Back-propagation through time¶

This is used in recurrent neural newtorks that have a recurrent relationship between the hidden state.

The basic idea is to unroll the computational graph and perform regular back-propagation.

Exponential distribution Comparing gaussian processes to other models