Out-of-Distributed Semantic Pruning for Robust Semi-Supervised Learning
This work addresses a semantic-level bottleneck in robust SSL, which is incremental as it builds on existing sample-level filtering methods.
The paper tackles the problem of corrupted semantic information in robust semi-supervised learning by proposing OOD Semantic Pruning (OSP), a framework that prunes out-of-distribution semantics from in-distribution features, resulting in a 13.7% accuracy improvement for ID classification and 5.9% AUROC gain for OOD detection on TinyImageNet.
Recent advances in robust semi-supervised learning (SSL) typically filter out-of-distribution (OOD) information at the sample level. We argue that an overlooked problem of robust SSL is its corrupted information on semantic level, practically limiting the development of the field. In this paper, we take an initial step to explore and propose a unified framework termed OOD Semantic Pruning (OSP), which aims at pruning OOD semantics out from in-distribution (ID) features. Specifically, (i) we propose an aliasing OOD matching module to pair each ID sample with an OOD sample with semantic overlap. (ii) We design a soft orthogonality regularization, which first transforms each ID feature by suppressing its semantic component that is collinear with paired OOD sample. It then forces the predictions before and after soft orthogonality decomposition to be consistent. Being practically simple, our method shows a strong performance in OOD detection and ID classification on challenging benchmarks. In particular, OSP surpasses the previous state-of-the-art by 13.7% on accuracy for ID classification and 5.9% on AUROC for OOD detection on TinyImageNet dataset. The source codes are publicly available at https://github.com/rain305f/OSP.