/DifferentialNeuralComputer

A Differential Neural Computer implementation in tensorflow

Primary LanguagePythonMIT LicenseMIT

Differential Neural Computer

A Differential Neural Computer (DNC) implementation. Currently, it just learns to repeat back a series of patterns.

Learn more about DNC's here.

This DNC implementation and visualizations are based on this implementation by Mostafa-Samir.

Using:
Tensorflow

To Run:

$ python DNC.py

During training, any visualizations will be exported to a local directory called: images/

Results

Losses Over 10,000 Iterations:

Memory View At Iteration 0:

Memory View At Iteration 10,000: