What Matters in Data Curation for Multimodal Reasoning? Insights from the DCVLR Challenge
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909992051802112 |
|---|---|
| author | Shin, Yosub Buriek, Michael Sobolev, Boris Bushuyeu, Pavel Kumar, Vikas Xu, Haoyang Watson, Samuel Molybog, Igor |
| author_facet | Shin, Yosub Buriek, Michael Sobolev, Boris Bushuyeu, Pavel Kumar, Vikas Xu, Haoyang Watson, Samuel Molybog, Igor |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_10922 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | What Matters in Data Curation for Multimodal Reasoning? Insights from the DCVLR Challenge Shin, Yosub Buriek, Michael Sobolev, Boris Bushuyeu, Pavel Kumar, Vikas Xu, Haoyang Watson, Samuel Molybog, Igor Artificial Intelligence 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. |
| title | What Matters in Data Curation for Multimodal Reasoning? Insights from the DCVLR Challenge |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2601.10922 |