OpenMatch: Open-set Consistency Regularization for Semi-supervised Learning with Outliers

Abstract

Semi-supervised learning (SSL) is an effective means to leverage unlabeled data to improve a model's performance. Typical SSL methods like FixMatch assume that labeled and unlabeled data share the same label space. However, in practice, unlabeled data can contain categories unseen in the labeled set, i.e., outliers, which can significantly harm the performance of SSL algorithms. To address this problem, we propose a novel Open-set Semi-Supervised Learning (OSSL) approach called OpenMatch. Learning representations of inliers while rejecting outliers is essential for the success of OSSL. To this end, OpenMatch unifies FixMatch with novelty detection based on one-vs-all (OVA) classifiers. The OVA-classifier outputs the confidence score of a sample being an inlier, providing a threshold to detect outliers. Another key contribution is an open-set soft-consistency regularization loss, which enhances the smoothness of the OVA-classifier with respect to input transformations and greatly improves outlier detection. OpenMatch achieves stateof-the-art performance on three datasets, and even outperforms a fully supervised model in detecting outliers unseen in unlabeled data on CIFAR10.Preprint. Under review.

Open-Set Semi-Supervised Learning for 3D Point Cloud Understanding

“…Our baseline is conducted with DGCNN backbone trained by FixMatch semi-supervised loss with all unlabeled weights fixed to 1. We compare our method with 4 most relevant and up-to-date methods, DS3L [8], Multi-OS [33], LTWA [22] and OP-Match [24]. We migrate them to our point cloud setting and combine them with the same FixMatch SSL.…”

“…out-of-distribution data (OOD) [4], [8]. SSL under such a setting is commonly referred to as open-set semi-supervised learning [33], [24].…”

“…The recent attempts to address open-set SSL can be classified into two categories. The first genre approaches from some heuristic criterion [4], [33], [24], e.g. the consistency of predictions upon multiple augmentations or training an outlier detector.…”

“…Open-Set Semi-Supervised Learning. Open-set semisupervised learning [4], [33], [8], [18], [24] refers to the challenging case where unlabeled data may contain harmful data, sometimes referred to as out-of-distribution data (OOD). Open-set SSL can be roughly classified into two genres.…”

“…Open-set SSL can be roughly classified into two genres. The heuristic OOD detection approaches often explicitly detect unlabeled OOD samples, through training an OOD detector [33], [24] or calculating consistency of pseudolabel predictions [4], [16], [18]. Some heuristic thresholding is required to distinguish outliers from inliers.…”