MCLR: Improving Conditional Modeling via Inter-Class Likelihood-Ratio Maximization and Unifying Classifier-Free Guidance with Alignment Objectives

TL;DR

This paper introduces MCLR, an alignment training objective that enhances diffusion models' conditional generation by internalizing classifier-free guidance effects, improving guidance-free performance and providing a theoretical understanding.

Contribution

The paper proposes MCLR, a novel training objective that maximizes inter-class likelihood ratios, unifying classifier-free guidance with principled training and offering theoretical insights.

Findings

Fine-tuning with MCLR improves guidance-free conditional generation.

MCLR narrows the gap between guidance-free and CFG-based sampling.

Theoretical analysis links CFG to an optimal solution of a weighted MCLR objective.

Abstract

Diffusion models achieve strong performance in generative modeling, but their success often relies heavily on classifier-free guidance (CFG), an inference-time heuristic that modifies the sampling trajectory. In theory, diffusion models trained with standard denoising score matching (DSM) should recover the target data distribution, raising two fundamental questions: (i) why is inference-time guidance necessary in practice, and (ii) can its underlying effect be internalized into a principled training objective? In this work, we argue that a key limitation of standard DSM is insufficient inter-class separation. To address this issue, we propose MCLR, an alignment objective that explicitly maximizes inter-class likelihood-ratios during training. Fine-tuning diffusion models with MCLR induces CFG-like improvements under standard sampling, substantially improving guidance-free conditional generation and narrowing the gap to inference-time CFG. Beyond these empirical benefits, we show theoretically that the CFG-guided score is exactly the optimal solution to a sample-adaptive weighted MCLR objective. This result connects CFG to alignment-based objectives, providing a mechanistic interpretation of CFG as an implicit inference-time contrastive alignment procedure.