Debunking Grad-ECLIP: A Comprehensive Study on Its Incorrectness and Fundamental Principles for Model Interpretation

TL;DR

This paper critically examines Grad-ECLIP, revealing it is not a novel approach and highlighting its flaws, while proposing fundamental principles for accurate model interpretation.

Contribution

It demonstrates Grad-ECLIP's equivalence to attention-based methods and emphasizes essential principles for reliable model interpretation.

Findings

Grad-ECLIP is equivalent to a simpler attention-based method.

Grad-ECLIP's interpretation results are misaligned with the original model.

Fundamental principles for correct model interpretation are proposed.

Abstract

Grad-ECLIP is published at ICML 2024 and represents a new Transformer interpretation technical route (intermediate features-based). First, this paper demonstrates that the intermediate features-based technical route is not a novel one. Based on the existing attention-based route, we have developed Attention-ECLIP, which is completely equivalent to Grad-ECLIP but with simpler computation. Both through formal derivation and experimental validation, we prove that the intermediate feature-based route represented by Grad-ECLIP is actually an equivalent variant of the attention-based route. Next, this paper demonstrates that the Grad-ECLIP method is flawed. The model interpretation results obtained by Grad-ECLIP are not those of the original model, and the interpretation results are misaligned with the model's performance. We analyze the causes of Grad-ECLIP's flaws and propose, or rather, explicitly emphasize two fundamental principles that model interpretation should adhere to in order to avoid similar errors.