Noisy-Pair Robust Representation Alignment for Positive-Unlabeled Learning

2025-09-30

AI Summary

In positive-unlabeled (PU) learning, unreliable supervision impedes discriminative representation learning. To address this, we propose NcPU—a non-contrastive framework that requires neither auxiliary negative samples nor prior parameters. Its core components are a noise-robust non-contrastive loss (NoiSNCL) and a regret-based pseudo-label disambiguation mechanism (PLD), jointly optimized within an EM framework to enable stable representation alignment without auxiliary information—achieved for the first time. NcPU integrates non-contrastive learning, representation alignment, and regret-driven label correction, supported by theoretical analysis and iterative optimization. On challenging benchmarks such as CIFAR-100, it significantly outperforms existing PU methods, narrowing the performance gap by 14.26%. Furthermore, it demonstrates practical utility in a real-world post-disaster building damage assessment task.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Adversarial Learning & Robustness
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

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

Abstract

Positive-Unlabeled (PU) learning aims to train a binary classifier (positive vs. negative) where only limited positive data and abundant unlabeled data are available. While widely applicable, state-of-the-art PU learning methods substantially underperform their supervised counterparts on complex datasets, especially without auxiliary negatives or pre-estimated parameters (e.g., a 14.26% gap on CIFAR-100 dataset). We identify the primary bottleneck as the challenge of learning discriminative representations under unreliable supervision. To tackle this challenge, we propose NcPU, a non-contrastive PU learning framework that requires no auxiliary information. NcPU combines a noisy-pair robust supervised non-contrastive loss (NoiSNCL), which aligns intra-class representations despite unreliable supervision, with a phantom label disambiguation (PLD) scheme that supplies conservative negative supervision via regret-based label updates. Theoretically, NoiSNCL and PLD can iteratively benefit each other from the perspective of the Expectation-Maximization framework. Empirically, extensive experiments demonstrate that: (1) NoiSNCL enables simple PU methods to achieve competitive performance; and (2) NcPU achieves substantial improvements over state-of-the-art PU methods across diverse datasets, including challenging datasets on post-disaster building damage mapping, highlighting its promise for real-world applications. Code: Code will be open-sourced after review.

Problem

• Learning binary classifiers with limited positive and abundant unlabeled data
• Addressing performance gap between PU learning and supervised methods
• Learning discriminative representations under unreliable supervision conditions

Innovation

• Non-contrastive framework for PU learning without auxiliary data
• Noisy-pair robust loss aligns intra-class representations reliably
• Phantom label disambiguation provides conservative negative supervision