Seminars in Hearing Research (09/24/20) - Brandon Coventry
Seminars in Hearing Research (09/24/20) - Brandon Coventry
Author: | M. Heinz |
---|---|
Event Date: | September 24, 2020 |
Hosted By: | Hari Bharadwaj |
Time: | 1030-1120 |
Location: | Zoom |
Contact Name: | Bharadwaj, Hari M |
Contact Email: | hbharadw@purdue.edu |
Open To: | All |
Priority: | No |
School or Program: | Biomedical Engineering |
College Calendar: | Show |
Seminars in Hearing Research at Purdue (SHRP)
Title: Towards closed-loop optical control of auditory thalamocortical circuits using deep reinforcement learning
Speaker: Brandon S Coventry, Ph.D. Candidate, Weldon School of Biomedical Engineering (Bartlett Lab)
Date: September 24, 2020
Time: 10:30 – 11:20 am
Zoom Info:
Topic: Seminars in Hearing Research at Purdue (SHRP)
Abstract:
Closed-loop neuromodulation, also called intelligent neural control, has become the holy grail of both neuroprostheses and brain computer interfaces for its promise to sense specific physiologic states and take targeted therapeutic actions when certain conditions are met. However, current closed loop systems are limited to primarily simple threshold measures only allowing limited control of neural circuits. A method that can learn relevant physiologic states would allow for more fine-tuned control of the neural circuit under control, having direct therapeutic use for both hearing restoration and cochlear implants as well as Parkinson’s disease and other neurological diseases and disorders. One such method exists in the class of machine learning algorithms known as reinforcement learning (RL). In RL tasks, the system under study is treated as a game environment where certain actions can lead to short or long-term rewards. The goal of the algorithm is to learn an approximate model of the system under study and the actions to take which maximize both long-term and short-term rewards.
In this talk, we will discuss my current work with infrared neural stimulation (INS), a label-free targeted optical neuromodulation technique, and introduce our new toolbox, SpikerNet, which can implement reinforcement learning-based closed loop control. We will also discuss how reinforcement learning can be used in physiological studies in general. Rats in our study were implanted with 16 channel recording arrays into auditory cortex and a fiber optic stimulating “optrode” into the ventral division of the medial geniculate body. Our optical stimulation results suggest that INS is highly localized to local microcircuit stimulation and can reliably stimulate auditory thalamocortical circuits. We also will show our initial work with SpikerNet which will lead to a targeted, closed-loop neuromodulation tool.