Learning Interpretable Queries for Explainable Image Classification with Information Pursuit
This addresses the problem of improving explainability in image classification for users by reducing reliance on expert curation, though it is incremental as it builds on the existing Information Pursuit framework.
The paper tackles the limitation of hand-crafted query dictionaries in Information Pursuit for explainable image classification by learning interpretable queries directly from the dataset, resulting in learned dictionaries that significantly outperform those generated with large language models.
Information Pursuit (IP) is an explainable prediction algorithm that greedily selects a sequence of interpretable queries about the data in order of information gain, updating its posterior at each step based on observed query-answer pairs. The standard paradigm uses hand-crafted dictionaries of potential data queries curated by a domain expert or a large language model after a human prompt. However, in practice, hand-crafted dictionaries are limited by the expertise of the curator and the heuristics of prompt engineering. This paper introduces a novel approach: learning a dictionary of interpretable queries directly from the dataset. Our query dictionary learning problem is formulated as an optimization problem by augmenting IP's variational formulation with learnable dictionary parameters. To formulate learnable and interpretable queries, we leverage the latent space of large vision and language models like CLIP. To solve the optimization problem, we propose a new query dictionary learning algorithm inspired by classical sparse dictionary learning. Our experiments demonstrate that learned dictionaries significantly outperform hand-crafted dictionaries generated with large language models.