Rethinking Deep Thinking: Stable Learning of Algorithms using Lipschitz Constraints
Jay Bear, Adam Prügel-Bennett, and Jonathon Hare
The University of Southampton
This is a repository containing the experiment code for Rethinking Deep Thinking: Stable Learning of Algorithms using Lipschitz Constraints.
The method of running experiments may seem unorthodox. This is due to Bear writing the vast majority of the code despite not being a Python programmer. Since the writing of this code, Bear is now much more familiar with Python programming practices, however this code remains mostly unchanged from the version used for the initial submission of the paper.
Training models involves describing the model and its training routine in
JavaScript Object Notation (JSON) files. Some JSON config files are provided
in the configs directory.
Inside of the configs directory is another README.md explaining each
part of the training config file, including acceptable values for each
item.