Stumbled on a stat that 90% of neural data is thrown away in real-time BCIs

I was reading a paper from the 2023 BCI Society meeting and found out that most commercial BCI headsets only use about 10% of the raw neural signal they pick up. The rest is filtered out as noise or just ignored to save processing power. That blew my mind because I always assumed we were capturing everything. On one hand, filtering could help focus on clear brain waves like P300 or motor imagery. But on the other hand, are we missing important patterns that could improve accuracy for things like typing or prosthetic control? I know some researchers argue we should keep more data and let machine learning sort it out, while others say it's just wasted bandwidth. Where do you stand on this trade-off between efficiency and potential? Has anyone seen real-world results from keeping more raw data?

2 comments

2 Comments

casey9096d ago

90% thrown out, so my brain's basically been showing up to work half asleep this whole time.

oliver_mitchell6d ago

Whoa, I used to be totally on the side of filtering everything out to keep things clean and fast. But honestly, this post and that 90% stat really changed my mind. It makes me wonder how much useful noise we're just tossing in the trash because it doesn't fit our old models. I mean, machine learning is pretty good at finding patterns in messy data, so maybe we should throw more of this raw stuff at it and see what sticks. Your mileage may vary, but I think we're leaving a lot of potential on the table by being too quick to call it noise.