Open junxnone opened 4 years ago
A variable-length String
set-selection
θc
k
+Anchor Point
N-1 sigmoids
N-1 Layer
Compilation Failures
Input Layer
0-padding
N-1
j
0 ~ N-1
junxnone/tech-io#805 junxnone/aiwiki#395
Neural Architecture Search with Reinforcement Learning
A variable-length String
set-selection
Pipeline
Training
θc
k
个子网络 AccuracySkip-Connections
+Anchor Point
=N-1 sigmoids
- 指定之前N-1 Layer
的连接关系Compilation Failures
- 当 Layer 与其他 Layer 不兼容, 或者 Layer 没有输入或者输出时Input Layer
0-padding
比较小的 LayerN-1
Anchor Point 函数j
层 Anchor Point 隐藏状态0 ~ N-1
RNN Cell
LSTM - xt/ht-1/ct-1
LSTM - ht/ct
Reference