← Back to archive day

Discovering Failure Modes in Vision-Language Models using RL

Kanishk Jain, Qian Yang, Shravan Nayak, Parisa Kordjamshidi, Nishanth Anand, Aishwarya Agrawal

35

Recommendation Score

significant🟡 IntermediateReasoning & AgentsLLM ReasoningBenchmarkUseful for both

Research context

Primary field

Reasoning & Agents

Reasoning, planning, tool use, and agentic workflows.

Topics

LLM Reasoning

Paper type

Benchmark

Best for

Useful for both

arXiv categories

cs.CVcs.AIcs.CV

Why It Matters

Finding specific weaknesses in vision-language models usually requires slow, manual testing. This paper uses reinforcement learning to automatically discover scenarios where models fail, such as spatial reasoning errors. This automation allows teams to rapidly identify and fix blind spots that human testers might miss.

Abstract

Vision-language Models (VLMs), despite achieving strong performance on multimodal benchmarks, often misinterpret straightforward visual concepts that humans identify effortlessly, such as counting, spatial reasoning, and viewpoint understanding. Previous studies manually identified these weaknesses and found that they often stem from deficits in specific skills. However, such manual efforts are costly, unscalable, and subject to human bias, which often overlooks subtle details in favor of salient objects, resulting in an incomplete understanding of a model's vulnerabilities. To address these limitations, we propose a Reinforcement Learning (RL)-based framework to automatically discover the failure modes or blind spots of any candidate VLM on a given data distribution without human intervention. Our framework trains a questioner agent that adaptively generates queries based on the candidate VLM's responses to elicit incorrect answers. Our approach increases question complexity by focusing on fine-grained visual details and distinct skill compositions as training progresses, consequently identifying 36 novel failure modes in which VLMs struggle. We demonstrate the broad applicability of our framework by showcasing its generalizability across various model combinations.

Published April 6, 2026
© 2026 A2A.pub — AI to Action. From papers to practice, daily.
Summaries are AI-assistedPrivacyTerms