Comment on Braincraft challenge – 1000 neurons, 100 seconds, 10 runs, 2 choices, no rewardComments−rhaps0dy1yThe 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).
Comments
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).