Code associated with Spiking Neural Networks: Considerations, Implementations, and Comparison with Convolutional Neural Networks. Jackson Borchardt, Samantha Coury, Stephanie Crater, Ashley Qin. VS 265, Fall 2024
See Neuron_Model_Comparison.ipynb for a comparison of a variety of neuron models: standard Leaky Integrate-and-Fire (LIF) neuron, Hodgkin-Huxley neuron, various Izhikevich neurons (regular spiking, intrinsically bursting, chattering, or fast spiking), and our novel Funky neuron (equivalent to an LIF neuron with threshold that randomly changes when a spike is fired).
See LIF_vs_Funky_snn.ipynb for a comparison of spiking neural networks comprised of snntorch's Leaky (LIF) neuron vs. our custom Funky neuron. Note that this notebook requires a custom version of snnTorch with our Funky neuron added, available at [https://github.com/calderast/snntorch]. The pull request detailing the exact functionality added to snnTorch is here.
See cnn_mnist.py for a implementation of a simple feedforward convolutional neural network (CNN) trained such that it achieved roughly equivalent (~94%) test set accuracy to our SNNs.
Much of our code is adapted from snntorch. Relevant snntorch tutorials are included in the snntorch_tutorials/ folder.