Code for Van der Plas, Vogels* and Manohar*, 2022, Proceedings Machine Learning Research

Presented at COLLAS, August 2022, conference.

https://proceedings.mlr.press/v199/plas22a.html

We welcome you to reuse this code, but please cite our paper if you do! Thank you!

See paper for explanation, this is the code that created all figures, including the saved trained networks.

• All packages are stated in py37.yml (use anaconda to create a new environment from this file)
• Train or load a single network.ipynb is an example notebook of how to train RNNs on 1 or multiple tasks.
• Figure generation notebook.ipynb creates all figures of the paper

Please note: because all networks are saved, the repository is quite large (approximately 750MB). Alternatively, you can download everything except the models/ folder (and the .git/ folder) to exclude pre-trained networks, which saves 740MB.