[ConvLSTM] Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting阅读笔记
1、主要创新相比于FC-LSTM,引入卷积层,1)可以充分利用空间关联,2)能够输出图像序列,3)减少参数冗余2、网络1)输入时空序列,特征数目=P,图像大小=MxN,即2)输出3)网络设计i:inputgate,f:forgetgate,c:cellstatus,h:finalstate,o:outputgate,*:convolutionoperator,o:Hadamardproduct->