论文题目 mixup: Beyond Empirical Risk Minimization
2017(ICLR2018),Hongyi Zhang et al. Mixup ,MIT和FAIR
一作的知乎回答 https://www.zhihu.com/question/67472285
论文笔记 https://blog.csdn.net/u013841196/article/details/81049968
mixup的应用:
论文名称:Bag of Tricks for Image Classification with Convolutional Neural Networks
这篇文章全都是tricks,讲的是图像分类问题里的技巧
笔记 https://www.jianshu.com/p/0e0bc5dc300a
关键词:
- Large-batch
- Linear scaling learning rate(线性缩放学习率)
- Learning rate warmup(学习率热身)
- FP32 FP16
- Cosine Learning Rate Decay(余弦学习率衰减)
- Label Smoothing(标签平滑)
- Knowledge Distillation(知识蒸馏)
- Mixup Training(混合训练)