Interpolation Consistency Training for Semi-Supervised Learning

TL;DR

Interpolation Consistency Training (ICT) is a computationally efficient semi-supervised learning method that enforces prediction consistency at interpolated data points, effectively improving neural network performance on benchmark datasets.

Contribution

ICT introduces a novel regularization technique that leverages data interpolation to enhance semi-supervised learning with neural networks.

Findings

Achieves state-of-the-art results on CIFAR-10 and SVHN datasets.

Effectively moves decision boundaries to low-density regions.

Reduces overfitting through data-adaptive regularization.

Abstract

We introduce Interpolation Consistency Training (ICT), a simple and computation efficient algorithm for training Deep Neural Networks in the semi-supervised learning paradigm. ICT encourages the prediction at an interpolation of unlabeled points to be consistent with the interpolation of the predictions at those points. In classification problems, ICT moves the decision boundary to low-density regions of the data distribution. Our experiments show that ICT achieves state-of-the-art performance when applied to standard neural network architectures on the CIFAR-10 and SVHN benchmark datasets. Our theoretical analysis shows that ICT corresponds to a certain type of data-adaptive regularization with unlabeled points which reduces overfitting to labeled points under high confidence values.