prikaz prve stranice dokumenta Benchmarking Attention-Based Interpretability of Deep Learning in Multivariate Time Series Predictions
  Download
PDF 712.29 KB
Scientific paper - Original scientific paper
Benchmarking Attention-Based Interpretability of Deep Learning in Multivariate Time Series Predictions
Entropy, 23 (2021), 2; 143. https://doi.org/10.3390/e23020143
Barić, Domjan; Fumić, Petar; Horvatić, Davor; Lipić, Tomislav

Cite this document

Barić, D., Fumić, P., Horvatić, D. & Lipić, T. (2021). Benchmarking Attention-Based Interpretability of Deep Learning in Multivariate Time Series Predictions. Entropy, 23. (2). doi: 10.3390/e23020143

Barić, Domjan, et al. "Benchmarking Attention-Based Interpretability of Deep Learning in Multivariate Time Series Predictions." Entropy, vol. 23, no. 2, 2021. https://doi.org/10.3390/e23020143

Barić, Domjan, Petar Fumić, Davor Horvatić and Tomislav Lipić. "Benchmarking Attention-Based Interpretability of Deep Learning in Multivariate Time Series Predictions." Entropy 23, no. 2 (2021). https://doi.org/10.3390/e23020143

Barić, D., et al. (2021) 'Benchmarking Attention-Based Interpretability of Deep Learning in Multivariate Time Series Predictions', Entropy, 23(2). doi: 10.3390/e23020143

Barić D, Fumić P, Horvatić D, Lipić T. Benchmarking Attention-Based Interpretability of Deep Learning in Multivariate Time Series Predictions. Entropy [Internet]. 2021 January 25 [cited 2024 October 05];23(2). doi: 10.3390/e23020143

D. Barić, P. Fumić, D. Horvatić and T. Lipić, "Benchmarking Attention-Based Interpretability of Deep Learning in Multivariate Time Series Predictions", Entropy, vol. 23, no. 2, January 2021. [Online]. Available at: https://urn.nsk.hr/urn:nbn:hr:217:063684. [Accessed: 05 October 2024]

Please login to the repository to save this object to your list.
Metadata
Title (english)Benchmarking Attention-Based Interpretability of Deep Learning in Multivariate Time Series Predictions
AuthorDomjan Barić
AuthorPetar Fumić
AuthorDavor Horvatić
AuthorTomislav Lipić
Author's institutionUniversity of Zagreb
Faculty of Science
(Department of Physics)
Scientific / art field,
discipline and subdiscipline		NATURAL SCIENCES
Physics
Abstract (english)
The adaptation of deep learning models within safety-critical systems cannot rely only on good prediction performance but needs to provide interpretable and robust explanations for their decisions. When modeling complex sequences, attention mechanisms are regarded as the established approach to support deep neural networks with intrinsic interpretability. This paper focuses on the emerging trend of specifically designing diagnostic datasets for understanding the inner workings of attention mechanism based deep learning models for multivariate forecasting tasks. We design a novel benchmark of synthetically designed datasets with the transparent underlying generating process of multiple time series interactions with increasing complexity. The benchmark enables empirical evaluation of the performance of attention based deep neural networks in three different aspects: (i) prediction performance score, (ii) interpretability correctness, (iii) sensitivity analysis. Our analysis shows that although most models have satisfying and stable prediction performance results, they often fail to give correct interpretability. The only model with both a satisfying performance score and correct interpretability is IMV-LSTM, capturing both autocorrelations and crosscorrelations between multiple time series. Interestingly, while evaluating IMV-LSTM on simulated data from statistical and mechanistic models, the correctness of interpretability increases with more complex datasets.
Keywords (english)
Languageenglish
Publication typeScientific paper - Original scientific paper
Publication statusPublished
Peer reviewPeer review - international
Publication versionPublished version
Journal titleEntropy
Numberingvol. 23, no. 2, 143
e-ISSN1099-4300
DOIhttps://doi.org/10.3390/e23020143
URN:NBNurn:nbn:hr:217:063684
Publication2021-01-25
ProjectNumber: KK.01.1.1.01.0004
Title (croatian): Provedba vrhunskih istraživanja u sklopu Znanstvenog centra izvrsnosti za kvantne i kompleksne sustave te reprezentacije Liejevih algebri
Title (english): Implementation of top research within the Scientific Center of Excellence for Quantum and Complex Systems, and Representations of Lie Algebras
Acronym: QuantiXLie
Leader: Hrvoje Buljan
Jurisdiction: eu
Funding stream: European Regional Development Fund
Document URLhttps://www.mdpi.com/1099-4300/23/2/143
Type of resourceText
Access conditionsOpen access
Terms of useLicence Attribution 4.0 International (CC BY 4.0) icon
Public note© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). Received: 19 December 2020. Accepted: 21 January 2021. Published: 25 January 2021.
Created on2024-09-22 12:41:01