FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence

21 Jan 2020Kihyuk SohnDavid BerthelotChun-Liang LiZizhao ZhangNicholas CarliniEkin D. CubukAlex KurakinHan ZhangColin Raffel

Semi-supervised learning (SSL) provides an effective means of leveraging unlabeled data to improve a model's performance. In this paper, we demonstrate the power of a simple combination of two common SSL methods: consistency regularization and pseudo-labeling... (read more)

PDF Abstract
TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT LEADERBOARD
Semi-Supervised Image Classification cifar-100, 10000 Labels FixMatch (CTA) Accuracy 76.82 # 2
Semi-Supervised Image Classification CIFAR-100, 2500 Labels FixMatch (CTA) Percentage error 28.64 # 1
Semi-Supervised Image Classification CIFAR-100, 400 Labels FixMatch (CTA) Percentage error 49.95 # 1
Semi-Supervised Image Classification CIFAR-10, 250 Labels FixMatch Accuracy 94.93 # 1
Semi-Supervised Image Classification CIFAR-10, 4000 Labels FixMatch (CTA) Accuracy 95.69 # 2
Semi-Supervised Image Classification CIFAR-10, 40 Labels FixMatch (CTA) Percentage error 11.39 # 1
Semi-Supervised Image Classification ImageNet - 10% labeled data FixMatch Top 5 Accuracy 89.13% # 11
Semi-Supervised Image Classification STL-10, 1000 Labels FixMatch (CTA) Accuracy 94.83 # 1
Semi-Supervised Image Classification SVHN, 1000 labels FixMatch (CTA) Accuracy 97.64 # 1
Semi-Supervised Image Classification SVHN, 40 Labels FixMatch (CTA) Percentage error 7.65 # 1

Methods used in the Paper


METHOD TYPE
🤖 No Methods Found Help the community by adding them if they're not listed; e.g. Deep Residual Learning for Image Recognition uses ResNet