Zero-Shot Textual Explanations via Translating Decision-Critical Features

TL;DR

TEXTER is a novel method that isolates decision-critical features in image classifiers and translates them into natural language explanations, improving faithfulness and interpretability.

Contribution

It introduces a technique to identify and emphasize decision-critical features and maps them into CLIP space for better zero-shot textual explanations.

Findings

TEXTER produces more faithful explanations than existing methods.

The approach enhances interpretability for Transformer-based models.

Code is publicly available at the provided GitHub URL.

Abstract

Textual explanations make image classifier decisions transparent by describing the prediction rationale in natural language. Large vision-language models can generate captions but are designed for general visual understanding, not classifier-specific reasoning. Existing zero-shot explanation methods align global image features with language, producing descriptions of what is visible rather than what drives the prediction. We propose TEXTER, which overcomes this limitation by isolating decision-critical features before alignment. TEXTER identifies the neurons contributing to the prediction and emphasizes the features encoded in those neurons -- i.e., the decision-critical features. It then maps these emphasized features into the CLIP feature space to retrieve textual explanations that reflect the model's reasoning. A sparse autoencoder further improves interpretability, particularly for Transformer architectures. Extensive experiments show that TEXTER provides more faithful and interpretable explanations than existing methods. The code is available at \url{https://github.com/tttt-0814/TEXTER}.