Section: New Results

Embodied and embedded systems

Participants : Yann Boniface, Hervé Frezza-Buet, Bernard Girau, Mathieu Lefort, Dominique Martinez, Jean-Charles Quinton, Nicolas Rougier.

InterCell

Our research in the field of dedicated architectures and connectionist parallelism mostly focuses on embedded systems (cf. § 3.5 ). Nevertheless we are also involved in a project that considers coarse-grain parallel machines as implementation devices. The core idea of this InterCell project (cf. http://intercell.metz.supelec.fr ) is to map fine grain computation (cells) to the actual structure of PC clusters. The latter rather fit coarse grain processing, using relatively few packed communication, which a priori contradicts neural computing. Another fundamental feature of the InterCell project is to promote interaction between the parallel process and the external world. Both features, cellular computing and interaction, allow to consider the use of neural architectures on the cluster on-line, for the control of situated systems, as robots.

Embodied/embedded olfactory systems

How can animals successfully locate odour sources?

Our goal is to investigate this question. Two different classes of strategies are possible for olfactory searches: those based on a spatial map, e.g. Infotaxis, and those where the casting-and-zigzagging behaviour observed in insects is purely reactive. We have implemented Infotaxis in a robot and shown that it produces trajectories that feature zigzagging and casting behaviours similar to those of moths. This result however should not be interpreted as evidence that the corresponding moth behaviour is driven by Infotaxis. Whether or not moths use infotactic or reactive strategies is still unclear. To compare both strategies, we have developped a cyborg using the antennae of a tethered moth as sensors (no artificial sensor for pheromone molecules is presently known). Experiments are in progress to compare the trajectories of the cyborg controlled by infotactic and reactive search strategies to those obtained with the same cyborg but driven by the moth's brain.

How can technology emulate biological olfactory processing?

Glomerular microcircuits in the first stage of the olfactory pathway reformat odor representation. First, many ORNs expressing the same receptor protein, yet presenting heterogeneous dose-response properties, converge onto each glomerulus [10] . Second, onset latency of glomerular activation is believed to play a role in encoding odor quality and quantity in the context of fast information processing [6] . Taking inspiration from biology, we designed a simple yet robust glomerular latency coding scheme for processing gas sensor data [7] . The proposed bio-inspired approach was evaluated using an Sn02 sensor array. Glomerular convergence was achieved by noting the possible analogy between receptor protein expressed in ORNs and metal catalyst used across the fabricated gas sensor array. Ion implantation was another technique used to account both for sensor heterogeneity and enhanced sensitivity. The response of the gas sensor array was mapped into glomerular latency patterns, whose rank order is concentration-invariant.

Hardware implementations of neural models

In the field of dedicated embeddable neural implementations, we use our expertise in both neural networks and FPGAs so as to propose efficient implementations of applied neural networks on FPGAs, as well as to define hardware-friendly neural models.

• Following our results on the design of spiking models back-engineered from spike recordings, recent works have focused on the analysis of the influence of precision onto asymptotic dynamics of FPGA-embedded integrate-and-fire neural models [13] .

• We design hardware-friendly adaptations of dynamic neural fields that use spiking neurons. In this field, we have derived a highly simplified version of such spiking neural fields, and we have experimentally shown that the main properties of standard neural fields are maintained in the context of visual attention [29] .

• We currently intend to minimize the topological constraints of FPGA-embedded spiking neural fields using reduced neighborhoods but randomly propagating spikes. A preliminary result has been obtained so as to implement massively distributed pseudo-random number generators based on cellular automata that use minimal areas [21] .

Towards brain-inspired hardware

Our activities on dedicated architectures have strongly evolved in the last years. We now focus on the definition of brain-inspired hardware-adapted frameworks of neural computation. Our current works aim at defining hardware-compatible protocols to assemble various perception-action modalities that are implemented and associated by different bio-inspired neural maps.

Anticipatory mechanisms in neural fields

We have defined first models of neural fields that include anticipatory mechanisms through the integration of spatiotemporal representations into the lateral interactions of a dynamic neural field [23] . This work targets increased robustness and goal-oriented action selection within sensori-motor systems.

Multimodal learning through joint dynamic neural fields

This work relates to the development of a coherent multimodal learning for a system with multiple sensory inputs.

• We have modified the BCM synaptic rule, a local learning rule, to obtain the self organization of our neuronal inputs maps and we use a CNFT based competition to drive the BCM rule. In practice, we introduce a feedback modulation of the learning rule, representing multimodal constraints of the environment [39] .

• We have introduced an unlearning term in the BCM equation to solve the problem of the different temporalities between the raise of the activity within modal maps and the multimodal learning of the organization of the maps [22] .