Learning in a neural network model in real time using real world stimuli

摘要

In this paper we present a model of the auditory system that is trained using real-world stimuli and running in real-time. The system consists of different sound sources, a microphone, an A/D board, a peripheral auditory system implemented in software and a central network of spiking neurons. The synapses formed by peripheral neurons on the central ones are subject to synaptic plasticity. We implemented a learning rule that depends on the precise temporal relation of pre- and post-synaptic action potentials. We demonstrate that this mechanism allows the development of receptive fields combining learning in real-time, learning with few ]stimulus presentations and robust learning I the presence of large imbalances in the probabil- ity of occurrence of individual stimuli.