Contrastive Positive Unlabeled Learning

Anish Acharya, sujay sanghavi

Primary Area: representation learning for computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: positive unlabeled learning

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: Positive Unlabeled(PU) learning refers to the task of learning a binary classifier given a few labeled positive samples, and a set of unlabeled samples (which could be positive or negative). Majority of the existing approaches rely on additional knowledge of the class prior, which is unavailable in practice. Furthermore, these methods tend to perform poorly in low-data regimes, especially when very few positive examples are labeled. In this paper, we propose a novel PU learning framework that overcomes these limitations. We start by learning a feature space through pretext-invariant representation learning and then apply pseudo-labeling to the unlabeled examples, leveraging the cluster-preserving property of the representation space. Overall, our proposed PU learning framework handily outperforms state-of-the-art PU learning methods across several standard PU benchmark datasets, while not requiring a-priori knowledge or estimate of class prior. Remarkably, our method remains effective even when labeled data is scant, where previous PU learning algorithms falter. We also provide simple theoretical analysis motivating our proposed algorithms.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 4808