PSPU: Enhanced Positive and Unlabeled Learning by Leveraging Pseudo Supervision

TL;DR

This work introduces a pseudo-supervised PU learning framework (PSPU), in which the PU model is train 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.

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.

Figures (4)

• Figure 1: Challenges in PU net: traditional PU net suffers from overfitted risk estimation between labeled and unlabeled positive data. Note that images of deer denote the positive samples, and that of goat denote the negative samples.
• Figure 2: Gap between PU risk estimation and oracle risk estimation among different settings.
• Figure 3: The Training Process of PSPU at Epoch $e$.
• Figure 4: Precision or F1 curve during training. (a) F1 score on imbalanced CIFAR-10. (b) F1 score on extremely imbalanced CIFAR-10.