arXiv cs.AI· Xiang Li, Yixuan Jia, Xiao Li, Jeffrey A. Fessler, Rongrong Wang, Qing Qu·· 13 小时前AI 评分31
MCLR：通过类间似然比最大化改进条件生成，并统一无分类器引导与对齐目标
MCLR: Improving Conditional Modeling via Inter-Class Likelihood-Ratio Maximization and Unifying Classifier-Free Guidance with Alignment Objectives
AI 导读
研究者提出微调目标 MCLR，通过显式最大化类间似然比增强类别特定结构，在多种模型和数据集上显著提升无引导条件生成效果并缩小与 CFG 的差距。理论上证明 CFG 引导得分是样本自适应加权、ELBO 近似的 MCLR 目标的唯一极小值点，从而将 CFG 解释为隐式的推理时对比对齐过程。研究还厘清了 MCLR 与 DPO、CCA、CFM 等对比目标的联系与差异。
正文 · 原文
Abstract:Diffusion models achieve strong generative performance but often rely on classifier-free guidance (CFG), an inference-time intervention that modifies the sampling trajectory. Since standard denoising score matching (DSM) should recover the target distribution when the score is learned accurately, the practical necessity of CFG suggests deficiencies in learned conditional scores. We identify weak inter-class separation as one such limitation: samples generated from the same noise under different labels can retain similar class-agnostic structure. To address this deficiency, we introduce MCLR, a principled fine-tuning objective that explicitly maximizes inter-class likelihood-ratios, thereby strengthening class-specific structures under standard sampling. Across diverse models and datasets, MCLR substantially improves guidance-free conditional generation and narrows the gap to CFG. Theoretically, we prove that the CFG-guided score is the unique minimizer of a sample-adaptively weighted, ELBO-approximated MCLR objective, interpreting CFG as an implicit inference-time contrastive alignment procedure. We further compare MCLR with related contrastive objectives, including DPO, CCA, and CFM, clarifying their connections and key differences in an otherwise fragmented literature.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2603.22364 [cs.LG] |
| (or arXiv:2603.22364v4 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2603.22364 arXiv-issued DOI via DataCite |
Submission history
From: Xiang Li [view email]
[v1]
Mon, 23 Mar 2026 05:01:47 UTC (45,122 KB)
[v2]
Thu, 9 Apr 2026 15:35:52 UTC (28,146 KB)
[v3]
Tue, 12 May 2026 23:23:12 UTC (21,817 KB)
[v4]
Tue, 20 Oct 2026 20:25:03 UTC (23,654 KB)
来源：arXiv cs.AI · arxiv.org