Unsupervised Learning for Class Distribution Mismatch

2025-05-11

AI Summary

This paper addresses unsupervised open-set recognition under class distribution mismatch (CDM) between training and target tasks. Methodologically, it introduces the first fully unsupervised CDM modeling framework: (1) a novel diffusion-based paradigm for generating semantically controllable positive–negative sample pairs, enabling precise addition or removal of class semantics; and (2) a confidence-driven dynamic pseudo-labeling iteration mechanism that eliminates reliance on labeled data or semi-supervised assumptions. Evaluated on Tiny-ImageNet with 60% distribution mismatch, the method significantly outperforms OpenMatch (trained with 40 labels per class), achieving absolute improvements of 35.1%, 63.7%, and 72.5% in accuracy for known classes, unknown classes, and novel classes, respectively. These results demonstrate superior joint identification capability in open-world scenarios under severe distribution shift.

Technology Category

• Machine Learning: Semi-Supervised Learning
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Class distribution mismatch (CDM) refers to the discrepancy between class distributions in training data and target tasks. Previous methods address this by designing classifiers to categorize classes known during training, while grouping unknown or new classes into an"other"category. However, they focus on semi-supervised scenarios and heavily rely on labeled data, limiting their applicability and performance. To address this, we propose Unsupervised Learning for Class Distribution Mismatch (UCDM), which constructs positive-negative pairs from unlabeled data for classifier training. Our approach randomly samples images and uses a diffusion model to add or erase semantic classes, synthesizing diverse training pairs. Additionally, we introduce a confidence-based labeling mechanism that iteratively assigns pseudo-labels to valuable real-world data and incorporates them into the training process. Extensive experiments on three datasets demonstrate UCDM's superiority over previous semi-supervised methods. Specifically, with a 60% mismatch proportion on Tiny-ImageNet dataset, our approach, without relying on labeled data, surpasses OpenMatch (with 40 labels per class) by 35.1%, 63.7%, and 72.5% in classifying known, unknown, and new classes.

Problem

• Addressing class distribution mismatch without labeled data
• Training classifiers using synthesized positive-negative pairs
• Improving classification of known and unknown classes

Innovation

• Uses diffusion model to synthesize training pairs
• Employs confidence-based pseudo-labeling for real-world data
• Trains classifier with positive-negative pairs from unlabeled data