Physical reservoir computing: a tutorial

Susan Stepney. Physical reservoir computing: a tutorial.
Natural Computing, 2024. doi: 10.1007/s11047-024-09997-y

A decade ago (I was going to write "a few years ago", then looked at the date!) my colleagues and I wrote a paper entitled "When does a physical system compute?", which gives a framework to distinguish systems that are computing, from ones that are just doing their thing.

Recently, I was invited to write a tutorial for Natural Computing on Physical Reservoir Computing.  That's about using weird physical materials, like a glob of carbon nanotubes, or a sheet of magnetic material, or whatever, to compute directly, according to the "reservoir computing" model, which is a form of neural network.

I decided to use the framework we previously developed to structure the tutorial.  This framework provides five things you need to consider: (1) the abstract computational model, here, reservoir computing; (2) the physical computing substrate, here, the purported reservoir computer; (3) how to encode abstract inputs and physically inject them into the computer; (4) how to observe physical outputs and decode them to abstract results; (5) and last but certainly not least, how to validate that the physical system is faithfully implementing the abstract model.

If you want to see more of the details, have a look at the paper: it's open access.