AI-driven multi-target neuroprosthetic optimization

We will design novel BO strategies to parallelize multi-target neurostimulation optimization.
The development of large-channel-count neural interfaces aims at obtaining high-throughput communication with the nervous system by parallelizing the flow of information over multiple subregions. For instance, the motor cortex, spinal cord and nerves host representations of multiple bodily movements within highly-interconnected topographies. Neural interfaces are currently unable to explore these arrangements to evoke diverse functional motor outputs. This mapping is restricted to the human capacity to focus on one target at a time. This technological gap severely limits the development of multivariate neuroscientific investigation and multimodal neuromodulation intervention.
We propose a new algorithm for dynamical multi-target Bayesian optimization, suitable to parallelize multi-target optimization, allowing full identification of the controllability space (i.e., all possible outputs that can be generated with the available inputs) of a single interface.

Faculty Supervisor:

Marco Bonizzato

Student:

Partner:

École polytechnique fédérale de Lausanne

Discipline:

Engineering

Sector:

Education

University:

Université de Montréal

Program: