Extended Category Learning with Spiking Nets and Spike Timing
Dependent Plasticity
by Chris Huyck and Carlos Samey
- I've been doing a lot of work with simulated biological neurons
and Cell Assemblies for about 30 years now.
- A couple of years ago, it occurred to me that a really simple way to do
machine learning was to use actual biological models; the world, and my
taught programme thesis students love machine learning.
- So, last year, I whacked together some stuff using
NEST, PyNN, and a simple feedforward topology, learning with
spike timing dependent plastiticy. I gave a poster at SGAI 2020 on
this.
- I thought this might actually be a way to learn about the work in
this area, get some projects for students, and maybe get to something
really novel. So, I put together a page
https://www.cwa.mdx.ac.uk/spikeLearn/spikeLearn.html, found
an MSc thesis student (Carlos), and did a bit more work on my own.
- That's what this paper is.
- We've stuck with reasonably biologically plausible neuron models,
and models of learning (STDP on synapses).
- We've done some work on document categorisation with this
kind of topology.
- We've extended it with extra layers to use combinations of
features explicitly.
