A machine learning competition, called Brain-to-text ’25, will challenge experts to predict the speech of a brain-computer interface (BCI) user who lost their ability to speak due to a neurodegenerative disease. The individual or team with the lowest error rate between predicted and actual sentences will win a US $5,000 prize.
The competition is hosted by the University of California Davis’s Neuroprosthetics Lab and uses brain data from 46-year-old Casey Harrell, whose speech is unintelligible except to his regular caregivers. The goal is to improve the accuracy of BCIs for patients with diseases like amyotrophic lateral sclerosis (ALS).
Decoding speech from brain data involves two steps: predicting phonemes and then words. Competitors will train their algorithms on 10,948 sentences and then predict 1,450 unseen sentences. The competition aims to attract machine learning experts who can help mature the technology.
However, ethical concerns surrounding the use of patient data must be addressed. Researchers have taken precautions to protect Harrell’s identity, but bioethicist Veljko Dubljević warns that even with safeguards, there is a risk of identifying the patient in 50 years.
The competition offers cash prizes for the lowest error rates and the most innovative approach. The top slots are likely to go to coders with no background in BCIs who focus on speed over ingenuity, but both traditional and new approaches will contribute to the science engineering ecosystem.
Source: https://spectrum.ieee.org/speech-bci-machine-learning-competition