DMAD recasts distribution matching as classification and learns log-density ratios directly to improve few-step student training.
DMAD improves upon DMD by learning log-density ratios directly, reducing the need for an auxiliary diffusion model.
Before reading this…
Applications
To understand this paper, make sure you know these concepts first:
Distribution Matching Distillation (DMD) trains a few-step student from the difference between separately estimated target and student scores, so it must keep an auxiliary diffusion model fitted to the student's evolving distribution at extra memory and computation cost. We introduce DMAD, Distribution Matching as Adversarial Distillation, which recasts distribution matching as classification and learns the required log-density ratios directly. Two discriminator heads on a shared backbone distinguish real data and teacher samples from the student's, and linear losses on their logits train the student without auxiliary score fitting. We prove that at the discriminator optimum these losses recover the distribution-matching gradient underlying DMD, through the classical identity linking discriminator logits to log-density ratios. We further introduce gap-based reweighting, which adapts teacher supervision across noise levels from the real-data head's empirical logit gap between real and teacher samples. DMAD reaches a Fréchet Inception Distance (FID) of 1.04 with one-step generation on ImageNet-64x64, 14.47 with four-step SDXL on COCO-10K, and a VBench total score of 85.15 with four-step Wan2.1-T2V-14B, the best values among the compared few-step methods and the multi-step teachers. On MiniMax-H3-33B, our four-step student achieves overall human preference rates of 79.1% over DMD2 and 84.6% over rCM for joint audio-video generation, excluding ties. Our code, models and demos are available at https://yzmblog.github.io/projects/DMAD.
DyMD: Preserving Interaction Dynamics through Distribution Matching Distillation in Few-Step Video W…
The paper proposes a new framework, DyMD, to improve video generation and embodi…
Sherpa: Teaching LLMs to Teach Adaptively
This paper introduces Sherpa, a multi-turn reinforcement learning framework that…
Fast Speech Foundation Model Distillation Using Interleaved Stacking
This paper proposes a novel method for training acceleration of speech foundatio…
$λ$-Controlled GRPO: Turning Flow-Matching Ratio Instability into a Budgeted Resource
This paper proposes a method to stabilize training of flow-matching models in re…
StreamHear: Domain-Adapted Pseudo-Labeling for Semi-Supervised Streaming Speech Recognition
The paper introduces StreamHear, a semi-supervised pipeline for adapting pretrai…
PRISM: Distribution-Gated Flow Matching for Controllable Unpaired Image Translation
This paper introduces PRISM, a framework for unpaired image-to-image translation…
Reinforcement Learning Disrupts Gradient-Based Adversarial Optimization
This paper investigates the use of reinforcement learning to disrupt gradient-ba…
UNITE-AUDIO: Joint Learning of Continuous Tokenization and Latent Flow Matching for Text-to-Audio Ge…
This paper introduces Unite-Audio, a system that jointly learns continuous audio…