---
title: "Multimodal In-Context Learning with Explicit Rule Induction"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.41754v1"
permalink: "https://feed.astralor.com/items/itm_258322d496462c5f5e1bffda"
published: "2026-10-30T13:57:30.000Z"
score: 54
channel: "tech"
tags: ["LLM","RAG","多模态","Multimodal In-Context Learning","Explicit Rule Induction","MICL","Existing","ERIM"]
---

# Multimodal In-Context Learning with Explicit Rule Induction

## At a glance

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 M…

## Key points

- 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.

## Full text

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.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_258322d496462c5f5e1bffda