sPykeFlow

Introduction

• This is a tool for unsupervised feature extraction with spiking neural networks.
• Neurons are non-leaky integrate and fire. There are methods to enforce lateral inhibition, STDP competition.
• It provides insights like spike activity, feature extraction, animation of synapse evoltuion for each layer etc.
• It also provides feature classification class and a few other jupyter notebooks with misc codes.

Organization

• AllDataSets folder should contain the data that you wish to work with.
• spykeflow folder contains all the important classes of SpykeFlow.
• notebooks contains miscelleneous jupyter notebooks for classification, plots, etc.
• outputs contains all the outputs/plots generated so far with SpykeFlow.
• notebooks/main_emnist.ipynb shows an example to train a spiking convolutional layer with two conv and pool layers with EMNIST dataset and effects of lateral inhibition on spike activity and feature extraction inside the network.
• notebooks/main_facebike.ipynb shows an example to train a spiking convolutional layer with three conv and pool layers with facebikes dataset and effects of lateral inhibition on spike activity and feature extraction inside the network.
• An example notebook to classify the extracted spike features is given in notebooks/classifierclass_usage.ipynb.
• Exhaustive list of requirements in listed in requirements.txt however most important requirements are
• NumPy==1.15.3, SciPy==1.1.0, Progressbar==2.3, Theano==0.8.0, Pandas==u'0.20.3, Pickle==$Revision:72223$, Matplotlib==2.2.3, Keras==2.2.4, Sklearn==0.19.1, OpenCv==3.3.0, Tensorflow==1.14.0.
• Unfortunately, the code is in Python2.7. I will work on porting it to Python3.7 soon. I started working on this tool from Summer 2018 and Python 2.7 was still in the game for many packages that I was experimenting with and I chose to dance with the same girl that I came with.

Publications

• A Deep Unsupervised Feature Learning Spiking Neural Network with Binarized Classification Layers for EMNIST Classification