Official implementation of the warmup initialisation procedure for RNNs.
This repository contains the datasets, recurrent cells and experiments presented the paper.
See python supervised.py -h
and python reinforcement.py -h
for usage details.
If you find it useful, please reference in your paper:
@article{lambrechts2023warming,
title={Warming up recurrent neural networks to maximise reachable multistability greatly improves learning},
author={Lambrechts, Gaspard and De Geeter, Florent and Vecoven, Nicolas and Ernst, Damien and Drion, Guillaume},
journal={Neural Networks},
year={2023},
publisher={Elsevier}
}
To learn more: