On Understanding of Spatiotemporal Prediction Model
- 28 December 2022
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 33 (7) , 3087-3103
Abstract
Recently, explainable artificial intelligence has received considerable attention. Most existing studies are focusing on the tasks of CNNs-based image classification and RNNs-based time series analysis. In this paper, we pay attention to the more complicated spatiotemporal predictive learning task (SPLT), where both the spatial and temporal information play important roles. To explain the internal mechanism of spatiotemporal prediction models, we propose a comprehensive analysis method. Specifically, with a typical encoder-decoder framework, we focus on two core issues of SPLT: image generation and spatiotemporal dynamics. For the first issue, we develop a quantitative channel perturbation method to explore the importance of features to prediction. Furthermore, we propose a technique called the synthesis of multiple independent components to analyze how these features generate the prediction. According to the experimental results, the coarse- and fine-grained synthesis (CFGS) mechanism is drawn for image generation in SPLT. For the second issue, we propose a state decomposition technique and a state expansion technique to disentangle coupled signals in the spatiotemporal dynamical system. This helps us to explore the mechanism of forming motion. Moreover, to diagnose the movement of a particular region during analysis, we propose a fluorescent stamp-based technique. By observing extensive experimental results, we summarize a collaboration mechanism to explain how the motion is formed in SPLT, namely, the extending the present and erasing the past (EPEP) mechanism. To the best of our knowledge, this is the first work to interpret the internal mechanism of SPLT models.Keywords
Funding Information
- National Natural Science Foundation of China (62272130, 61972111)
- Shenzhen Science and Technology Program (KCXFZ20211020163403005, JCYJ20210324120208022, JCYJ20200109113014456)
- European Space Agency (ESA) and National Remote Sensing Center of China (NRSCC) Dragon 5 Cooperation Program (ID 59333)
This publication has 48 references indexed in Scilit:
- Interpreting Deep Visual Representations via Network DissectionIEEE Transactions on Pattern Analysis and Machine Intelligence, 2018
- Exact and Consistent Interpretation for Piecewise Linear Neural Networks2018
- Representation of Linguistic Form and Function in Recurrent Neural NetworksComputational Linguistics, 2017
- Interpretable Explanations of Black Boxes by Meaningful Perturbation2017
- Learning Deep Features for Discriminative Localization2016
- “Why Should I Trust You?”: Explaining the Predictions of Any Classifier2016
- Visualizing and Understanding Convolutional Networks2014
- Scale-Selective Verification of Rainfall Accumulations from High-Resolution Forecasts of Convective EventsMonthly Weather Review, 2008
- Actions as Space-Time ShapesIEEE Transactions on Pattern Analysis and Machine Intelligence, 2007
- Recognizing human actions: a local SVM approach2004