A Distributional View for Visual Mechanistic Interpretability: KL-Minimal Soft-Constraint Principle
For researchers in visual mechanistic interpretability, this work provides a principled framework to address the trade-off between human interpretability and model faithfulness.
The paper identifies statistical biases in existing visual mechanistic interpretability methods that either produce perceptually uninterpretable or mechanistically unfaithful results. It proposes a KL-minimal soft-constraint principle realized via energy-guided diffusion posterior sampling, achieving a theoretical balance between interpretability and faithfulness, validated on DINOv3.
Most current paradigms in visual mechanistic interpretability (MI) remain confined to interpreting internal units of the vision model via heuristic methods (e.g., top-$K$ activation retrieval or optimization with regularization). In this work, we establish a theoretical distributional view for visual MI, which models the influence of a feature activation on the natural image distribution, thereby formulating a Kullback-Leibler (KL)-minimal optimization problem to model the MI task. Under this framework, statistical biases are identified within previous MI paradigms, which reveal that they may either be perceptually uninterpretable to humans (i.e., deviate from the natural image distribution), or mechanistically unfaithful to the vision models (i.e., unable to activate model features). To resolve the biases under the distributional view, we propose a model with a KL-minimal soft-constraint principle for visual MI that theoretically balances interpretability and faithfulness. We realize this principle via energy-guided diffusion posterior sampling. Extensive experiments validate the theoretical soundness of the proposed distributional view and demonstrate the practical effectiveness of our paradigm on the DINOv3 vision model.