🤖 AI Summary
To address the limitations of expert- or large-language-model (LLM)-based manual construction of query dictionaries in interpretable image classification, this paper proposes a data-driven, learnable query dictionary framework. Methodologically, it introduces the first formulation of the query dictionary as an optimized parameter embedded within the latent space of multimodal models (e.g., CLIP), trained end-to-end and deployed for posterior inference under an Information Pursuit framework. The approach integrates variational optimization, latent-space mapping, and a sparsity-inspired query learning algorithm grounded in dictionary learning principles. Evaluated on multiple benchmark datasets, the learned query dictionary consistently outperforms LLM-generated handcrafted queries—achieving higher classification accuracy and improved human interpretability—as validated by both quantitative metrics and human evaluation. This work overcomes fundamental bottlenecks inherent in conventional prompt engineering and human-defined priors, establishing a new paradigm for scalable, adaptive, and interpretable vision-language alignment.
📝 Abstract
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.
