Rethinking Data Selection as Distribution Matching on Semantic Manifolds
Paper · https://arxiv.org/abs/2611.08495
arXiv:2611.08495v1 Announce Type: new Abstract: Instruction tuning of large language models and large vision language models relies on massive training data, which incurs substantial computational and storage costs. Meanwhile, training data often contains redundancy, noise, and low quality samples. Existing data selection methods mostly rely on diversity, importance, or a combination of both, without explicitly constraining the distributional consistency between the selected subset and the full data in the embedding space. We propose QMAC, a quality aware multiband characteristic function matching method that formulates data selection as a joint optimization problem of distribution matching and quality filtering. QMAC matches the empirical characteristic functions of the selected subset and the full data across multiple frequency scales, capturing both global semantic coverage and local perturbations, while using task related quality scores to filter noisy and low quality samples. We systematically evaluate QMAC on 4 data types, 8 target models, and 44 benchmarks. Experimental results show that QMAC outperforms existing data selection baselines on standard image-text instruction data, substantially surpasses random selection on standard text instruction, image-text reasoning, and text reasoning data, and exceeds full data training in multiple model and budget combinations. The code will be made publicly available upon acceptance of the paper.
First seen by Oriel on .
AI summary
The paper presents QMAC, a data selection method that combines semantic distribution matching with quality filtering to reduce instruction-tuning data while preserving or improving model performance.
- QMAC matches empirical characteristic functions of selected and full datasets across multiple frequency scales, capturing global semantic coverage and local perturbations.
- Task-related quality scores filter noisy and low-quality samples from large language model and vision-language model instruction data.
- Evaluation covers 4 data types, 8 target models, and 44 benchmarks.
- QMAC outperforms existing baselines on image-text instruction data and substantially beats random selection on text instruction, image-text reasoning, and text reasoning data.
- Selected subsets exceed full-data training in multiple model and budget combinations; code availability is planned upon paper acceptance.
#data selection #instruction tuning #distribution matching #multimodal models
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| arXiv AI Papers | oai:arXiv.org:2611.08495v1 | https://arxiv.org/abs/2611.08495 Read discussion |