Dynamics, Plasticity and Learning in a Neural Network Model of the Basal Ganglia

Abstract Organisms are able to learn from reward and punishment to cope with unknown situations, in psychology this phenomena is called reinforcement learning. Experimental studies provide strong evidence that the mechanisms behind reinforcement learning in the mammal brain are located in the basal ganglia, an subcortical area known to be involved in decision making and the planning of movement [squire2013fundamental, purves2004Neuro]. The neurotransmitter dopamine plays a key role in reinforcement learning [schultz1997neural]. The adaption of behavior is believed to be driven by dopaminergic plasticity in the striatum, the input nucleus of the basal ganglia. In order to investigate the neural mechanisms behind reinforcement learning, an existing spiking neural network model of the basal ganglia [jitsev2012learning2] is improved and extended. The mathematical and structural basis of this model is provided by temporal difference learning, a reinforcement learning algorithm known from the field of machine learning [barto1998reinforcement]. Abstract While being an effective learning algorithm, though, temporal difference learning is not motivated from a biological perspective. Understanding how learning from reward and punishment takes place in the mammal brain requires biological constraints. This thesis addresses this point by implementing homeostatic regulation mechanisms and investigating neural circuits which implement value computation and action selection in the striatum. To improve the biological plausibility of the model, the synaptic plasticity rules of the dopaminergic feedback in the model are re-calculated and simplified. Finally, the plausibility of winner-takes-all circuits in the dorsal striatum by strong lateral connection between fast spiking interneurons is discussed. Abstract This work establishes a basis for future research on the origin of abnormal dopaminergic plasticity leading to dysfunctional states of the basal ganglia like those observed in diseases like Parkinson or Tourette syndrome