Spikelearn

Implementation of spiking neural networks capable of online learning tailored for machine learning workflows and neuromorphic computing applications.

Motivation

We needed a SNN model with the following requirements:

• Capable of handling traditional ML workflows
• Heterogeneous, with the ability to integrate both mathematical models and neurons or synapses inspired on neuromorphic computing and emergent devices
• That could be easily parametrizable, in order to explore a large number of configurations in high performance computing environments.
• That could reproduce models in existing neuromorphic chips such as Loihi.
• That could handle neuromodulators and other neuroscience-inspired goodies.
• That could be easily extensible.
• That is capable of online learning through a variety of synaptic plasticity rules.

Spikelearn intends to fill that role.

Status

Spikelearn is still in development. Please check spikelearn's documentation in readthedocs.

Quick install

Through pypi:

pip install spikelearn

Usage

from spikelearn import SpikingNet, SpikingLayer, StaticSynapse
import numpy as np

snn = SpikingNet()
sl = SpikingLayer(10, 4)
syn = StaticSynapse(10, 10, np.random.random((10,10)))

snn.add_input("input1")
snn.add_layer(sl, "l1")
snn.add_synapse("l1", syn, "input1")
snn.add_output("l1")

u = 2*np.random.random(10)
for i in range(10):
    s = snn(2*np.random.random(10))
    print(s)

Acknowledgements

• Threadwork, U.S. Department of Energy Office of Science, Microelectronics Program.

Copyright and license

Copyright © 2022, UChicago Argonne, LLC

Spikelearn is distributed under the terms of BSD License. See LICENSE

Argonne Patent & Intellectual Property File Number: SF-22-154