通过改进的数据-噪声耦合实现流基生成模型的更快推理
Faster Inference of Flow-Based Generative Models via Improved Data-Noise Coupling
AI总结：
本文提出LOOM-CFM方法，通过优化 minibatch OT 跨 minibatch 的 assignments，提升流基生成模型在样本速度与质量间的平衡，支持高分辨率潜在空间合成和蒸馏初始化。
摘要：
Conditional Flow Matching (CFM)，一种无需模拟的连续归一化流训练方法，为图像和视频生成等关键任务提供了高效的替代方案。CFM在解决这些任务中的性能取决于数据与噪声的耦合方式。最近的方法利用minibatch最优传输（OT）重新分配每个训练步骤中的噪声-数据对，以优化采样轨迹并加速推理。然而，其优化仅限于单个minibatch，限制了其在大规模数据集上的效果。为解决这一不足，我们引入LOOM-CFM（Looking Out Of Minibatch-CFM），一种新的方法，通过在训练过程中跨minibatch保持和优化这些分配，扩展minibatch OT的范围。我们的方法在多个数据集上均显示出在样本速度与质量之间的平衡改进。LOOM-CFM还增强了蒸馏初始化，并支持潜在空间训练中的高分辨率合成。
英文摘要：
Conditional Flow Matching (CFM), a simulation-free method for training continuous normalizing flows, provides an efficient alternative to diffusion models for key tasks like image and video generation. The performance of CFM in solving these tasks depends on the way data is coupled with noise. A recent approach uses minibatch optimal transport (OT) to reassign noise-data pairs in each training step to streamline sampling trajectories and thus accelerate inference. However, its optimization is restricted to individual minibatches, limiting its effectiveness on large datasets. To address this shortcoming, we introduce LOOM-CFM (Looking Out Of Minibatch-CFM), a novel method to extend the scope of minibatch OT by preserving and optimizing these assignments across minibatches over training time. Our approach demonstrates consistent improvements in the sampling speed-quality trade-off across multiple datasets. LOOM-CFM also enhances distillation initialization and supports high-resolution synthesis in latent space training.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
