Time-series attribution maps with regularized contrastive learning

TL;DR

This paper introduces xCEBRA, a novel contrastive learning-based method for generating identifiable attribution maps in time-series data, improving interpretability of deep learning models with theoretical guarantees and empirical validation.

Contribution

It presents a new regularized contrastive learning algorithm and an attribution method called Inverted Neuron Gradient, providing identifiability guarantees for time-series attribution maps.

Findings

Theoretically guarantees Jacobian matrix identification.

Empirically accurately distinguishes zero and non-zero ground-truth attributions.

Outperforms existing attribution methods on synthetic datasets.

Abstract

Gradient-based attribution methods aim to explain decisions of deep learning models but so far lack identifiability guarantees. Here, we propose a method to generate attribution maps with identifiability guarantees by developing a regularized contrastive learning algorithm trained on time-series data plus a new attribution method called Inverted Neuron Gradient (collectively named xCEBRA). We show theoretically that xCEBRA has favorable properties for identifying the Jacobian matrix of the data generating process. Empirically, we demonstrate robust approximation of zero vs. non-zero entries in the ground-truth attribution map on synthetic datasets, and significant improvements across previous attribution methods based on feature ablation, Shapley values, and other gradient-based methods. Our work constitutes a first example of identifiable inference of time-series attribution maps and opens avenues to a better understanding of time-series data, such as for neural dynamics and decision-processes within neural networks.