Why Are DMD Students Lazy? Understanding the Copying Behavior in Few-Step Distillation

• URL: http://arxiv.org/abs/2606.02237v1
• Date: Mon, 01 Jun 2026 13:30:19 GMT
• Title: Why Are DMD Students Lazy? Understanding the Copying Behavior in Few-Step Distillation
• Abstract summary: Distribution Matching Distillation (DMD) compresses pretrained diffusion models into efficient few-step generators by aligning their noised distributions across all scales.<n>Surprisingly, we find that in high-dimensional settings, distilled students spontaneously reproduce the original noise-data pairings of the teacher, a phenomenon we term copying.
• Score: 66.15462638744619
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Distribution Matching Distillation (DMD) compresses pretrained diffusion models into efficient few-step generators by aligning their noised distributions across all scales. In principle, such distribution-level supervision remains agnostic to specific noise-data pairings of the teacher; this provides the student the freedom to remap latent noise, a behavior consistently observed in low-dimensional settings. Surprisingly, we find that in high-dimensional settings, distilled students spontaneously reproduce the original noise-data pairings of the teacher, a phenomenon we term copying. We demonstrate that copying is neither a byproduct of adversarial objectives nor a result of teacher memorization. Instead, our evidence suggests that copying is an emergent property arising from the limited geometric freedom of the student model during high-dimensional distillation.

Related papers

• DistillAlign: Coordinating Mode Covering and Mode Seeking in Autoregressive Video Distillation [29.992105358526715]
  We introduce a distributional evaluation protocol that measures precision and coverage between student and teacher distributions.<n>We propose joint distillation, which combines DMD's mode-seeking objective with a Consistency Distillation-based mode-covering constraint.
  arXiv  Detail & Related papers  (2026-07-29T11:58:47Z)
• Provably Learning Diffusion Models under the Manifold Hypothesis: Collapse and Refine [60.669081685261965]
  Diffusion models generate high-dimensional data with remarkable quality.<n>How their training efficiently learns the score function remains theoretically unexplained.<n>We instantiate this principle as Score-induced Latent Diffusion (SiLD)
  arXiv  Detail & Related papers  (2026-05-16T16:51:10Z)
• A Survey of On-Policy Distillation for Large Language Models [19.95776080082138]
  This survey provides the first comprehensive overview of On-Policy Distillation for Large Language Models (LLMs)<n>We introduce a unified $f$-divergence framework over on-policy samples and organize the landscape along three dimensions: emphfeedback signal (logit-based, outcome-based, or self-play), emphteacher access (white-box, black-box, or teacher-free), and emphloss minimization (token-level, sequence-level, or hybrid)
  arXiv  Detail & Related papers  (2026-04-01T08:32:34Z)
• Knowledge Distillation of Uncertainty using Deep Latent Factor Model [10.148306002388196]
  We introduce a new method of distribution distillation called Gaussian distillation.<n>It estimates the distribution of a teacher ensemble through a special Gaussian process called the deep latent factor model (DLF)<n>By using multiple benchmark datasets, we demonstrate that the proposed Gaussian distillation outperforms existing baselines.
  arXiv  Detail & Related papers  (2025-10-22T06:46:59Z)
• Learning from Stochastic Teacher Representations Using Student-Guided Knowledge Distillation [64.15918654558816]
  Self-distillation (SSD) training strategy is introduced for filtering and weighting teacher representation to distill from task-relevant representations only.<n> Experimental results on real-world affective computing, wearable/biosignal datasets from the UCR Archive, the HAR dataset, and image classification datasets show that the proposed SSD method can outperform state-of-the-art methods.
  arXiv  Detail & Related papers  (2025-04-19T14:08:56Z)
• Distribution Backtracking Builds A Faster Convergence Trajectory for Diffusion Distillation [19.88187051373436]
  We propose Distribution Backtracking Distillation (DisBack) to speed up the sampling speed of diffusion models.<n>DisBack achieves faster and better convergence than the existing distillation method, with FID score of 1.38 on ImageNet 64x64 dataset.
  arXiv  Detail & Related papers  (2024-08-28T17:58:17Z)
• Improved Distribution Matching Distillation for Fast Image Synthesis [54.72356560597428]
  We introduce DMD2, a set of techniques that lift this limitation and improve DMD training. First, we eliminate the regression loss and the need for expensive dataset construction. Second, we integrate a GAN loss into the distillation procedure, discriminating between generated samples and real images.
  arXiv  Detail & Related papers  (2024-05-23T17:59:49Z)
• On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes [44.97759066341107]
  Generalized Knowledge Distillation (GKD) trains the student on its self-generated output sequences by leveraging feedback from the teacher. We demonstrate the efficacy of GKD for distilling auto-regressive language models on summarization, translation, and arithmetic reasoning tasks.
  arXiv  Detail & Related papers  (2023-06-23T17:56:26Z)
• HomoDistil: Homotopic Task-Agnostic Distillation of Pre-trained Transformers [49.79405257763856]
  This paper focuses on task-agnostic distillation. It produces a compact pre-trained model that can be easily fine-tuned on various tasks with small computational costs and memory footprints. We propose Homotopic Distillation (HomoDistil), a novel task-agnostic distillation approach equipped with iterative pruning.
  arXiv  Detail & Related papers  (2023-02-19T17:37:24Z)
• Teacher's pet: understanding and mitigating biases in distillation [61.44867470297283]
  Several works have shown that distillation significantly boosts the student's overall performance. However, are these gains uniform across all data subgroups? We show that distillation can harm performance on certain subgroups. We present techniques which soften the teacher influence for subgroups where it is less reliable.
  arXiv  Detail & Related papers  (2021-06-19T13:06:25Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.