hjq310/neuropack

NeuroPack, a tool for emulating memristor-based neuromorphic architectures

NeuroPack

Paper available here: NeuroPack: An Algorithm-level Python-based Simulator for Memristor-empowered Neuro-inspired Computing

experimentalResults.npz gives experimental results shown in the paper.

NeuroPack is an algorithm-level simulator for memrsitor-based neuromorphic designs. It incorporates an empirical memristor model here to simulate how memristor states change during an online-training process. To get memristor fitting parameters, please use ArC One instrument module 'ParameterFit'.

---

Components

• NeuroPack.py implements the main panel and the built-in analysis tool for NeuroPack
• NeuroCores/core_X.py implements the neuron model and the learning rule

---

Input files

• Configuration file: NeuroData/XX.json is an input file that users need to provide for setting configuration
• Matrix Connectivity file: NeuroData/XX.txt is an input file that defines how neurons are connected and which memristor is mapped to the connectivity.
• Stimuli file for training: NeuroData/XX.txt is an input file that provides input spikes and output labels (for supervised learning only) for training.
• Stimuli file for testing (optional): NeuroData/XX.txt is an input file that provide input spikes and output labels (for supervised learning only) for test.

---

Install and run

• Please install the updated Arc One interface with all required packages.
• Download NeuroPack and put it in the following directory: C:/Users/AppData/Roaming/arc1pyqt\ProgPanels
• If you use Windows, open command prompt and go to the path where you install arc1_pyqt.
• Use the following commands to run Arc One interface:

SET NNDBG=1 if you want to print out debug information in command prompt.

python setup.py build

python run.py

• Now ArC One interface pops up. Select 'VirtualArC' from the portal dropdown list, and click 'connect'.
• Click 'Read All'. Now the state indicator changes to 'Ready'.
• Select 'NeuroPack' from module dropdown list, and click 'Add'.
• Now NeuroPack main panel should pop up. Add all necessary input files and click 'Training network' to run.

---

Example usage

If you want to reproduce the results showcased in the paper 'NeuroPack: An Algorithm-level Python-based Simulator for Memristor-empowered Neuro-inspired Computing', please do the following:

• Click 'NeuroBase.json' and select 'MNIST_LIF.json'.
• Click 'Load Conn. Matrix' and select 'MNIST_LIF_connmat.txt'.
• Click 'Load Stim. File' and select 'MNIST_LIF_stim.txt'.
• Tick 'Load test file', click and select 'MNIST_LIF_test_stim.txt'.
• Tick 'Save to', click 'Load test file', and create a file to store inference results.
• Select 'LIF_supervisedLearning_wta' for 'Network core'.
• Press 'Train Network'

---

Citation

@ARTICLE{NeuroPack,
AUTHOR={Huang, Jinqi and Stathopoulos, Spyros and Serb, Alexantrou and Prodromakis, Themis},   
TITLE={NeuroPack: An Algorithm-Level Python-Based Simulator for Memristor-Empowered Neuro-Inspired Computing},      	
JOURNAL={Frontiers in Nanotechnology},      
VOLUME={4},      	
YEAR={2022},       
URL={https://www.frontiersin.org/article/10.3389/fnano.2022.851856},       	
DOI={10.3389/fnano.2022.851856},      	
ISSN={2673-3013}
}

neuropack's Issues

Datasets

Hey, How can I train datasets like paper and where I can get it.?

Thank you.

Exception encountered while importing module cannot import name 'pulse' from 'arc1pyqt.VirtualArC'