Brain Machine Interfaces as a Testbed For Computational Modeling

From Santa Fe Institute Events Wiki

By Jose Principe, U of Florida.

The present generation of motor BMIs are simply signal translators (from multielectrode array recordings to kinematic variables). Recently we proposed a new co-adaptive close loop paradigm for BMIs based on reinforcement learning. The subject’s motor cortical neural activity is translated into a value function for a computer agent (CA) running Q learning that controls the actions of a robot to deliver a water reward to the subject (a rodent). Preliminary results show that the subject and the agent are able to cooperate and improve their joint performance across task difficulty. We will briefly describe the current tests and address the computational modeling challenges.

Back to Agenda.