Brief
A central problem in neuroscience is to understand how the brain drives commands for movement, and how the brain re-enters these neural states during learning. This knowledge is also fundamental for engineering new therapies such as brain-machine interfaces (BMIs) to treat movement disorders. First, I will present my work using closed-loop BMIs as tools to study the brain, as BMIs allow the experimenter to choose and study the neurons that directly control movement of a neuroprosthetic computer cursor. I discovered that states that are shared across neurons follow predictable dynamics in time, and that while individual neurons are variable, the same shared dynamics are used to control a range of different movements. In addition, I found that these shared dynamics are learned with experience and reinforced with dopamine. These findings have applications for building better BMI algorithms to decode movement from the brain. Second, I will present my work developing a read-write optical neural interface to not only decode the real-time activity of neurons, but also activate targeted neurons in closed-loop based on ongoing neural activity and physical movement. I applied this system based on 2-photon microscopy and optogenetics to stimulate neurons deep in the brain through a GRIN lens in the striatum. Excitingly, I discovered that manipulating specific dimensions of shared neural dynamics modulates the force of specific physical actions. Altogether, my work has revealed that predictable, shared neural dynamics are used to control movement, learned with experience, reinforced with dopamine, and can be manipulated to drive specific physical actions.
I will conclude with discussion of my future research program, studying how neuronal connectivity gives rise to dynamics, how these dynamics constrain and enable learning, and how they descend to drive muscles and movement. Critically, the system I developed to photostimulate targeted neurons will enable measurement of connectivity in vivo, a long-standing challenge in neuroscience. Combining this read-write neural interface with computational models of neural dynamics promises to reveal principles for how the brain generates and learns behavior, with application to future neurotechnology and artificial intelligence.
Bio
Vivek Athalye is a senior scientist at the Allen Institute for Neural Dynamics, where he runs his own lab studying how neural dynamics emerge from connectivity and control action. Vivek did his bachelor’s degree in Electrical Engineering at Stanford University where he worked in Krishna Shenoy’s lab on implementing brain-machine interfaces in FPGA hardware for clinical application. Vivek did his PhD in Electrical Engineering and Computer Sciences at UC Berkeley with Dr. Jose Carmena, developing new brain-machine interfaces and statistical models of neural activity to study how the brain learns new skills from reinforcement. Vivek was also a visiting scientist at the Champalimaud Neuroscience Programme in Lisbon, Portugal, where he learned to perform neuroscience experiments in mice and develop new behavioral tasks. Vivek did his postdoc in Neuroscience at Columbia University with Dr. Rui Costa and Dr. Darcy Peterka, developing an optical system to stimulate targeted neurons and to study how neural dynamics control muscles and physical movement. He has received a K99/R00 Pathway to Independence Award and a BRAIN Initiative F32 postdoctoral fellowship. In addition, Vivek loves tennis and is a die-hard Roger Federer fan.
