Hacker News new | ask | show | jobs
by rhaps0dy 284 days ago
The easiest thing to do is probably use A* to find the solution, then imitation learning on the NN to learn it. (The immediate feedback gets rid of the vanishing/exploding gradients problem).