Rethinking Classifier-Free Guidance in On-Policy Diffusion Distillation

• URL: http://arxiv.org/abs/2607.24731v2
• Date: Thu, 30 Jul 2026 05:33:15 GMT
• Title: Rethinking Classifier-Free Guidance in On-Policy Diffusion Distillation
• Abstract summary: On-policy distillation (OPD) adapts diffusion models by querying a teacher along trajectories generated by the current student.<n>Existing OPD methods naturally extend velocity matching to the CFG-composed prediction.<n>We show that this objective is under-identified at the branch level.<n>We introduce Positive--Direction Matching (PDM), a branch-aware OPD objective that separately constrains the positive prediction and the CFG conditional direction.
• Score: 14.183839885490372
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: On-policy distillation (OPD) adapts diffusion models by querying a teacher along trajectories generated by the current student, but how it should behave under classifier-free guidance (CFG), a default component of modern diffusion systems, remains poorly understood. Existing OPD methods naturally extend velocity matching to the CFG-composed prediction, directly matching teacher and student guided velocities. We show that this objective is under-identified at the branch level: positive- and negative-branch errors can compensate in the guided prediction. Through two contrasting cases, we find that naive matching remains effective under shared negative conditioning, where both branch errors decrease jointly. When the model's native CFG schema retains privileged information in the teacher's negative branch that is unavailable to the student, however, this joint reduction breaks down and the composed objective induces antagonistic branch-error dynamics, reducing the positive-branch error while increasing the negative-branch error. We term this failure mode Negative Branch Asymmetry (NBA). To address NBA, we introduce Positive--Direction Matching (PDM), a branch-aware OPD objective that separately constrains the positive prediction and the CFG conditional direction. We apply PDM to dense-to-sparse video control, where naive guided matching is highly sensitive to inference guidance scales, while branch-aware supervision enables more robust and effective knowledge transfer.

Related papers

• Learning from Consensus and Disagreement: Unsupervised On-Policy Self-Distillation with Minority-Trajectory Contrast [15.292903130335974]
  On-policy self-distillation improves language-model reasoning by querying a teacher on states actually visited by the student.<n>Recent methods create a powerful information asymmetry by exposing the teacher to privileged context.<n>We introduce CoDA, a fully unsupervised framework that creates reliable privileged information entirely from the latent uncertainty structure of a model's own rollouts.
  arXiv  Detail & Related papers  (2026-08-09T15:23:25Z)
• Adaptive Supervised Anchoring for On-Policy Self-Distillation [15.445625053566244]
  On-policy self-distillation adapts a language model by distilling guidance from a frozen teacher on trajectories sampled from the student.<n>We show that when student rollouts drift from target trajectories, conditioning the teacher on off-target prefixes substantially weakens its task-relevant supervision.<n>We propose a unified training framework that separates two complementary supervision pathways.
  arXiv  Detail & Related papers  (2026-08-08T05:46:32Z)
• Purified OPSD: On-Policy Self-Distillation Without Losing How to Think [52.879387744920415]
  We find that on-policy self-distillation (OPSD) consistently fails on long chain-of-thought (long-CoT) reasoning models.<n>We propose a two-step solution: first, we construct a reference-only teacher to isolate the non-transferable component of the supervision signal.<n>Second, we use pointwise mutual information (PMI) as the mechanism to transform this residual into a well-formed PMI target distribution.
  arXiv  Detail & Related papers  (2026-07-02T14:33:07Z)
• Diffusing Blame: Task-Dependent Credit Assignment in Biologically Plausible Dual-Stream Networks [20.760754362327933]
  We introduce modulo error routing to extend Error Diffusion beyond binary classification.<n>We show that a dual-stream excitatory/inhibitory architecture trained with this method achieves 96.7% on MNIST.<n>In reinforcement learning, we integrate ED with Proximal Policy Optimization (PPO) and evaluate it on continuous-control tasks in Google Brax and on Craftax.
  arXiv  Detail & Related papers  (2026-06-30T14:09:40Z)
• RLCSD: Reinforcement Learning with Contrastive On-Policy Self-Distillation [46.554172283247915]
  On-policy self-distillation provides dense, token-level supervision for reasoning models.<n>We show that the learning signal drawn from this distributional gap concentrates on style tokens rather than task-bearing ones.<n>We propose textbfRLCSD (Reinforcement Learning with Contrastive on-policy Self-Distillation) to mitigate this drift.
  arXiv  Detail & Related papers  (2026-06-10T06:31:59Z)
• Mitigating Error Amplification in Fast Adversarial Training [58.74042726356826]
  Fast Adversarial Training (FAT) has proven effective in enhancing model robustness by encouraging networks to learn perturbation-invariant representations.<n>FAT often suffers from catastrophic overfitting (CO), where the model overfits to the training attack and fails to generalize to unseen ones.<n>We propose a Distribution-aware Dynamic Guidance (DDG) strategy that dynamically adjusts both the perturbation budget and supervision signal.
  arXiv  Detail & Related papers  (2026-04-27T11:23:18Z)
• Stable On-Policy Distillation through Adaptive Target Reformulation [7.361248172930405]
  Veto is an objective-level reformulation that constructs a geometric bridge in the logit space.<n>Veto consistently outperforms supervised fine-tuning and existing on-policy baselines.
  arXiv  Detail & Related papers  (2026-01-12T02:57:39Z)
• Normalized Attention Guidance: Universal Negative Guidance for Diffusion Models [57.20761595019967]
  We present Normalized Attention Guidance (NAG), an efficient, training-free mechanism that applies extrapolation in attention space with L1-based normalization and refinement.<n>NAG restores effective negative guidance where CFG collapses while maintaining fidelity.<n>NAG generalizes across architectures (UNet, DiT), sampling regimes (few-step, multi-step), and modalities (image, video)
  arXiv  Detail & Related papers  (2025-05-27T13:30:46Z)
• Contrastive CFG: Improving CFG in Diffusion Models by Contrasting Positive and Negative Concepts [55.298031232672734]
  As-Free Guidance (CFG) has proven effective in conditional diffusion model sampling for improved condition alignment. We present a novel method to enhance negative CFG guidance using contrastive loss.
  arXiv  Detail & Related papers  (2024-11-26T03:29:27Z)
• Perturbation-Invariant Adversarial Training for Neural Ranking Models: Improving the Effectiveness-Robustness Trade-Off [107.35833747750446]
  adversarial examples can be crafted by adding imperceptible perturbations to legitimate documents. This vulnerability raises significant concerns about their reliability and hinders the widespread deployment of NRMs. In this study, we establish theoretical guarantees regarding the effectiveness-robustness trade-off in NRMs.
  arXiv  Detail & Related papers  (2023-12-16T05:38:39Z)
• ELODI: Ensemble Logit Difference Inhibition for Positive-Congruent Training [110.52785254565518]
  Existing methods to reduce the negative flip rate (NFR) either do so at the expense of overall accuracy by forcing a new model to imitate the old models, or use ensembles. We analyze the role of ensembles in reducing NFR and observe that they remove negative flips that are typically not close to the decision boundary. We present a method, called Ensemble Logit Difference Inhibition (ELODI), to train a classification system that achieves paragon performance in both error rate and NFR.
  arXiv  Detail & Related papers  (2022-05-12T17:59:56Z)
• Robustness through Cognitive Dissociation Mitigation in Contrastive Adversarial Training [2.538209532048867]
  We introduce a novel neural network training framework that increases model's adversarial robustness to adversarial attacks. We propose to improve model robustness to adversarial attacks by learning feature representations consistent under both data augmentations and adversarial perturbations. We validate our method on the CIFAR-10 dataset on which it outperforms both robust accuracy and clean accuracy over alternative supervised and self-supervised adversarial learning methods.
  arXiv  Detail & Related papers  (2022-03-16T21:41:27Z)

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.