Section: New Results

Coding by spikes

Our goal here is a better understanding of the extent to which computing and modeling with spiking neuron networks might be biologically plausible and computationally efficient. Based on a thorough characterization of the main constraints on spiking neural networks dynamics this has led us to propose new algorithms to infer the structure of the network from its spike trains and to propose an FPGA implementation of spiking neural networks.

Reverse-engineering of spiking neural networks parameters

Participants : Bruno Cessac [correspondent] , Horacio Rostro-Gonzalez, Thierry Viéville [Cortex] .

We consider the deterministic evolution of a time-discretized spiking network of neurons with connection weights having delays, modeled as a discretized neural network of the generalized integrate and fire (gIF) type. The purpose is to study a class of algorithmic methods allowing to calculate the proper parameters (synaptic weights) to reproduce exactly a given spike train generated by an hidden (unknown) neural network. This problem is linear (L) if the membrane potentials are observed and LP (Linear-Programming) if only spike times are observed, in the context of gIF models. The L or LP adjustment mechanism is local to each unit and has the same structure as an "Hebbian" rule. This paradigm is easily generalizable to the design of input-output spike train transformations. This means that we have a practical method to "program" a spiking network, i.e. find a set of parameters allowing us to exactly reproduce the network output, given an input

This work has been submitted in the Journal of Neural Engineering, 2011[25] .

Development of FPGA-based efficient reconfigurable architectures for spiking neural networks

Participants : Bruno Cessac, Bernard Girau [INRIA Cortex] , Horacio Rostro-Gonzalez, Cesar Torres-Huitzil [Information Technology Department, Polytechnic University of Victoria (UPV), Tamaulipas, Mexico] , Thierry Viéville [Cortex, correspondent] .

Spiking neural networks are able to perform very powerful computations with precise timed spikes. We are developing an FPGA (Field Programmable Gate Array) reconfigurable platform that enables the simulation of in silico models of spiking neural networks. Since the model is directly mapped into a FPGA device, the neural processing is accelerated and the time consumption reduced. We use VHDL and Handel-C to design the reconfigurable architecture of a discrete time Integrate-and-Fire model coded in CUDA, running on GPU.

This work has been accepted in Journal of Physiology, Paris [24] .

Towards biologically inspired image coders

Participants : Marc Antonini [Laboratoire I3S, Sophia Antipolis, France] , Pierre Kornprobst, Khaled Masmoudi [Laboratoire I3S, Sophia Antipolis, France] .

In [51] we presented a novel bio-inspired and dynamic coding scheme for static images. Our coder aims at reproducing the main steps of the visual stimulus processing in the mammalians retina taking into account its time behavior. The main novelty of this work is to show how to exploit the time behavior of the retina cells to ensure, in a simple way, scalability and bit allocation. To do so, our main source of inspiration has been the biologically plausible retina model Virtual Retina described in Section  5.1 . Following a similar structure, our model has two stages. The first stage is an image transform which is performed by the outer layers in the retina. Here it is modelled by filtering the image with a bank of difference of Gaussians with time-delays. The second stage is a time-dependent analog-to-digital conversion which is performed by the inner layers in the retina. Thanks to its conception, our coder enables scalability and bit allocation across time. Also, compared to the JPEG standards, our decoded images do not show annoying artefacts such as ringing and block effects. As a whole, this article shows how to capture the main properties of a biological system, here the retina, in order to the design a new efficient coder.