Learning Robust Diffusion Models from Imprecise Supervision

2025-10-03

AI Summary

Conditional diffusion models suffer from conditional mismatch under imprecise supervision—e.g., noisy, blurry, or missing labels—leading to degraded generation quality. This work presents the first systematic study of imprecise supervision in diffusion models and proposes DMIS, a robust training framework. DMIS decouples the generative objective from the classification objective, introduces a diffusion-based classifier to explicitly model class posterior distributions, and designs an optimized timestep sampling strategy to enhance posterior inference efficiency; its objective is formulated via likelihood maximization. Experiments demonstrate that DMIS consistently generates high-fidelity, class-discriminative samples across diverse tasks—including image synthesis, weakly supervised learning, and noisy data distillation—while significantly improving robustness to label noise compared to standard conditional diffusion baselines.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Conditional diffusion models have achieved remarkable success in various generative tasks recently, but their training typically relies on large-scale datasets that inevitably contain imprecise information in conditional inputs. Such supervision, often stemming from noisy, ambiguous, or incomplete labels, will cause condition mismatch and degrade generation quality. To address this challenge, we propose DMIS, a unified framework for training robust Diffusion Models from Imprecise Supervision, which is the first systematic study within diffusion models. Our framework is derived from likelihood maximization and decomposes the objective into generative and classification components: the generative component models imprecise-label distributions, while the classification component leverages a diffusion classifier to infer class-posterior probabilities, with its efficiency further improved by an optimized timestep sampling strategy. Extensive experiments on diverse forms of imprecise supervision, covering tasks of image generation, weakly supervised learning, and noisy dataset condensation demonstrate that DMIS consistently produces high-quality and class-discriminative samples.

Problem

• Addressing imprecise supervision in conditional diffusion models training
• Mitigating condition mismatch from noisy ambiguous or incomplete labels
• Improving generation quality under diverse imprecise supervision scenarios

Innovation

• Training robust diffusion models with imprecise supervision
• Decomposing objective into generative and classification components
• Using diffusion classifier with optimized timestep sampling