Spiking-hybrid-plasticity-neural-network

• Our code will be coming soon.
• Hybrid plasticity (HP) models provides a generic framework for training global-local hybrid SNNs using pytorch.
• HP model is designed to support multiple spike coding methods (rate-based and temporal based), multiple neuron models, and learning rules (Hebbian-based, STDP-based etc.)
• Note: If the environment configurations are different, the results may fail to work properly. You may need to modify the package version or adjust code details according to your needs.

Setup

All the codes of this project have been debugged and passed on Python 3.5.4 and Pycharm platforms.

• Installation link for Pycharm: https://www.jetbrains.com/pycharm/download/
• Choose your operating system and Python version 3.5
• Download and install

Requirements

Linux: Ubuntu 16.04

Cuda 9.0 & cudnn6.0

NVIDIA Titan Xp and NVIDIA GTX 1080.

torch 1.2.0

torchvision 0.2.2

numpy 1.17.2

scipy 1.2.1

scikit-image 0.15.0

Instructions for use

• File names starting with ‘main_*’ can be run to reproduce the results in this paper.

An example demo

We provide a simple example code to help you quickly run our model and compare with other single-learning models.

How to run: please load the folder of "simple-example" and run the "main_*" functions.

Expected run time: 30s for one epoch (GTX 1080, one core).

Expected results are shown in the readme file in the ‘simple-example’ folder.

Reference

1. Wu Y, Zhao R, Zhu J, et al. Brain-inspired global-local hybrid learning towards human-like intelligence[J]. arXiv preprint arXiv:2006.03226, 2020.