---
title: "重新思考在线策略扩散蒸馏中的分类器自由引导"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.24731v1"
permalink: "https://feed.astralor.com/items/itm_2502596a980056125090073e"
published: "2026-07-27T17:57:02.000Z"
score: 79
channel: "ai"
tags: ["分类器自由引导","在线策略蒸馏","扩散模型蒸馏","正方向匹配","负分支不对称"]
---

# 重新思考在线策略扩散蒸馏中的分类器自由引导

## AI 点评

> 这篇论文直击扩散模型蒸馏中一个容易被忽视的细节——分类器自由引导的负分支不对称问题，理论分析清晰且实验验证有力。对于正在做扩散模型微调或知识蒸馏的工程师和研究者来说，是值得一读的参考。

## 速览

在线策略蒸馏（OPD）通过查询教师模型来适配扩散模型，但分类器自由引导（CFG）作为现代扩散系统的默认组件，其行为在OPD中尚未被充分理解。现有方法直接匹配教师和学生的引导速度，但本文发现该目标在分支层面存在欠识别问题：正负分支错误可相互补偿。通过对比案例，作者发现当教师负分支包含学生无法获取的特权信息时，朴素匹配会导致负分支不对称（NBA）失败模式。为此，提出正方向匹配（PDM），一种分支感知的OPD目标，单独约束正预测和CFG条件方向。在稠密到稀疏视频控制任务中，PDM显著提升知识迁移的鲁棒性和有效性。

## 关键点

- 现有在线策略蒸馏方法直接匹配引导速度，但该目标在分支层面欠识别，正负分支错误可相互抵消。
- 发现负分支不对称（NBA）失败模式：教师负分支特权信息导致朴素匹配时正分支错误下降而负分支错误上升。
- 提出正方向匹配（PDM），分支感知的蒸馏目标，单独约束正预测和CFG条件方向。
- 在稠密到稀疏视频控制任务中验证PDM对推理引导尺度更鲁棒，知识迁移更有效。

## 全文

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.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_2502596a980056125090073e