Role of unconditioned stimulus prediction in the operant learning: a Neural networks model.

S.E. LEW, C. WEDEMEYER, B.S. ZANUTTO

PROCEEDINGS OF THE IEEE

IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC

Lugar: Estados Unidos; Año: 2001 p. 331 - 336

0018-9219

A neural network model of operant conditioning for appetitive and aversive stimuli is proposed. From neurobiological and behavioural bases it is assumed that animals are able to compute the prediction of the unconditioned stimulus. The prediction controls the learning of the correct response to obtain reward and to avoid punishment. The model has as inputs: all the conditioned stimuli and the unconditioned stimulus. The outputs are all the possible responses of the animal; each one is computed by one neuron. Based on Hebbian or anti-Hebbian learning, depending on the prediction, the synaptic weights of the response neurons are calculated. The synaptic weights of the neuron computing the prediction are calculated based on the Rescorla-Wagner model. The simulated and experimental data have been compared, showing that the model predicts relevant features of operant conditioning. This model is a theory of operant conditioning and provides principles to design autonomous systems