We have shown that the Reservoir Computing framework transfers to complex substrates, and that performance can increase significantly when we control and manipulate input-output mappings and external perturbation through computer-controlled evolution. We have implemented a new example of the hardware-based reservoir methodology. We have two new types of reservoirs based on Light Emitting Diodes (LEDs) and resistors. Results show that unconstrained computer-controlled evolution can exploit the net effect of variations in components (resistors and diodes) to form a single reservoir competitive to previous findings.
full abstract : pdf
@inproceedings(Dale2017:UCNC:leds, author = "Matthew Dale and Julian F. Miller and Susan Stepney and Martin A. Trefzer", title = "Reservoir Computing \emph {in materio} with LEDs", crossref = "UCNC-2017-p" ) @proceedings(UCNC-2017-p, title = "Poster abstracts, UCNC 2017, Fayetteville, Arkansas, USA", booktitle = "Poster abstracts, UCNC 2017, Fayetteville, Arkansas, USA", year = 2017 )