logoalt Hacker News

kazinatoryesterday at 1:52 AM0 repliesview on HN

Ironically, you could use the sigmoid function instead of horses. The training stimulus slowly builds over multiple iteration and then suddenly, flip: the wrong prediction reverses.