VCL-PL:Semi-Supervised Learning from Noisy Web Data with Variational Contrastive Learning

Abstract

We address the problem of web supervised learning, in particular for face attribute classification. Web data suffers from image set noise, due to unrelated images that may be retrieved in response to the query. We propose a semi-supervised pseudo-labeling approach where the embedding space distribution is learnt via variational contrastive learning. We use 40 Gaussian sampling heads for the 40 attributes in the CelebA dataset and apply supervised contrastive learning over a limited amount of labelled data, to address the multi-label face attribute classification problem. Soft pseudo-labeling is then used to label the unlabelled data at attribute level, followed by two-stage domain adaptation. We show that the proposed method using noisy web data brings improvements in accuracy over supervised multi-label face attribute classification in all experimental settings (over 2% points for very low-data setting). We suggest that learning the embedding distribution and the subsequent soft pseudo-labeling according to the nearest neighbors help in overcoming the noise in the unlabeled data.

New Horizons in Web Search, Web Data Mining, and Web-Based Applications

“…Advanced autoencoder deep learning models make extracting information from heterogeneous contexts more efficient [6]. In web image search, semi-supervised pseudo-labeling and variational contrastive learning can be used to overcome the influence of noise and obtain better retrieval performance [7]. Embracing location-based social networks into web applications enables the users to register whenever they visit a specific point-of-interest (POI) through the so-called check-ins, or to establish social links with other users in the system [8].…”