BRAIN-Net: Spiking Neural Networks for Real-Time Processing of Brain Signals
Large-scale neural recordings using high-density electrode arrays are key to understanding brain dynamics and designing brain-computer interfaces for rehabilitation. These devices produce large data flows that raise new challenges to extract relevant information in real time with limited power consumption to be embedded into autonomous implantable devices. Toward this end, BrainNet proposes to use artificial Spiking Neural Networks (SNNs). Different SNN architectures based on unsupervised, supervised and hybrid learning will be developed and applied to automatic preprocessing of cortical spiking activity (unsupervised spike sorting), extraction of cortical activity patterns, and decoding of these patterns in term of behavior (vocal and speech production). These SNN architectures will then be implemented in FPGA and neuromorphic integrated circuits to run in real-time paradigms. An ethical reflexion will acompany these developments. BrainNet should thus bring new routes toward intelligent brain implants for neural prostheses.