- 博客(3)
- 收藏
- 关注
RSS订阅原创 论文阅读:Unsupervised Deep Learning by Neighbourhood Discovery
Unsupervised Deep Learning by Neighbourhood Discovery论文的ideaUnsupervised Feature Learning via Non-Parametric Instance Discrimination提出将每一个sample看作一个instance，这样便可以learn出比较好的结果。这样即便是类别一致的instance也要被强制...
2019-12-11 15:51:16
974
原创 论文阅读：Representation Learning with Contrastive Predictive Coding
Representation Learning with Contrastive Predictive Coding 参考一些NLP方法 理解Contrastive LearningCPCMotivation: 该论文认为，随着预测未来更多的帧，一些noisy的低层信息会被忽略掉，更多的共享信息(该论文称之为slow features)会被提取到，这也是一些更应被关注的信息。论文举了几...
2019-12-09 16:18:10
4621
2
原创 Mutual Information的理解及应用
Understanding To Definitions in Information TheoryKL Divergence The KL Divergence depict the divergence of two distribution, for example. Actually, we do not know the correlation between the two ran...
2019-12-07 16:38:43
649
空空如也
空空如也
TA创建的收藏夹 TA关注的收藏夹
TA关注的人