跳转至

文章背景与核心概要

合成数据在机器学习的训练和数据增强中得到了越来越广泛的应用。然而，当前的许多策略通常依赖于外部的基础模型或第三方数据集，由于政策或法律限制，这些资源在许多场景下的使用受到限制。为此，本文提出了 ScoreMix——一种独立的合成数据生成方法，它利用扩散模型的得分可加性（score compositionality），直接在现有数据集中沿着反向扩散轨迹混合类条件得分，从而为识别任务生成困难的合成样本。

研究发现，选择在判别器嵌入空间中距离较远的类别进行混合，相比于基于邻近性的选择，能够带来显著更大的性能提升（平均额外提升高达 3%）。此外，标准的对齐度量表明条件空间与嵌入空间在很大程度上是不相关的，并且生成器的条件空间对下游性能的影响微乎其微。在 8 个公开的人脸识别基准测试中，ScoreMix 在无需进行超参数搜索的情况下，将准确率提升了最多 7 个百分点，充分凸显了其强大的鲁棒性和实用价值。该研究成果已发表于第 43 届国际机器学习会议（ICML 2026）。

ScoreMix: Synthetic Data Generation by Score Composition in Diffusion Models Improves Recognition

Summary

ScoreMix is a self-contained synthetic data generation method designed to improve recognition tasks (such as face recognition) without relying on external foundation models or third-party datasets. By leveraging the score compositionality of diffusion models, ScoreMix mixes class-conditioned scores along reverse diffusion trajectories to produce hard synthetic samples directly from the available dataset.

Key findings and results include: * Class Selection Strategy: Mixing classes that are distant in the discriminator's embedding space yields significantly larger performance gains (up to 3% additional average improvement) compared to proximity-based selection. * Space Decorrelation: Condition and embedding spaces are largely uncorrelated under standard alignment metrics, and the generator's condition space has a negligible effect on downstream performance. * Robust Performance: Across 8 public face recognition benchmarks, ScoreMix improves accuracy by up to 7 percentage points without requiring hyperparameter searches. * Conference Publication: Published in the Proceedings of the 43rd International Conference on Machine Learning (ICML 2026).

Paper Metadata / 论文元数据

属性 (Attribute) 详情 (Details)
arXiv ID arXiv:2506.10226 [cs.CV]
作者 (Authors) Parsa Rahimi, Sebastien Marcel
主学科 (Primary Subject) 计算机视觉与模式识别 (cs.CV)
其他学科 (Other Subjects) 人工智能 (cs.AI), 机器学习 (cs.LG)
会议 (Conference) ICML 2026 (第 43 届国际机器学习会议论文集)
提交日期 (Submitted Date) 2025年6月11日（最后修订：2026年9月3日，版本 v3）
论文网站 (Paper Website) 官方项目主页

Abstract / 摘要

Synthetic data generation is increasingly used in machine learning for training and data augmentation. Yet, current strategies often rely on external foundation models or datasets, whose usage is restricted in many scenarios due to policy or legal constraints. We propose ScoreMix, a self-contained synthetic generation method to produce hard synthetic samples for recognition tasks by leveraging the score compositionality of diffusion models. The approach mixes class-conditioned scores along reverse diffusion trajectories, yielding domain-specific data augmentation without external resources. We systematically study class-selection strategies and find that mixing classes distant in the discriminator's embedding space yields larger gains, providing up to 3% additional average improvement, compared to selection based on proximity. Interestingly, we observe that condition and embedding spaces are largely uncorrelated under standard alignment metrics, and the generator's condition space has a negligible effect on downstream performance. Across 8 public face recognition benchmarks, ScoreMix improves accuracy by up to 7 percentage points, without hyperparameter search, highlighting both robustness and practicality. Our method provides a simple yet effective way to maximize discriminator performance using only the available dataset, without reliance on third-party resources.