsight along with nine other commonly used attribution methods on eight different time series datasets to validate its efficacy. The evaluation results show that TSInsight naturally achieves output space contraction; therefore, it is an effective tool for the interpretability of deep time series models.
format	article
author	Shoaib Ahmed Siddiqui Dominique Mercier Andreas Dengel Sheraz Ahmed
author_facet	Shoaib Ahmed Siddiqui Dominique Mercier Andreas Dengel Sheraz Ahmed
author_sort	Shoaib Ahmed Siddiqui
title	TSInsight: A Local-Global Attribution Framework for Interpretability in Time Series Data
title_short	TSInsight: A Local-Global Attribution Framework for Interpretability in Time Series Data
title_full	TSInsight: A Local-Global Attribution Framework for Interpretability in Time Series Data
title_fullStr	TSInsight: A Local-Global Attribution Framework for Interpretability in Time Series Data
title_full_unstemmed	TSInsight: A Local-Global Attribution Framework for Interpretability in Time Series Data
title_sort	tsinsight: a local-global attribution framework for interpretability in time series data
publisher	MDPI AG
publishDate	2021
url	https://doaj.org/article/881a494110f14e32aac9ec0578d41da6
work_keys_str_mv	AT shoaibahmedsiddiqui tsinsightalocalglobalattributionframeworkforinterpretabilityintimeseriesdata AT dominiquemercier tsinsightalocalglobalattributionframeworkforinterpretabilityintimeseriesdata AT andreasdengel tsinsightalocalglobalattributionframeworkforinterpretabilityintimeseriesdata AT sherazahmed tsinsightalocalglobalattributionframeworkforinterpretabilityintimeseriesdata
_version_	1718431595701993472

TSInsight: A Local-Global Attribution Framework for Interpretability in Time Series Data

Ejemplares similares