Building an embedding layer

I would like to build an embedding layer to implement time to vector embedding per this paper:

What is the right starting point to implement my class? Should I extend org.deeplearning4j.nn.conf.layers.FeedForwardLayer or extend org.deeplearning4j.nn.layers.BaseLayer? What are the methods that I need to override? Thanks for any pointer.

The easiest way to do this these days is to subclass SameDiffLambdaLayer. See the examples here: