Inductive Moment Matching
This addresses the trade-off between sample quality and inference speed in generative models, offering a stable and efficient alternative to distillation methods.
The paper tackles the problem of slow inference in diffusion and flow matching models by proposing Inductive Moment Matching (IMM), a new generative model for one- or few-step sampling with a single-stage training procedure, achieving a 1.99 FID on ImageNet-256x256 with 8 steps and a state-of-the-art 2-step FID of 1.98 on CIFAR-10.
Diffusion models and Flow Matching generate high-quality samples but are slow at inference, and distilling them into few-step models often leads to instability and extensive tuning. To resolve these trade-offs, we propose Inductive Moment Matching (IMM), a new class of generative models for one- or few-step sampling with a single-stage training procedure. Unlike distillation, IMM does not require pre-training initialization and optimization of two networks; and unlike Consistency Models, IMM guarantees distribution-level convergence and remains stable under various hyperparameters and standard model architectures. IMM surpasses diffusion models on ImageNet-256x256 with 1.99 FID using only 8 inference steps and achieves state-of-the-art 2-step FID of 1.98 on CIFAR-10 for a model trained from scratch.