Application of local learning and biological activation functions to networks of neurons for motor control

Models of networks of neurons involved in motor control have been largely based on concepts derived for artificial neural networks such as global learning and idealized activation functions. The neurons in these models frequently fail to incorporate measured spike rates and baseline, background firing and thus the neuronal outputs may be less useful for testing and developing analysis techniques that can eventually be used on experimental data. In this paper we present an approach for creating large-scale networks of neurons that include local learning and more biological features of neuronal spiking and demonstrate that the models are able to learn a generalized two-dimensional reaching task. This approach opens the possibility for the development of more biologically realistic network models with an increased capacity for adaptation, with a possible tradeoff of reduced learning rates.