{"ID":2843747,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2511.06722","arxiv_id":"2511.06722","title":"Revisiting the Data Sampling in Multimodal Post-training from a Difficulty-Distinguish View","abstract":"Recent advances in Multimodal Large Language Models (MLLMs) have spurred significant progress in Chain-of-Thought (CoT) reasoning. Building on the success of Deepseek-R1, researchers extended multimodal reasoning to post-training paradigms based on reinforcement learning (RL), focusing predominantly on mathematical datasets. However, existing post-training paradigms tend to neglect two critical aspects: (1) The lack of quantifiable difficulty metrics capable of strategically screening samples for post-training optimization. (2) Suboptimal post-training paradigms that fail to jointly optimize perception and reasoning capabilities. To address this gap, we propose two novel difficulty-aware sampling strategies: Progressive Image Semantic Masking (PISM) quantifies sample hardness through systematic image degradation, while Cross-Modality Attention Balance (CMAB) assesses cross-modal interaction complexity via attention distribution analysis. Leveraging these metrics, we design a hierarchical training framework that incorporates both GRPO-only and SFT+GRPO hybrid training paradigms, and evaluate them across six benchmark datasets. Experiments demonstrate consistent superiority of GRPO applied to difficulty-stratified samples compared to conventional SFT+GRPO pipelines, indicating that strategic data sampling can obviate the need for supervised fine-tuning while improving model accuracy. Our code will be released at https://github.com/qijianyu277/DifficultySampling.","short_abstract":"Recent advances in Multimodal Large Language Models (MLLMs) have spurred significant progress in Chain-of-Thought (CoT) reasoning. Building on the success of Deepseek-R1, researchers extended multimodal reasoning to post-training paradigms based on reinforcement learning (RL), focusing predominantly on mathematical dat...","url_abs":"https://arxiv.org/abs/2511.06722","url_pdf":"https://arxiv.org/pdf/2511.06722v1","authors":"[\"Jianyu Qi\",\"Ding Zou\",\"Wenrui Yan\",\"Rui Ma\",\"Jiaxu Li\",\"Zhijie Zheng\",\"Zhiguo Yang\",\"Rongchang Zhao\"]","published":"2025-11-10T05:31:59Z","proceeding":"cs.CV","tasks":"[\"cs.CV\",\"cs.AI\",\"cs.CL\"]","methods":"[\"Reinforcement Learning\",\"Large Language Model\",\"Language Model\"]","has_code":false,"code_links":[{"ID":607246,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_id":2843747,"paper_url":"https://arxiv.org/abs/2511.06722","paper_title":"Revisiting the Data Sampling in Multimodal Post-training from a Difficulty-Distinguish View","repo_url":"https://github.com/qijianyu277/DifficultySampling","is_official":false,"mentioned_in_paper":false,"mentioned_in_github":true,"github_stars":0}]}