Interpretable Neural Networks with Frank-Wolfe: Sparse Relevance Maps and Relevance Orderings

TL;DR

This paper introduces a constrained optimization approach using Frank-Wolfe algorithms to produce sparse and ordered relevance maps for neural network interpretability, outperforming existing methods.

Contribution

It reformulates RDE as a constrained optimization problem, enabling multi-rate and relevance-ordering variants that improve interpretability and performance.

Findings

Reformulated RDE with Frank-Wolfe for sparsity control

Proposed multi-rate and relevance-ordering RDE variants

Empirically outperforms standard RDE and baselines

Abstract

We study the effects of constrained optimization formulations and Frank-Wolfe algorithms for obtaining interpretable neural network predictions. Reformulating the Rate-Distortion Explanations (RDE) method for relevance attribution as a constrained optimization problem provides precise control over the sparsity of relevance maps. This enables a novel multi-rate as well as a relevance-ordering variant of RDE that both empirically outperform standard RDE and other baseline methods in a well-established comparison test. We showcase several deterministic and stochastic variants of the Frank-Wolfe algorithm and their effectiveness for RDE.