Time-series attribution maps with regularized contrastive learning
This work addresses the problem of unreliable attribution in time-series data for researchers in fields like neural dynamics and AI interpretability, offering a first example of identifiable inference, though it is incremental as it builds on existing gradient-based methods.
The paper tackles the lack of identifiability guarantees in gradient-based attribution methods for deep learning models by proposing xCEBRA, a method combining regularized contrastive learning and Inverted Neuron Gradient to generate attribution maps with theoretical guarantees. Empirically, it shows robust approximation of ground-truth attribution maps on synthetic data and significant improvements over previous methods like feature ablation and Shapley values.
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.