Visual Instruction Bottleneck Tuning

Changdae Oh, Jiatong Li, Shawn Im, Sharon Li

Keywords: Multimodal Large Language Models, Vision-Language Models, Instruction Tuning, Distribution Shifts, Robustness, Information Bottleneck

TL;DR: We propose Vittle, a new visual instruction tuning framework that improves robustnessof MLLMs to data distribution shifts by pursuing the minimal sufficient representation.

Abstract: Despite widespread adoption, multimodal large language models (MLLMs) suffer performance degradation when encountering unfamiliar queries under distribution shifts. Existing methods to improve MLLM generalization typically require either more instruction data or larger advanced model architectures, both of which incur non-trivial human labor or computational costs. In this work, we take an alternative approach to enhance the generalization and robustness of MLLMs under distribution shifts, from a representation learning perspective. Inspired by information bottleneck (IB) principle, we derive a variational lower bound of the IB for MLLMs and devise a practical implementation, Visual Instruction Bottleneck Tuning (Vittle). We then provide a theoretical justification of Vittle by revealing its connection to an information-theoretic robustness metric of MLLM. Empirical validation of multiple MLLMs on open-ended and closed-form question answering and object hallucination detection tasks over 45 datasets, including 30 shift scenarios, demonstrates that Vittle consistently improves the MLLM's robustness under shifts by pursuing the learning of a minimal sufficient representation.

Primary Area: Deep learning (e.g., architectures, generative models, optimization for deep networks, foundation models, LLMs)

Submission Number: 6759