Prompt-driven efficient Open-set Semi-supervised Learning

Haoran Li, Chun-Mei Feng, Tao Zhou, Yong Xu, Xiaojun Chang

Abstract: Open-set Semi-Supervised Learning (OSSL) has always been vulnerable to the unseen categories, i.e., outliers, that have never been seen in the labeled set. Then a out-of-distribution (OOD) detector is introduced to identify outliers unseen in the labeled training data that the unlabeled data may contain, to reduce the damage to the SSl algorithm. In this work, we suggest that using a visual prompting driven mechanism to obtain higher effectiveness in the OSSL task. To this end, we propose a prompt-driven efficient OSSL framework, called OpenPrompt, which can propagate class information from labeled to unlabeled data with only a small amount of trainable parameters in the input space. Besides, a prompt-driven joint space learning mechanism is proposed to detect OOD data by maximizing the distribution gap between ID and OOD samples in unlabeled data. The experimental results on three public datasets show that OpenPrompt outperforms state-of-the-art methods with less than 1% of trainable parameters. More importantly, OpenPrompt achieves a 4% improvement in term of AUROC on outlier detection over a fully supervised model on CIFAR10.

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.

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

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Applications (eg, speech processing, computer vision, NLP)