Sensory Decision Making and Behavioral Control in Multi-Area Cortical Attractor Networks
Playing an action video game requires fast processing of sensory cues, rapid decision-making, and the accurate control of movements. Goal of this project is to construct biologically realistic functional network models of cortical processing that accumulate sensory evidence, form decisions and select appropriate actions in behavioral reaction tasks that mimic the challenges faced in an action game. Theoretical work has identified cortical attractor networks as core processing units and there is increasing experimental evidence that neural activity in the cortex signifies attractor dynamics. We will use recurrent neural networks of spiking neurons with locally balanced clusters of excitatory and inhibitory neurons as substrate for computation with attractors. The neural dynamics in simulations will be compared to neural dynamics recorded from behaving primates. Downsized models will be implemented on real-time neuromorphic hardware allowing for application in the control of autonomous robots.
Methodology & Research Model
Modelling of spiking neural networks, advanced neural data analysis methods.
Requirements for Application
Proficient programming skills, preferably in a high level language such as Python. Basic understanding of dynamical systems theory. Experience with spikig network simulators and with the analysis of neural data is of advantage.