Code used for the paper 'A Sparse Quantized Hopfield Network for Online-Continual Memory'.
Run with Python 3.7.6 and Pytorch 1.10.0.
All datasets besides the TinyImagenet dataset are downloaded automatically via Pytorch. To download TinyImagenet see https://github.com/tjmoon0104/pytorch-tiny-imagenet?tab=readme-ov-file
To reproduce data from a training run/experiment:
main.py --test argument
To reproduce plots for a training run/experiment:
main.py --plot argument
Here are the arguments for the various experiments used to reproduce tests and plots:
associative memory comparisons: assoc_comp
online-continual tests one hidden layer: OnCont-L1
online-continual tests three hidden layer: OnCont-L3
noisy encoding tests one hidden layer: nsEncode-L1
noisy encoding tests one hidden layer: nsEncode-L3
noisy encoding tests one hidden layer: recog
architecture comparisons: arch compare
For example, to reproduce the plots for the online-continual task with the one hidden layer models, run
main.py --test OnCont-L1
followed by
main.py --plot OnCont-L1