ScanMix: Learning from Severe Label Noise via Semantic Clustering and Semi-Supervised Learning

by   Ragav Sachdeva, et al.

In this paper, we address the problem of training deep neural networks in the presence of severe label noise. Our proposed training algorithm ScanMix, combines semantic clustering with semi-supervised learning (SSL) to improve the feature representations and enable an accurate identification of noisy samples, even in severe label noise scenarios. To be specific, ScanMix is designed based on the expectation maximisation (EM) framework, where the E-step estimates the value of a latent variable to cluster the training images based on their appearance representations and classification results, and the M-step optimises the SSL classification and learns effective feature representations via semantic clustering. In our evaluations, we show state-of-the-art results on standard benchmarks for symmetric, asymmetric and semantic label noise on CIFAR-10 and CIFAR-100, as well as large scale real label noise on WebVision. Most notably, for the benchmarks contaminated with large noise rates (80 above), our results are up to 27 available at


page 1

page 2

page 3

page 4


ReLaB: Reliable Label Bootstrapping for Semi-Supervised Learning

Reducing the amount of labels required to trainconvolutional neural netw...

Semi-Supervised Learning with Self-Supervised Networks

Recent advances in semi-supervised learning have shown tremendous potent...

Reliable Label Correction is a Good Booster When Learning with Extremely Noisy Labels

Learning with noisy labels has aroused much research interest since data...

Building One-Shot Semi-supervised (BOSS) Learning up to Fully Supervised Performance

Reaching the performance of fully supervised learning with unlabeled dat...

EvidentialMix: Learning with Combined Open-set and Closed-set Noisy Labels

The efficacy of deep learning depends on large-scale data sets that have...

Semi-supervised Learning for Discrete Choice Models

We introduce a semi-supervised discrete choice model to calibrate discre...

Adversarial Ladder Networks

The use of unsupervised data in addition to supervised data in training ...