MixMatch - Pytorch Implementation

This repository is an unofficial implementation of MixMatch with Pytorch.

Original Paper: MixMatch: A Holistic Approach to Semi-Supervised Learning, by David Berthelot, Nicholas Carlini, Ian Goodfellow, Nicolas Papernot, Avital Oliver, Colin Raffel.

Original Repo (tensorflow): here

Directory structure of used dataset: CIFAR10

Key mechanisms implemented in this code

1. Augmentation
2. Guess Label
3. Entropy Minimization by Sharpen
4. MixUp
5. Consistensy Loss
6. Exponential Moving Average

The Mean-Teacher model used in this code follows the original implementation, found here

** This repo applies varied ratio of labels instead of the absolute label amount

Reference

MixMatch: A Holistic Approach to Semi-Supervised Learning, by David Berthelot, Nicholas Carlini, Ian Goodfellow, Nicolas Papernot, Avital Oliver, Colin Raffel.

Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results, by Antti Tarvainen, Harri Valpola