竞争流匹配扩展质量-覆盖度前沿
原标题：Expanding the Quality-Coverage Frontier with Competitive Flow Matching
AI 摘要
该论文提出 Competitive Flow Matching 方法，通过在训练中引入低维高斯辅助条件码，让模型对同一噪声状态生成 K 个候选速度预测，并仅对最匹配的候选施加损失（赢家通吃），从而缓解流匹配生成器在质量与覆盖度之间的权衡。在 ImageNet-256 上，该方法以不到原始 SiT 预训练 3% 的计算量进行后训练，在 67% Recall 点将 FD_r,6 从 19.22 降至 5.91；在 RAEv2 上 70% Recall 点从 16.83 降至 3.18，显著扩展了质量-覆盖度前沿。
正文节选
Expanding the Quality–Coverage Frontier with Competitive Flow Matching Abstract Flow matching (FM) generators often face a quality-coverage tradeoff. Inference-time techniques such as classifier-free guidance raise quality, but at the cost of coverage. We hypothesize this limitation stems from a basic property of the objective – an intermediate noisy image could correspond to many clean images, yet the model can only predict a single velocity, at best averaging over the multiple possibilities. T