Spiking Reservoir Computing Architectures for Model-based Epileptic Brain State Recognition.

Area of Science:

• Computational Neuroscience
• Neuromorphic Engineering
• Machine Learning

Background:

• Bio-inspired networks, particularly Recurrent Spiking Neural Networks (RSNNs), leverage brain circuit recurrence for complex spatio-temporal pattern generation with low energy demands.
• Current applications of RSNNs often lack biological relevance, despite their sophisticated capabilities.
• Neuromorphic hardware offers efficient implementation for these dynamic neural network architectures.

Purpose of the Study:

• To present a spiking reservoir computing architecture, a Liquid State Machine (LSM) using Leaky Integrate-and-Fire (LIF) neurons, for recognizing and decoding internal brain states.
• To simulate epileptic activity and develop a pipeline for inferring source state by processing Local Field Potentials (LFPs) with the LSM.
• To enhance LSM performance using biologically inspired synaptic plasticity for task-specific adaptation.

Main Methods:

• Implementation of a fully spiking Liquid State Machine (LSM) with Leaky Integrate-and-Fire (LIF) neurons.
• Simulation of epileptic activity and generation of Local Field Potentials (LFPs).
• Application of population coding for signal encoding and regression on biophysical parameters to infer brain state, enhanced by unsupervised synaptic plasticity.

Main Results:

• The spiking reservoir computing architecture successfully infers the source state by performing regression on biophysical parameters controlling epileptic dynamics.
• A simple, unsupervised synaptic plasticity mechanism optimizes the internal parameters of the reservoir, especially in smaller networks.
• The approach demonstrates hardware efficiency and suitability for edge-device implementations.

Conclusions:

• Spiking reservoir computing offers a promising, energy-efficient strategy for real-time decoding of complex brain dynamics, such as epileptic activity.
• Biologically inspired, unsupervised plasticity mechanisms can optimize general-purpose neuromorphic circuits for specific tasks.
• This technology has significant clinical implications for brain-computer interfaces and neurostimulation therapies in epilepsy management.