Data Descriptions from Large Language Models with Influence Estimation
Chaeri Kim, Jaeyeon Bae, Taehwan Kim
TL;DR
The paper tackles the interpretability of data in deep learning by turning data descriptions into human-readable language via large language models. It introduces a pipeline that (i) generates class-level textual descriptions using two-stage prompting with external knowledge from Wikipedia, (ii) scores and selects informative proponent texts through a novel Influence Score that combines training-instance influence and CLIP similarity, and (iii) trains a vision model in a shared embedding space with weighted proponent texts to realize cross-modal transfer classification. The approach achieves consistent gains over baselines in zero-shot and cross-modal transfer settings across nine image datasets, and GPT-4o-based qualitative evaluations corroborate the higher helpfulness, informativeness, and relevance of the generated descriptions. Together, these results provide evidence that carefully selected textual descriptions can improve model training and offer intrinsic interpretability of the decision-making process through cross-modal learning.
Abstract
Deep learning models have been successful in many areas but understanding their behaviors still remains a black-box. Most prior explainable AI (XAI) approaches have focused on interpreting and explaining how models make predictions. In contrast, we would like to understand how data can be explained with deep learning model training and propose a novel approach to understand the data via one of the most common media - language - so that humans can easily understand. Our approach proposes a pipeline to generate textual descriptions that can explain the data with large language models by incorporating external knowledge bases. However, generated data descriptions may still include irrelevant information, so we introduce to exploit influence estimation to choose the most informative textual descriptions, along with the CLIP score. Furthermore, based on the phenomenon of cross-modal transferability, we propose a novel benchmark task named cross-modal transfer classification to examine the effectiveness of our textual descriptions. In the experiment of zero-shot setting, we show that our textual descriptions are more effective than other baseline descriptions, and furthermore, we successfully boost the performance of the model trained only on images across all nine image classification datasets. These results are further supported by evaluation using GPT-4o. Through our approach, we may gain insights into the inherent interpretability of the decision-making process of the model.