Unconventional Computing manages to make the weirdest materials compute to a greater or lesser degree: slime moulds, chemicals, black holes, gold nanoparticles, swarms of crabs, carbon nanotubes. But if I’m given a blob of goo, can I work out how well it can compute?
That’s what our latest paper, just up on the arXiv, sets out to do: it provides a framework for evaluating how well some arbitrary substrate can be configured to be a Reservoir Computer:
Matthew Dale, Julian F. Miller, Susan Stepney, Martin A. Trefzer
A Substrate-Independent Framework to Characterise Reservoir Computers
arXiv:1810.07135 [cs.ET]
The Reservoir Computing (RC) framework states that any non-linear, input-driven dynamical system (the reservoir) exhibiting properties such as a fading memory and input separability can be trained to perform computational tasks. This broad inclusion of systems has led to many new physical substrates for RC. Properties essential for reservoirs to compute are tuned through reconfiguration of the substrate, such as change in virtual topology or physical morphology. As a result, each substrate possesses a unique “quality” – obtained through reconfiguration – to realise different reservoirs for different tasks.
Here we describe an experimental framework that can be used to characterise the quality of any substrate for RC. Our framework reveals that a definition of quality is not only useful to compare substrates, but can also help map the non-trivial relationship between properties and task performance. And through quality, we may even be able to predict the performance of similarly behaved substrates. Applying the framework, we can explain why a previously investigated carbon nanotube/polymer composite performs modestly on tasks, due to a poor quality. In the wider context, the framework offers a greater understanding to what makes a dynamical system compute, helping improve the design of future substrates for RC.
No comments:
Post a Comment