Remember-R1: Mitigating Long-Context Visual Forgetting through Reinforcement Learning

2026-08-02

AI Summary

This work addresses the tendency of multimodal large language models to overlook visual evidence during long-chain reasoning, a phenomenon termed “long-context visual forgetting.” To mitigate this issue, the authors propose a reinforcement learning–based process-level supervision approach that introduces three novel reward mechanisms operating at the reasoning trajectory level: visual keyword coverage, persistence of visual dependency, and focus on question-relevant image regions. These rewards collectively guide the model to consistently and effectively leverage visual information throughout extended reasoning processes. Evaluated across multiple model scales and multimodal benchmarks, the proposed method significantly enhances reasoning performance and effectively alleviates the decay of visual attention during text generation.

Abstract

Multimodal large language models (MLLMs) increasingly rely on long chain-of-thought reasoning for complex tasks. However, as reasoning sequences lengthen, models may gradually rely less on visual evidence and more on accumulated textual context, leading to visual forgetting. Existing approaches do not directly constrain how visual evidence is used and maintained along the original reasoning trajectory, leaving long-context visual forgetting insufficiently addressed. To address this issue, we propose Remember-R1, a reinforcement learning framework that mitigates long-context visual forgetting by applying process-level supervision directly on the original reasoning trajectory. Specifically, Remember-R1 introduces rewards that encourage broader coverage of matched visual keywords, stronger persistence of visual dependence in later reasoning steps, and greater focus on question-relevant image regions. Experiments across multiple model scales and diverse multimodal benchmarks demonstrate that Remember-R1 consistently improves reasoning performance. Additional analyses further show that it slows the decline of visual attention during generation, supporting its effectiveness in mitigating long-context visual forgetting.

Problem

• visual forgetting
• long-context reasoning
• multimodal large language models
• chain-of-thought

Innovation

• visual forgetting
• reinforcement learning
• multimodal reasoning
• process-level supervision
• long-context