PSPU: Enhanced Positive and Unlabeled Learning by Leveraging Pseudo Supervision

TL;DR

This paper introduces PSPU, a novel pseudo-supervised framework for positive and unlabeled learning that improves model performance by leveraging confident samples and consistency loss, outperforming existing methods on multiple datasets.

Contribution

The paper proposes a pseudo-supervised PU learning framework that enhances model accuracy by iteratively gathering confident samples and applying non-PU objectives, with added consistency loss.

Findings

Significant performance improvements on MNIST, CIFAR-10, CIFAR-100 datasets.

Effective handling of both balanced and imbalanced data scenarios.

Competitive results on industrial anomaly detection with MVTecAD.

Abstract

Positive and Unlabeled (PU) learning, a binary classification model trained with only positive and unlabeled data, generally suffers from overfitted risk estimation due to inconsistent data distributions. To address this, we introduce a pseudo-supervised PU learning framework (PSPU), in which we train the PU model first, use it to gather confident samples for the pseudo supervision, and then apply these supervision to correct the PU model's weights by leveraging non-PU objectives. We also incorporate an additional consistency loss to mitigate noisy sample effects. Our PSPU outperforms recent PU learning methods significantly on MNIST, CIFAR-10, CIFAR-100 in both balanced and imbalanced settings, and enjoys competitive performance on MVTecAD for industrial anomaly detection.