CogniGron Seminar: Susan Stepney (University of York, UK) - "Unconventional design of unconventional computers"

There are many proposed unconventional computational models: reservoir computing, general purpose analogue computing, membrane computing, quantum computing, morphogenetic computing, and more. There are just as many proposed unconventional computing substrates: slime moulds, carbon nanotubes, gene engineered bacteria, gold nanoparticle networks, memristors, optical relays, and more. But how to match model and substrate to get an effective unconventional computer?

We have developed CHARC, a framework to characterise how well some material substrate can instantiate and implement a given computational model.  We have applied this to a reservoir computing model instantiated in carbon nanotubes, and also in simulations. This works by comparing candidate instantiations with a baseline system, and using novelty search over computational behaviour space to quantify the achievable scope of the instantiation.

I will introduce this framework, and how we are developing it further: to characterise how faithfully some abstract model represents the computing properties of a given substrate; and to co-design unconventional models and substrates that are natural computational partners.