Causal Probing for Internal Visual Representations in Multimodal Large Language Models

TL;DR

This paper introduces a causal probing framework to analyze how multimodal large language models encode visual concepts, revealing different encoding strategies for entities and abstract concepts and their scaling behaviors.

Contribution

It presents a novel activation steering method to systematically investigate internal visual representations and their relation to model scaling and reasoning capabilities.

Findings

Entities are encoded with localized memorization.

Abstract concepts are globally distributed and require deeper models.

Recognition of geometric relations does not equate to procedural reasoning.

Abstract

Despite the remarkable success of Multimodal Large Language Models (MLLMs) across diverse tasks, the internal mechanisms governing how they encode and ground distinct visual concepts remain poorly understood. To bridge this gap, we propose a causal framework based on activation steering to actively probe and manipulate internal visual representations. Through systematic intervention across four visual concept categories, our results reveal a divergence in concept encoding: entities exhibit distinct localized memorization, whereas abstract concepts are globally distributed across the network. Critically, this divergence uncovers a mechanistic driver of scaling laws: increasing model depth is indispensable for encoding distributed and complex abstract concepts, whereas entity localization remains remarkably invariant to scale. Furthermore, reverse steering uncovers that blocking explicit output triggers a surge in latent activations, exposing a compensatory mechanism between perception and generation. Finally, extending our analysis to visual reasoning, we expose a disconnect between perception and reasoning although MLLMs successfully recognize geometric relations, they treat them merely as static visual features, failing to trigger the procedural execution necessary for abstract problem-solving.