Towards Temporal Information Processing -- Printed Neuromorphic Circuits with Learnable Filters

This github repository is for the paper at NanoArch'23 - Towards Temporal Information Processing -- Printed Neuromorphic Circuits with Learnable Filters

cite as

Towards Temporal Information Processing -- Printed Neuromorphic Circuits with Learnable Filters
Zhao, H.; Pal, P.; Hefenbrock, M.; Beigl, M.; Tahoori, M.
Proceedings of the 18th ACM International Symposium on Nanoscale Architectures. 2023.

Usage of the code:

1. Training of printed Temporal Processing Neuromorphic Circuit (pTPNC)

$ sh run_LearnableFilter.sh

Alternatively, the experiments can be conducted by running command lines in run_LearnableFilter.sh separately, e.g.,

$ sbatch exp_LearnableFilters.py --DATASET 0 --SEED 0 --task temporal --loss celoss --metric temporal_acc --projectname LearnableFilters
$ sbatch exp_LearnableFilters.py --DATASET 0 --SEED 1 --task temporal --loss celoss --metric temporal_acc --projectname LearnableFilters
...

Additionally, the baselines, e.g., the previsous printed Neuromorphic Circuits (pNCs) can be experimented by running run_baseline_pNN.sh.

2. After training printed neural networks, the trained networks are in ./LearnableFilters/model/, the log files for training can be found in ./LearnableFilters/log/. If there is still files in ./LearnableFilters/temp/, you should run the corresponding command line to train the networks further. Note that, each training is limited to 48 hours, you can change this time limitation in configuration.py

Similarly, the baselines can be found in the folder ./Baseline/.

3. Evaluation can be done by running the Evaluation.ipynb in the corresponding folders for pTPNC or baselines.