A Pytorch-based library for simulation of rate-encoded deep spiking neural networks. This library mostly implements the ANN to SNN conversion method described in Conversion of Continuous-Valued Deep Networks to Efficient Event-Driven Networks for Image Classification. It also supports a hybrid SNN-ANN simulation in which the initial layers are spiking and the latter layers are non-spiking.
If using for an academic publication, please consider citing "NEBULA: A Neuromorphic Spin-Based Ultra-Low Power Architecture for SNNs and ANNs", in International Symposium on Computer Architecture, 2020.
@inproceedings{10.1109/ISCA45697.2020.00039,
author = {Singh, Sonali and Sarma, Anup and Jao, Nicholas and Pattnaik, Ashutosh and Lu, Sen and Yang, Kezhou and Sengupta, Abhronil and Narayanan, Vijaykrishnan and Das, Chita R.},
title = {NEBULA: A Neuromorphic Spin-Based Ultra-Low Power Architecture for SNNs and ANNs},
year = {2020},
isbn = {9781728146614},
publisher = {IEEE Press},
url = {https://doi.org/10.1109/ISCA45697.2020.00039},
doi = {10.1109/ISCA45697.2020.00039},
}
- Validating a pre-trained pytorch model (ANN).
- Converting ANN to SNN and simulating it.
- Plotting graphs showing the correlation between SNN and ANN models.
- Simulating hybrid SNN-ANN models.
Python 3.6 is needed along with other packages mentioned in requirements.txt. CUDA acceleration is supported but not required to run the code. To install the required packages run the following command:
pip install -r requirements.txt
In order to run the code, enter the following command from the terminal:
python main.py --config-file configs/lenet5.ini
The configs folder contains .ini files in which appropriate flags need to be set to achieve the desired task. Location of the-pretrained networks along with simulation parameter details are also specified here. Please refer to configs/tutorial.ini for a description of all the flags. Some of the pre-trained models used in this implementation can be found here.