Overview / Usage

We are trying to generalize LSTMs and GRUs with an automated approach that constructs the update equations recursively as the recurrent network trains. This combines elements of recursive and recurrent neural networks.

Methodology / Approach

We are using PyTorch to construct our model and training on Intel CPUs. Our code recursively searches for the optimal update equations during training of the RNN.

Technologies Used

PyTorch with Python 3
Numpy
Intel AI DevCloud

Repository

https://github.com/matthewkennedy5/RRNN