Use Perturbations when Learning from Explanations
This addresses the issue of aligning model predictions with human reasoning in MLX, offering a more effective approach for applications requiring interpretable AI, though it builds incrementally on prior methods.
The paper tackles the problem of sub-optimal performance in machine learning from explanations (MLX) by recasting it as a robustness problem, using human explanations to define a manifold for perturbations, which improves performance and yields state-of-the-art results on benchmarks.
Machine learning from explanations (MLX) is an approach to learning that uses human-provided explanations of relevant or irrelevant features for each input to ensure that model predictions are right for the right reasons. Existing MLX approaches rely on local model interpretation methods and require strong model smoothing to align model and human explanations, leading to sub-optimal performance. We recast MLX as a robustness problem, where human explanations specify a lower dimensional manifold from which perturbations can be drawn, and show both theoretically and empirically how this approach alleviates the need for strong model smoothing. We consider various approaches to achieving robustness, leading to improved performance over prior MLX methods. Finally, we show how to combine robustness with an earlier MLX method, yielding state-of-the-art results on both synthetic and real-world benchmarks.