Table 3.

The ConvLSTM architecture (ConvLSTM is our proposed model that combines convolutional and recurrent models). To keep the architecture clear, we omitted the input layer (layer 00) and the dropout layers (the even layer indices) applied after each convolutional neural network (CNN) layer. N was set to 128.

Layer Index	01	03	05	07	09	11	12
type of filter	CNN	CNN	CNN	CNN	CNN	LSTM	Dense
number of filters	N	N	N	$\frac{3}{4}N$	$\frac{3}{4}N$	N	2