Protein Structured Reservoir Computing for Spike-Based Pattern Recognition

Nowadays we witness a miniaturisation trend in the semiconductor industry backed up by groundbreaking discoveries and designs nanoscale characterisation fabrication. To facilitate produce ever smaller, faster cheaper computing devices, size of nanoelectronic devices is now reaching scale atoms or molecules - technical goal undoubtedly demanding for novel devices. Following trend, explore an unconventional route implementing reservoir on single protein molecule introduce neuromorphic connectivity with small-world networking property. We have chosen Izhikevich spiking neurons as elementary processors, corresponding to verotoxin protein, its 'hardware' architecture communication networks connecting processors. apply readout layer various training methods supervised fashion investigate whether molecular structured Reservoir Computing (RC) system capable deal machine learning benchmarks. start Remote Supervised Method, based Spike-Timing-Dependent-Plasticity, carry linear regression scaled conjugate gradient back-propagation methods. The RC network evaluated proof-of-concept handwritten digit images from MNIST dataset demonstrates acceptable classification accuracy comparison other similar approaches.