Probabilistic Spiking Neural Network Research Internship

References

Papers

Probabilistic Spiking Neural Networks
Expectation Maximization for PSSNs
Variational Bayes Algorithm 4
Variational Auto-Encoders Useful for understanding Partially Observed training methods

Resources

Norse Pytorch SSN Library

Abstract

This repo contains code and results I obtained as part of an internship researching SNNs

PSSN/ Julia PSSN implementation
Norse-Pytorch/ Norse demo for comparison
TestData.xlsx Training results and parameters. Useful for comparing alternative implementations
PSSN_Presentation.pptx A presentations with some visuals describing obtained results and findings

Set Up

Run the PSSN/DataEncoding.jl file to generate the encoded data
Run Norse-Pytorch/norse-3layer.ipynb to train the norse demo, you can select Iris or MNIST dataset
Run PSSN/Main.jl to train the julia model. PSSN/PSSN.jl contains the various funtions and structs to define and train a network

Next Steps

Test PSSN implementation on more diverse training problems
Fully and Partially Observed models can potentially be combined into a single function using EM
Port to FluxML or Pytorch for potential performance gain and possibility to utilize back-propagation with the reparameterization trick