AnE: Pushing the Reasoning Frontier of Multimodal LLMs via Anchor Evolution
2026-05-25 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors identify that current methods to improve reasoning in multimodal large language models hit a limit because they rely on fixed, sometimes low-quality data. To fix this, they propose a new approach called Anchor Evolution, which uses accurate, truth-based data to guide learning and a special training method to help the model internalize reasoning skills more effectively. Their experiments show this method improves the model's reasoning abilities significantly across multiple tasks. Overall, their approach helps the model learn to reason better and more reliably with different types of inputs.
Multimodal Large Language ModelsSupervised Fine-Tuning (SFT)Reinforcement Learning (RL)Truth Anchor ExpansionScaffold-Stripping MechanismReasoning PathData CurationCognitive DriftHallucinated ReasoningTrajectory Rollouts
Authors
Zehao Wang, Yihan Zeng, Zidong Gong, Yuanfan Guo, Feng Zhu, Hongzhi Zhang, Wei Zhang, Wangmeng Zuo
Abstract
Post-training via Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) is crucial for enhancing reasoning in Multimodal Large Language Models (MLLMs), yet existing paradigms often reach a performance bottleneck due to the limitations of static data. While current methods leverage self-reflection or self-evolution to push these boundaries, they still suffer from cognitive drift and hallucinated reasoning paths caused by low-quality synthetic data. To address these challenges, we propose Anchor Evolution (AnE), a new paradigm that integrates truth-anchored data curation and model evolution, achieving faithful and steady performance gains at the reasoning frontier. Specifically, we propose Truth Anchor Expansion, which pinpoints the model failing frontier via trajectory rollouts and leverages ground-truth databases to retrieve high-fidelity anchors for faithful data curation. Subsequently, we introduce the Scaffold-Stripping Mechanism to internalize reasoning capabilities. This mechanism first anchors reasoning paths via scaffold-augmented supervision to mitigate the learning complexity and distribution drift of direct SFT on raw data, then leverages RL to strip the scaffold template, thereby effectively transitioning the reasoning paths into intrinsic model capabilities. Experimental results on multimodal reasoning benchmarks show that our method substantially advances the model performance frontier, improving the base model by 10.3\% across eight multimodal benchmarks and achieving state-of-the-art results. The code will be made publicly available.