Convolutional LSTM Network is a variant of lstm network. In this tutorial, we will discuss it and help you understand and use it.
Convolutional LSTM Network
Convolutional LSTM Network is improved based on LSTM with peephole connections.
Understand LSTM Peephole Connections: A Beginner Guide
The equations of it are:
We can find: the key idea of convolutional lstm network is to replace hadamard product between xt and ht-1 with convolutional operation.
To understand how to implement convolutional opeartion in tensorflow, we can use tf.nn.conv2d()
Understand the Shape of Tensor Returned by tf.nn.conv2d()
If you plan to implement convolutional opeartion on two vectors, you can read:
Understand Vector Convolution: A Beginner Guide