Deciphering Functions of Neurons in Vision-Language Models

TL;DR

This paper investigates the internal neuron functions of vision-language models, revealing specialized neurons and developing tools to interpret and evaluate their roles, enhancing transparency and trustworthiness.

Contribution

It introduces a framework for neuron interpretation in VLMs, classifies neurons into visual, text, and multi-modal types, and proposes an activation simulator for visual neuron explanations.

Findings

Identification of visual, text, and multi-modal neurons

Development of an automated neuron explanation framework

Analysis of neuron behaviors in LLaVA model

Abstract

The burgeoning growth of open-sourced vision-language models (VLMs) has catalyzed a plethora of applications across diverse domains. Ensuring the transparency and interpretability of these models is critical for fostering trustworthy and responsible AI systems. In this study, our objective is to delve into the internals of VLMs to interpret the functions of individual neurons. We observe the activations of neurons with respects to the input visual tokens and text tokens, and reveal some interesting findings. Particularly, we found that there are neurons responsible for only visual or text information, or both, respectively, which we refer to them as visual neurons, text neurons, and multi-modal neurons, respectively. We build a framework that automates the explanation of neurons with the assistant of GPT-4o. Meanwhile, for visual neurons, we propose an activation simulator to assess the reliability of the explanations for visual neurons. System statistical analyses on top of one representative VLM of LLaVA, uncover the behaviors/characteristics of different categories of neurons.