On the Out-Of-Distribution Generalization of Multimodal Large Language Models

TL;DR

This paper evaluates the out-of-distribution generalization of Multimodal Large Language Models, identifying mapping deficiency as a key challenge and demonstrating that in-context learning can improve their robustness, though vulnerabilities remain under domain shifts.

Contribution

It provides a comprehensive analysis of MLLMs' generalization limits, identifies the primary cause as mapping deficiency, and shows that in-context learning can enhance their performance in OOD scenarios.

Findings

MLLMs struggle with out-of-distribution generalization.

Mapping deficiency is identified as the main challenge.

In-context learning improves generalization but remains vulnerable to domain shifts.

Abstract

We investigate the generalization boundaries of current Multimodal Large Language Models (MLLMs) via comprehensive evaluation under out-of-distribution scenarios and domain-specific tasks. We evaluate their zero-shot generalization across synthetic images, real-world distributional shifts, and specialized datasets like medical and molecular imagery. Empirical results indicate that MLLMs struggle with generalization beyond common training domains, limiting their direct application without adaptation. To understand the cause of unreliable performance, we analyze three hypotheses: semantic misinterpretation, visual feature extraction insufficiency, and mapping deficiency. Results identify mapping deficiency as the primary hurdle. To address this problem, we show that in-context learning (ICL) can significantly enhance MLLMs' generalization, opening new avenues for overcoming generalization barriers. We further explore the robustness of ICL under distribution shifts and show its vulnerability to domain shifts, label shifts, and spurious correlation shifts between in-context examples and test data.