(SimCLRv2) Big Self-Supervised Models are Strong Semi-Supervised Learners
# (SimCLRv2)Big Self-Supervised Models are Strong Semi-Supervised Learners
# 作者:Google Hinton组
# 摘要
# 阅读
# 论文的目的及结论
# 论文的实验
# 论文的方法
# 论文的背景
# 总结
# 论文的贡献
Big CNN 用无监督学习来pretraining,然后去fine-tuning
再做一个无监督的蒸馏,train一个student model
无监督的蒸馏:用finetune的model来计算一个soft label,然后去train这个student model
# 论文的不足
# 论文如何讲故事
两个利用无标记数据的范式
- Task-agnostic use of unlabeled data
- Unsupervised pre-training + supervised fine-tuning
- Task-specific use of unlabeled data
- Self-training,pseudo-labeling
- Label consistency regularization
- Other label propogation
# 参考资料
上次更新: 2021/11/03, 23:35:28
- 02
- README 美化05-20
- 03
- 常见 Tricks 代码片段05-12