從分類器中心視角探討（無）分類器引導機制

Studying Classifier(-Free) Guidance From a Classifier-Centric Perspective

March 13, 2025

作者: Xiaoming Zhao, Alexander G. Schwing

cs.AI

摘要

無分類器指導已成為條件生成與去噪擴散模型結合的標準方法。然而，對於無分類器指導的全面理解仍顯不足。在本研究中，我們進行了一項實證研究，旨在提供對無分類器指導的新視角。具體而言，我們不僅限於探討無分類器指導，而是追溯其根源，即分類器指導，明確其推導過程中的關鍵假設，並系統性地研究分類器所扮演的角色。我們發現，無論是分類器指導還是無分類器指導，都是通過將去噪擴散軌跡遠離決策邊界來實現條件生成的，這些邊界區域通常是條件信息相互糾纏且難以學習的地方。基於這一以分類器為核心的理解，我們提出了一種基於流匹配的通用後處理步驟，旨在縮小預訓練去噪擴散模型所學分佈與真實數據分佈之間的差距，尤其是在決策邊界附近。多個數據集上的實驗驗證了所提方法的有效性。

English

Classifier-free guidance has become a staple for conditional generation with denoising diffusion models. However, a comprehensive understanding of classifier-free guidance is still missing. In this work, we carry out an empirical study to provide a fresh perspective on classifier-free guidance. Concretely, instead of solely focusing on classifier-free guidance, we trace back to the root, i.e., classifier guidance, pinpoint the key assumption for the derivation, and conduct a systematic study to understand the role of the classifier. We find that both classifier guidance and classifier-free guidance achieve conditional generation by pushing the denoising diffusion trajectories away from decision boundaries, i.e., areas where conditional information is usually entangled and is hard to learn. Based on this classifier-centric understanding, we propose a generic postprocessing step built upon flow-matching to shrink the gap between the learned distribution for a pre-trained denoising diffusion model and the real data distribution, majorly around the decision boundaries. Experiments on various datasets verify the effectiveness of the proposed approach.