Biological learning in key-value memory networks
All figures from the paper can be recreated by running notebooks/plot_all_figs.ipynb.
Some figures require pre-trained models (see directions on training and analysis below).
Below are the figures that require pre-trained models and the names of the
corresponding experiments that need to be run:
| Figure | Experiments to Run |
|---|---|
| 2b | tvt |
| 3a | train_random_capacity |
| 3c | train_prepost_zero_init |
| 4a | train_continual |
| 4c | train_corr |
| 5a | heteroassociative* |
| 5b | seqrecall* |
| 5c | copy* |
The functions found in the experiments.py file correspond to the name of
each experiment. Experiments can be run with:
python main.py --train {experiment name}
Generic analysis of training progress (or specialized analysis for some experiments)
can be done by running the following after training:
python main.py --analyze {experiment name}