Truth in the Few: High-Value Data Selection for Efficient Multi-Modal Reasoning

2025-06-05

AI Summary

Multimodal large language models (MLLMs) suffer from high data redundancy and excessive computational costs during training. Method: This paper proposes a cognitive-value-driven data curation paradigm. We introduce Reasoning Activation Potential (RAP) as a novel metric to identify “cognitive samples,” and design a dual-path filtering mechanism comprising a Causal Difference Estimator (CDE) and an Attention Confidence Estimator (ACE), augmented by a Difficulty-aware Replacement Module (DRM). Our approach integrates causal inference via the potential outcomes framework, cross-modal and textual output contrastive analysis, token-level self-attention interpretability modeling, and dynamic difficulty enhancement. Contribution/Results: Using only 9.3% of the original training data, our method surpasses full-data training performance across six mainstream multimodal reasoning benchmarks, reduces computational overhead by over 43%, and significantly improves both training efficiency and generalization capability.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Multi-modal Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

While multi-modal large language models (MLLMs) have made significant progress in complex reasoning tasks via reinforcement learning, it is commonly believed that extensive training data is necessary for improving multi-modal reasoning ability, inevitably leading to data redundancy and substantial computational costs. However, can smaller high-value datasets match or outperform full corpora for multi-modal reasoning in MLLMs? In this work, we challenge this assumption through a key observation: meaningful multi-modal reasoning is triggered by only a sparse subset of training samples, termed cognitive samples, whereas the majority contribute marginally. Building on this insight, we propose a novel data selection paradigm termed Reasoning Activation Potential (RAP), which identifies cognitive samples by estimating each sample's potential to stimulate genuine multi-modal reasoning by two complementary estimators: 1) Causal Discrepancy Estimator (CDE) based on the potential outcome model principle, eliminates samples that overly rely on language priors by comparing outputs between multi-modal and text-only inputs; 2) Attention Confidence Estimator (ACE), which exploits token-level self-attention to discard samples dominated by irrelevant but over-emphasized tokens in intermediate reasoning stages. Moreover, we introduce a Difficulty-aware Replacement Module (DRM) to substitute trivial instances with cognitively challenging ones, thereby ensuring complexity for robust multi-modal reasoning. Experiments on six datasets show that our RAP method consistently achieves superior performance using only 9.3% of the training data, while reducing computational costs by over 43%. Our code is available at https://github.com/Leo-ssl/RAP.

Problem

• Identifying high-value data for efficient multi-modal reasoning
• Reducing computational costs by selecting sparse cognitive samples
• Enhancing reasoning ability with minimal training data

Innovation

• RAP identifies high-value cognitive samples
• Uses CDE and ACE for sample selection
• DRM replaces trivial with challenging samples