Code for "Modelling the influence of data structure on learning in neural networks"

Here we provide the code used to run all the experiments of our recent paper on the hidden manifold model [1]. There are two parts:

1. A simulator for training two-layer neural networks trained on the hidden manifold using online learning. It is written in C++ and uses the Armadillo library for linear algebra.
2. A set of python scripts to run the memorisation experiments presented in Sec. V. They are implemented in Python using the pyTorch library.

Compilation of the C++ code

To compile locally, simply type

make hmm_online.exe hmm_ode.exe

This assumes that you have installed the Armadillo library on your machine.

Usage

To see the options of each program, use the -h flag.

To run the unit tests of the Python code, go the memorisation directory and simply type

References

[1] S. Goldt, M. Mézard, F. Krzakala and L. Zdeborová: Modelling the influence of data structure on learning in neural networks arXiv:1909.11500