VL-SAE: Interpreting and Enhancing Vision-Language Alignment with a Unified Concept Set

TL;DR

VL-SAE introduces a unified concept set to interpret and improve vision-language model alignments by correlating neurons with semantic concepts, enhancing interpretability and downstream task performance.

Contribution

The paper proposes VL-SAE, a sparse autoencoder that maps vision-language representations to a concept set, enabling interpretability and alignment enhancement in VLMs.

Findings

VL-SAE effectively interprets vision-language representations.

VL-SAE improves zero-shot image classification accuracy.

VL-SAE reduces hallucinations in VLM outputs.

Abstract

The alignment of vision-language representations endows current Vision-Language Models (VLMs) with strong multi-modal reasoning capabilities. However, the interpretability of the alignment component remains uninvestigated due to the difficulty in mapping the semantics of multi-modal representations into a unified concept set. To address this problem, we propose VL-SAE, a sparse autoencoder that encodes vision-language representations into its hidden activations. Each neuron in its hidden layer correlates to a concept represented by semantically similar images and texts, thereby interpreting these representations with a unified concept set. To establish the neuron-concept correlation, we encourage semantically similar representations to exhibit consistent neuron activations during self-supervised training. First, to measure the semantic similarity of multi-modal representations, we perform their alignment in an explicit form based on cosine similarity. Second, we construct the VL-SAE with a distance-based encoder and two modality-specific decoders to ensure the activation consistency of semantically similar representations. Experiments across multiple VLMs (e.g., CLIP, LLaVA) demonstrate the superior capability of VL-SAE in interpreting and enhancing the vision-language alignment. For interpretation, the alignment between vision and language representations can be understood by comparing their semantics with concepts. For enhancement, the alignment can be strengthened by aligning vision-language representations at the concept level, contributing to performance improvements in downstream tasks, including zero-shot image classification and hallucination elimination. Codes are available at https://github.com/ssfgunner/VL-SAE.