Competitive and Temporal Inhibition Structures with Spiking Neurons E. ROS, F. J. PELAYO, P. MARTIN-SMITH, I. ROJAS, D. PALOMAR and A. PRIETO Department of Architecture and Computer Technology, University of Granada, Spain. E-mail: eros@atc.ugr.es, aprieto@goliat.ugr.es Abstract. The paper describes the implementation of competitive neural structures based on a spiking neural model that includes multiplicative or shunting synapses enabling non-saturated stable states in response to different stationary inputs as well as controllable transient responses. A VLSI-viable implementation of this model has been previously proposed and tested [1]. It has the possibility of modulating the output spike frequency by an additional input without affect- ing other neuron variables such as the membrane potential. This feature is exploited in the simu- lation of a Selective Temporal Inhibition network that is suitable for implementing attentional control systems. Key words: spiking neurons, competitive processing, temporal inhibition, attentional control mechanisms, bio-inspired neural systems. 1. Introduction In the approach presented here, the main motivation for using spiking neurons arises from the implementation issues. An inter-neuron communication scheme based on instantaneous events that can be multiplexed in shared connection channels enables to take advantage of inter-chip communication schemes [2^6] that make multiple-chip architectures viable. Furthermore, the incoming pulses can be directly integrated at the target synapses, thus enabling single synapses to process spikes from several source neurons, therefore reducing the number of synapse modules required to implement convergent synaptic trees, like the many-to-many connections required in on-centre^off-surround topologies. A competitive structure based on the cell model described in Section 2 is simulated. In some perception processing schemes like saliency maps [7], competition between the different features extracted makes it possible to perform selective attentional control that can drastically reduce the reaction times in complex environments. An attentional shifting mechanism, based on Selective Temporal Inhibition (STI) structures, is also simulated in Section 3. This attentional mechanism takes advan- tage of the representation provided by a competitive structure in which different intermediate states of active nodes are preserved and taken into account. A com- petitive layer made up of the proposed STI elements selectively inhibits the ¢ring nodes in response to a global reset signal giving a chance to other less active nodes. Neural Processing Letters 11: 197^208, 2000. 197 # 2000 Kluwer Academic Publishers. Printed in the Netherlands.