Researchers used the San Diego Supercomputer Center's Comet (SDSC) system in applying a novel algorithm to replicate neural pathways for controlling limb movement.
The research lays the foundation for "biomimetic neuroprosthetics"--implants that reproduce brain circuits and their function--that could replace lost or damaged neurons or tissue.
State University of New York professor W.W. Lytton suggests such implants could enable motor-paralysis patients to direct a prosthetic by engaging directly with healthy pre-motor neurons.
The researchers trained a biomimetic model via spike-timing dependent plasticity and reinforcement learning, while evolutionary algorithms were employed to isolate reinforcement learning parameters that yielded the best governance of a virtual arm.
The team evolved 60 individuals over 1,000 generations, using an "island model" strategy to boost the odds of returning an optimal solution.
"Integrating the parallel 'island model' on Comet required some work, but we made it work to further speed the process," says SDSC's Subhashini Sivagnanam.
From UCSD News (CA)
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found