Electronic neural network for modelling the Pavlovian conditioning

Spiking neural networks are designed for better modeling the natural neural tissue physiology in order to increase the biological plausibility of the artificial neural structures. In this paper we will present a simple structure of electronic spiking neurons that is able to model the classical conditioning starting from the Pavlov observations of the dog's central nervous system. For modeling the conditioned reflex formation and extinction the artificial neural network uses the associative learning mechanisms implemented by the electronic synapses. The results show that using just a few artificial neurons implemented in analogue hardware the network is able to build new neural paths between areas of electronic neurons when the trained neural paths are activated concurrently with untrained ones modeling in this way the reflex formation. Thus, after the learning phase the input neural areas that initially were not able to activate the output neural areas gain this ability due to simultaneous activation with the trained neural paths. On the other hand by using a couple of inhibitory neurons the neural network learns to inhibit the formed reflex. Using inhibition to reduce the output activity of the neural network represents a new approach in modeling the conditional reflex extinction. Also, from our knowledge this represents the neural structure with the lowest number of neurons that is able to model the principles of Pavlovian conditioning. The significant reduction of the number of neurons was possible because the analogue neurons implement intrinsically high complexity functions.