Deep learning shapes cortical stimulation in a blind participant
Matt Perko, UC Santa BarbaraPress kit
A deep neural network trained to predict how the visual cortex answers electrical stimulation designed better stimulation than the standard approaches, in a study published in Neuron on 7 August 2026 under the title “Deep learning-based control of electrically evoked activity in human visual cortex”.
The work was carried out on one participant: a 27-year-old man who lost his sight after a traumatic brain injury and received a 96-channel electrode array in his visual cortex at Hospital IMED Elche in Spain. The implant was temporary and scheduled for removal after six months.
The network was given recordings of the participant’s own resting brain activity and learned to predict the activity produced by different combinations of stimulation settings. Patterns it designed reproduced the targeted brain activity more accurately, and needed less electrical current, than the approaches compared against. The team also found that measuring the brain’s response predicted what the participant reported seeing better than the stimulation settings alone did.
The authors are Michael Beyeler and Jacob Granley at UC Santa Barbara, Pehuén Moure and Shih-Chii Liu at ETH Zurich, and Fabrizio Grani and Eduardo Fernández at Miguel Hernández University. The work was funded by a 2022 NIH Director’s New Innovator Award worth $2 million over five years.
This is a single-participant proof of concept on a temporary implant, not a device anyone can receive.