chullhwan-song / Reading-Paper

152 stars 27 forks source link

Self-training with Noisy Student improves ImageNet classification #246

Open chullhwan-song opened 5 years ago

chullhwan-song commented 5 years ago
chullhwan-song commented 4 years ago

Abstract

방법 간단요약

1. Teacher model을 labeled data로 학습
2. Teacher model로 unlabeled data의 label를 예측 > pseudo labels
3. Student model로 labeled data와 unlabeled data를 학습
     * pseudo labels > hard or soft
4. 학습이 끝난 student model은 teacher model이 되고, 새로운 student model으로 2, 3 을 반복

Introduction

Self-training with Noisy Student

Noising Student

Experiments

데이터셋