arXiv cs.AI· Yuquan Li, Xinyang Jiang, Yuhao Xue, Cairong Zhao·· 1 天前AI 评分32
ERIM：用显式规则归纳增强多模态上下文学习
Multimodal In-Context Learning with Explicit Rule Induction
AI 导读
研究者提出 ERIM 框架，通过显式归纳演示样本背后的潜在规则来改进多模态上下文学习（MICL）。该方法利用多模态大模型（MLLM）的推理能力构建以规则为中心的数据集，引入规则归纳的显式监督，并采用混合冷启动监督微调与强化学习的两阶段策略微调 MLLM。实验显示，在多种 MICL 设置下性能一致提升，且正确归纳潜在规则可直接带来更强、更稳定的表现。
正文 · 原文
Abstract:A central challenge in multimodal in-context learning (MICL) arises from tasks that are ill-defined without context and require uncovering the underlying rules from demonstrations. Existing methods predominantly treat MICL as direct input-output mapping, overlooking the importance of explicitly inducing and reasoning over such rules, limiting both performance and interpretability. We propose ERIM, Explicit Rule-Induction for Multimodal In-Context Learning, a unified framework that enhances MICL by explicitly inducing the latent rules governing demonstration samples. Leveraging the reasoning ability of multi-modal large language model (MLLM), ERIM constructs rule-centric MICL datasets, introduces explicit supervision for rule induction, and fine-tunes MLLM to perform explicit rule-induction via a two-stage training strategy that combines mixed cold-start supervised fine-tuning with reinforcement learning. Extensive experiments show consistent performance improvements across diverse MICL settings and demonstrate that correctly inducing underlying rules directly leads to stronger and more stable performance.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.41754 [cs.AI] |
| (or arXiv:2610.41754v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.41754 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yuquan Li [view email]
[v1]
Fri, 30 Oct 2026 13:57:30 UTC (2,668 KB)
来源：arXiv cs.AI · arxiv.org