3 ms·
Do they use gradient descent? Otherwise how do they train?
Do they use gradient descent? Otherwise how do they train?
From the paper, it looks like they apply a long random stimulus when it misses the ball, and a short predictable stimulus when it doesn't. The theory is that the neurons will try to organize in a way that avoids the random stimulus.