Machine learning masters can do it for the next five months Try to best predict your Brain-Computer Interface (BCI) user's speech People who have lost their ability to speak due to neurodegenerative diseases. Competitors design algorithms that predict words from patient brain data. The individual or team whose algorithm makes the least error between the predicted and actual tried statements will win a prize of USD 5,000.
The competition, called Brain-text '25, is a second annual public brain-to-text competition hosted by part of the Braingate Consortium lab, a pioneer in BCI clinical trials since the early 2000s. This year, the contest is run by the Institute for Neurosuppression at the University of California, Davis. (Brain data from another BCI user was used to hold the first competition by a group from Stanford University.)
For two years, the UC Davis Research team Casey Hallell, a 46-year-old man. That speech is hard to understand except for his regular caregiver. Speech BCI After Harrell's Training Brain data, it can decode what he was trying to say Over 97% Time can be done I'll synthesize it immediately His own voice, as before It has been reported by IEEE Spectrum.
Deciphering speeches from brain data
Analyzing words from brain data is a two-step process. The algorithm must first predict the sound of a voice called a phonem from neural data. You then need to predict words from phonemes. Competitors train algorithms with corresponding brain data 10,948 sentences containing the attached transcript That's what Harrell was trying to say.
Next, the actual test is performed: the algorithm must predict the word of Starting from 1,450 sentences Brain data withheld from training data. The difference between the final set of predicted words and the word Harrell tried to say is called the word error rate.– The lower the error rate of a word, the overall functioning of the speech BCI.
Researchers reported a word error rate of 6.70%, which the public hopes to win. The goal of the competition is to attract machine learning experts, Nick Card, a postdoctoral researcher at UC Davis, who leads both clinical trials and competition, to attract machine learning experts who may not understand how valuable their skills are to BCIS.
“We can sit and hide this data internally and discover more over time,” says Card. “But if your goal is to mature this technology faster and help people who need to benefit from it right now, we want to share it.
Public invitations to the research world are “long postponed” and “great development” in the BCI field, University of Pennsylvania professor Konrad Kording says he uses machine learning to study the brain and is not involved in research or competition.
This year, Card and his fellow researchers raised the bar by lowering the starting word error rate with their own high-performance algorithms. The first brain competition in 2024 ended with competition winners who achieved 5.77%, with Stanford University Group posting an error rate of 11.06%. This year, we also offer cash awards for new error rates and the most innovative approaches, as well as BlackRock NeuroTech, a BCI company whose Braingate Hardware has been used in Braingate clinical trials since 2006.
BCI has long served as a bridge between neuroscience, medicine and machine learning. Additionally, machine learning has a tradition of open source research, but medical research is bound by patient confidentiality.
The main concern with public brain data is patient identification, according to bioethicist Veliko Davjevich, a professor of both philosophy and science, technology and society at North Carolina State University.
In this case, this concern comes as Harrell was published in August 2024, about five years after he began losing muscle tone due to amyotrophic lateral sclerosis (also known as Lugelig's disease). 2023, A neurosurgeon at UC Davis implanted four electrode arrays There are a total of 256 electrodes at the top of his brain. Hallell I used his speech BCI in an interview in New England Journal of Medicine How to explain the illness last year It feels like you're in a “slow motion car crash.” Harrell said at the time, “It was extremely painful to lose the ability to communicate, especially with her daughter.”
Speech BCI is trained with collected data, while Harrell conducts experiments during love and casually speaks with family and friends. However, Brain-tostext '25 competitors don't see “personal use” data recorded during Harrell's casual discussion, Card says.
This is a “good precaution,” but I think Harrell understands what it means to have someone's sensitive medical data in the public domain for many years. For example, in a similar way to how blood was donated in 1955, today's BCIS “noise” could be decoded into meaningful personal information in 50 years. (DNA profiling was not established until the 1980s.) Dubljević recommends limiting data storage to five years.
Speech BCIS deciphers the intended movement of a person's jaw and mouth muscles, just as BCIs deciphers of arms and hand prosthetics decipher the intended movement. However, I feel that the speech BCI is more personal than the BCI, which controls the prosthetics of the hand, says Dubljević. He says that speeches are closer to “the innermost sanctuary of a person.” “There's a lot of fear about reading the mind, right?”
“As a researcher who wants to see science and technology being deployed in the public good, I want to avoid hype about it,” says Dubljević to avoid backlash.
Cash Awards for Innovative BCI Solutions
The lowest two word error rates have cash awards of $5,000 and $3,000 respectively, with the most innovative approach winning $1,000.
The final category is intended to promote highly potential ideas given more data or more time. Stacking ten times the same algorithm is a common way to force a more accurate overall performance, but it takes ten times more computing power and says, “Frankly, isn't that a very creative solution?” card says.
The innovative categories are likely to attract the usual crowd of academic and industry BCI scientists who enjoy finding creative solutions, Kording says.
But the top slots could go to the coder without a background in BCIS and sports “street fighting” style machine learning, as Kolding calls it. These “street fighters” focus on speed rather than ingenuity. In fact, the best BCI algorithms say “usually they don't actually drive because of a deep knowledge of how the brain works. They drive because they have a deep understanding of how machine learning works.”
That said, Kolding says both traditional BCIS and newcomers are important parts of the science and engineering ecosystem. Once the corners are full, the competition is set to be an exciting battle.
From the article on the site
Related articles on the web
