26th Annual Computational Neuroscience Meeting ({CNS}{\ast}2017): Part 3

Neuro computational models represent a powerful tool for bridging the gap between functions of the neural circuits and observable behaviors [1]. Once the model has been built, its output is compared with the observations either to validate the model itself or to propose new hypotheses. This approach has led to building a multi-scale model of the sensorimotor system from muscles, proprioceptors to skeletal joints, spinal regulating centers and central control circuits [2–6]. In this framework, we propose a neural network architecture to simulate the selection of actions performed by the motor cortex in response to a sensory input during a reward-based movement learning. The network has as many input nodes as the number of different stimuli, each node being a combination of the sensory inputs, and as many output nodes as the number of different actions that can be performed, each node being a combination of the motor commands. The network is fully connected, so that each stimulus concurs to the selection of each action and each action is selected concurrently by all the stimuli. The weights are updated by taking into account both the expected reward and the actual reward, as suggested in [7]. By adopting this architecture, the percept is represented by a combination of sensory inputs, while the action is represented by a combination of motor commands. Thus, it reproduces faithfully the condition of experiments of motor learning when a set of sensory inputs, such as semantically neutral visual stimuli, are presented to the subject whose response is merely a motor action, such as pushing a button. Under such conditions, it then becomes possible to fit the data provided by the experiments with the model to both estimate the validity of the model and to infer the role of the parameter on behavioral traits. The simulations were compared to the behaviors of human subjects while learning which out of two buttons to press in response to a collection of visual stimuli containing edges and geometric shapes in a reward based setting. The results showed that the behavior of the complete system is the one expected under the hypothesis that the reward acts by modulating the action selection triggered by the input stimuli during motor learning. Moreover, differently from most literature models, the learning rate varies with the complexity of the task, i.e. the number of input stimuli. It can be argued that the decrease in learning rate seen in humans learning large set of stimuli could be due to an attenuation of memory traces in real synapses over time. In our future investigations, we will work to improve the model by adding such an effect in our network.