May 23, 2022
I am pleased to invite you to this week’s Neural Engineering Seminar! These seminars are focused on fostering engagement between speakers and students. This is a great opportunity to learn about and get more engaged in neural engineering, regardless of what your previous background may be.
Friday, May 27th, 4-5 PM in CSE2 382 (Bill and Melinda Gates Center): Reinforcement Learning for Deep Brain Stimulation
Dr. Joseph West, PhD
Visiting Researcher with Dr. Jeff Herron’s lab
RSVP (strongly encouraged): https://forms.gle/tZnL8TR9dY85L8Fv6
Abstract:
Deep Brain Stimulation (DBS) is used to manage the symptoms of severe, medically intractable neurological disorders which include Parkinson’s disease, essential tremor, epilepsy and obsessive-compulsive disorder. DBS consists of a stimulator which is placed in the chest cavity with wires running under the skin into electrodes which are implanted deep in the brain. The stimulator generates periodic electrical pulses to the electrodes which in turn stimulate the portions of the brain around the electrodes. Although DBS is effective for many patients the stimulation parameters are pre-set by a clinician, and as such the stimulation is always on and does not adapt for changes in the patient i.e. it is open loop. The pursuit of closed-loop DBS control is an important challenge as it could provide stimulation on-demand and also continual parameter adaption to account for changes in the patient’s illness or physical condition. Reinforcement learning has recently been shown to achieve superhuman performance in learning optimal sequential decision making in game playing (board and video) and numerous virtual and physical robotics problems including learning how to walk bipedally. These reinforcement learning problems although seemingly diverse are essentially closed loop control problems where the artificial decision making agent observes the environment and makes decisions to maximise some reward. My hypothesis is that a reinforcement learning agent can learn the optimum stimulation parameters for DBS therapy and through continual monitoring provide adaptive closed loop control. In this presentation I will discuss reinforcement learning for games and how we plan to reformulate the DBS stimulation problem into a continuous game suitable for a reinforcement learning agent as well as the practical limitations with this approach.
There will be a Q&A at the end of the talk, and refreshments provided!
Best,
The Neural Engineering Seminar series at the University of Washington