Reproducibility study of "LICO: Explainable Models with Language-Image Consistency"

TL;DR

This study critically examines the reproducibility of LICO, a method claiming to enhance interpretability and classification in vision models, finding that results could not be consistently replicated, emphasizing the need for rigorous validation.

Contribution

This paper provides a comprehensive reproducibility assessment of LICO, revealing inconsistencies and highlighting the importance of transparent evaluation in interpretability research.

Findings

Failed to reproduce LICO's claimed performance improvements

Did not observe consistent interpretability enhancements

Undermines reliability of LICO's reported benefits

Abstract

The growing reproducibility crisis in machine learning has brought forward a need for careful examination of research findings. This paper investigates the claims made by Lei et al. (2023) regarding their proposed method, LICO, for enhancing post-hoc interpretability techniques and improving image classification performance. LICO leverages natural language supervision from a vision-language model to enrich feature representations and guide the learning process. We conduct a comprehensive reproducibility study, employing (Wide) ResNets and established interpretability methods like Grad-CAM and RISE. We were mostly unable to reproduce the authors' results. In particular, we did not find that LICO consistently led to improved classification performance or improvements in quantitative and qualitative measures of interpretability. Thus, our findings highlight the importance of rigorous evaluation and transparent reporting in interpretability research.