paper-with-me

Papers

What Matters in Data Curation for Multimodal Reasoning? Insights from the DCVLR Challenge

2026-01-16 · Yosub Shin, Michael Buriek, Boris Sobolev, Pavel Bushuyeu, Vikas Kumar, Haoyang Xu, Samuel Watson, Igor Molybog arxiv

We study data curation for multimodal reasoning through the NeurIPS 2025 Data Curation for Vision-Language Reasoning (DCVLR) challenge, which isolates dataset selection by fixing the model and training protocol. Using a compact curated dataset derived primarily from Walton Multimodal Cold Start, our submission placed first in the challenge. Through post-competition ablations, we show that difficulty-based example selection on an aligned base dataset is the dominant driver of performance gains. Increasing dataset size does not reliably improve mean accuracy under the fixed training recipe, but mainly reduces run-to-run variance, while commonly used diversity and synthetic augmentation heuristics provide no additional benefit and often degrade performance. These results characterize DCVLR as a saturation-regime evaluation and highlight the central role of alignment and difficulty in data-efficient multimodal reasoning.

📄 PDF Abstract BibTeX arXiv:2601.10922

Code (0)

등록된 구현이 없습니다.

Tasks

Multimodal Reasoning

Similar Papers 제목 키워드 기반

StoryTR: Narrative-Centric Video Temporal Retrieval with Theory of Mind Reasoning

2026-04-25 · Xuanyue Zhong, Yuqiang Xie, Guanqun Bi, Jiangping Yang 외 arxiv

Current video moment retrieval excels at action-centric tasks but struggles with narrative content. Models can see \textit{what is happening} but fail to reason \textit{why it matters}. This semantic gap stems from the l…

Moment Retrieval

Building Agent Harnesses for Scientific Curation from Multimodal Sources

2026-06-19 · Sheng Zhang, Qin Liu, Renqian Luo, Shufang Xie 외 arxiv

Scientific discovery workflows often depend on structured curation from the literature. This is difficult for current agents because the key evidence is scattered across long text, dense tables, and figures, and the fina…

Dual Tuning for Reasoning Efficacy-Driven Data Curation in Multimodal LLM Training

2026-02-04 · Ruobing Zheng, Tianqi Li, Jianing Li, Qingpei Guo 외 arxiv

Reasoning post-training improves Large Language Models (LLMs) on complex tasks such as mathematics and coding, but its benefits across diverse multimodal tasks remains uncertain. The trend of releasing parallel "Instruct…

Reinforcement Learning

LEEVLA: Seeing What Matters in Latent Environment Evolution for Vision-Language-Action

2026-07-09 · Qi Lyu, Baicheng Liu, Xudong Wang, Jiahua Dong 외 arxiv

Vision-language-action (VLA) models aim to map multimodal inputs to robot actions. However, most existing approaches struggle to cover complex dynamic scenarios due to treating all visual tokens uniformly and reasoning w…

LLAVADI: What Matters For Multimodal Large Language Models Distillation

2024-07-28 · Shilin Xu, Xiangtai Li, Haobo Yuan, Lu Qi 외

The recent surge in Multimodal Large Language Models (MLLMs) has showcased their remarkable potential for achieving generalized intelligence by integrating visual understanding into Large Language Models.Nevertheless, th…

Knowledge Distillation