BMI

Memristive Spiking Neural Networks for Artificial Intelligence and Brain-Machine-Interfaces

The present project is aimed at the implementation of a Brain-Machine-interface (BMI) by integrating memristive Spiking Neural Networks (SNN) with an intact vertebrate animal model, the zebrafish larva, by means of an optogenetic neural link. The SNN will have learning capabilities, hence the BMI may endow the larva with novel cognitive skills that are not naturally present. Conversely, the BMI may also allow to probe and search for plasticity in neural system of the animal model. The BMI will require to implement SNN able to

perform the neurocomputations and implement learning with a speed that cannot be reached by software running on conventional computer hardware, such as FPGAs. We shall leverage the recent development of a novel compact spiking neuron model that is based on a new type of memristors to implement SNNs directly on hardware. The concept of memristive devices are similar to that of Mott neuristors, but the key breakthrough was to achieve the same resistive commutation properties by means of a conventional electronic

device, the thyristor. This allows for simplicity in the circuit design and functional reliability. The SNNs will incorporate learning capabilities by means of electronic synaptic circuits that implement the spike-time-dependent-plasticity (STDP) rule. The project is structured along three Work Packages. The first one is to extend a mathematical theory of Rectified Linear Unit (ReLU) neuron networks, which are firing-rate coded, to SNN, which operate with explicit spikes. We shall implement sequences and other dynamical neural attractor

systems. The second WP will be devoted to implement learning with STDP. We shall aim to extend the notion of associative learning to the learning of spatio-temporal sequences. The third WP is the main one, where we implement the BMI using the experimental setup of the Sumbre Lab at ENS. The BMI will allow to explore the behavior of the neural system of larva in novel ways.