【Distill 系列：三】On the Efficacy of Knowledge Distillation
https://arxiv.org/pdf/1910.01348.pdf teacher不是性能越高越好（这个我实验验证得到的结果一致） teacher训练中early stop（未尝试），蒸馏中early stop（我这里不work）有利于提高蒸馏效果 Method 一个潜意识的猜想：性能越高的teacher的蒸馏效果越好 可以看到，随着teacher模型变大，蒸馏的student性能并没有依次
相关文章
- 1. On the Efficacy of Knowledge Distillation
- 2. Awesome Knowledge-Distillation
- 3. 【Distill 系列：一】bmvc2019 Learning Efficient Detector with Semi-supervised Adaptive Distillation
- 4. Knowledge Distillation
- 5. Knowledge Distillation 笔记
- 6. Tutorial: Knowledge Distillation
- 7. 【distill.&transfer】Deep Face Recognition Model Compression via Knowledge Transfer and Distillation
- 8. Knowledge Distillation by On-the-Fly Native Ensemble论文解读
- 9. 在线多分支融合——Knowledge Distillation by On-the-Fly Native Ensemble
- 10. 深入浅出：Knowledge Distillation by On-the-Fly Native Ensemble
- 更多相关文章...
- • XSLT
- • XSLT
- • Docker容器实战(七) - 容器眼光下的文件系统
- • RxJava操作符（三）Filtering
相关标签/搜索
每日一句
-
每一个你不满意的现在，都有一个你没有努力的曾经。
欢迎关注本站公众号,获取更多信息
相关文章
- 1. On the Efficacy of Knowledge Distillation
- 2. Awesome Knowledge-Distillation
- 3. 【Distill 系列：一】bmvc2019 Learning Efficient Detector with Semi-supervised Adaptive Distillation
- 4. Knowledge Distillation
- 5. Knowledge Distillation 笔记
- 6. Tutorial: Knowledge Distillation
- 7. 【distill.&transfer】Deep Face Recognition Model Compression via Knowledge Transfer and Distillation
- 8. Knowledge Distillation by On-the-Fly Native Ensemble论文解读
- 9. 在线多分支融合——Knowledge Distillation by On-the-Fly Native Ensemble
- 10. 深入浅出：Knowledge Distillation by On-the-Fly Native Ensemble