Our
new paper published today:
Matthew Dale, Julian F. Miller, Susan Stepney, Martin A. Trefzer.
A substrate-independent framework to characterize reservoir computers.
Proceedings of the Royal Society A, 475(2226), 2019
Somewhat amazingly, various blobs of “goo” can be made to compute simple tasks. But, given a new blob of goo, can we tell how well it will compute, without having to train it on specific tasks?
That’s what we set out to address in our new paper. We describe a framework for evaluating the “quality” of a proposed computing substrate, in comparison to a “reference” reservoir computer (an unconventional model of computing that fits well with gooey substrates).
We then use of the framework to evaluate a (physical) carbon nanotube system (it computes, as we knew, but not very much, as we also knew, but now we know exactly how much). We also use it to evaluate a (simulated) optical delay line, and show that it can be used for many reservoir tasks, but not necessarily all.
We are now going on to generalise this framework to a wider set of computational models and physical substrates, as part of out
EPSRC-funded SpInspired project. Watch this space!