Choreographing neural networks (2011-2013)
The role of myself and my colleagues at ISS in the CORONET project was to develop neuromorphic hardware which implements a "mesoscopic" model of brain dynamics. By mesoscopic we meant somewhere above the microscopic models of neurons and synapses that neuromorphic engineers normally work with. The units we were modelling represented whole populations of neurons, in particular capturing their "bistable attractor dynamics"; this means the way a group of closely interconnected neurons may tend to be either all virtually inactive or else all in a state of high activity, and how they tend to switch between these states probabilistically, sometimes with very low frequencies.
These dynamics arise from the population of neurons as a whole rather than from the properties of individual neurons. When we then couple many of these units together in a network, the resulting behaviour can be very rich, and is a promising candidate for explaining certain observations from neuropsychology. The form of the model I was working with had the working name "Gelatinous-State Network", in homage to the "Liquid-State Network".
Papers to which I made an acknowledged contribution
- "Robust working memory in an asynchronously spiking neural network realized with neuromorphic VLSI" Giulioni M, Camilleri P, Mattia M, Dante V, Braun J, Del Giudice P, Frontiers in Neuromorphic Engineering, 2011. pdf