Parts | Days | Selection | Search | Updates | Downloads | Help
BP: Fachverband Biologische Physik
BP 27: Modelling of non-linear dynamics in biological movement (focus session, joint BP/DY)
BP 27.5: Talk
Wednesday, April 2, 2014, 15:15–15:30, ZEU 250
COMPUTATIONAL MODEL FOR A FLEXIBLE SENSORIMOTOR MEMORY BASED ON A RECURRENT NEURAL NETWORK — •Kim Joris Boström and Heiko Wagner — Motion Science, University of Münster, Germany
The motor system has the unique capacity to learn complex movements in a flexible manner. Using recent recurrent network architecture based on the reservoir computing approach, we propose a computational model of a flexible sensorimotor memory for the storage of motor commands and sensory feedback into the synaptic weights of a neural network. The stored patterns can be retrieved, modulated, interpolated, and extrapolated by simple static commands. The network is trained in a manner that corresponds to a realistic exercising scenario, with experimentally measured muscular activations and with kinetic data representing proprioceptive feedback. The model may help to explain how complex movement patterns can be learned and then executed in a fluent and flexible manner without the need for detailed attention. Furthermore, it may help to understand the reafference principle in a new way, as an internal feedforward model for the prediction of expected sensory reafference would no longer be necessary. Instead, the reafference would be learned together with the motor commands by one and the same network, so that neural resources were exploited in a highly efficient way.