Hacker News new | ask | show | jobs
by skybrian 2522 days ago
Well, maybe. Since most of this data will be very repetitive, I don't think we can know for sure how much it helps?
2 comments
imglorp 2522 days ago
You (and neural nets) learn the most when a boring repetition is violated. The error function was activated and that's worth feeding back into learning.

Ask any kid, they probably couldn't say what happened at school all week. But if they do remember something, it would have been something outside of their routine.

link
imglorp 2522 days ago
You (and neural nets) learn the most in the smaller number of cases when a boring repetition is violated. The error function was activated and that's worth feeding back into learning.

Ask any kid, they probably couldn't say what happened at school all week. But if they do remember something, it would have been something outside of their routine.

link
skybrian 2522 days ago
Yes, exactly! We don't know the amount or quality of interesting data from the total amount of data.
link