MIT 6.S191 Lecture 6: Deep Reinforcement Learning

preview_player
Показать описание
MIT 6.S191 Lecture 6: Deep Reinforcement Learning
Рекомендации по теме
Комментарии
Автор

Inspiring! You are very energetic and making other energies. Thanks

delowerhossain
Автор

The point that should be stressed is that the Q value not only depends on the score but on the game state that includes the score.

someguyfromafrica
Автор

Didn't gerry tesauro use NN in mid-90s to solve backgammon. So why do you say DeepMind was the first one to apply NN for state approximation? (27 min)

dubeya
Автор

+1 When lecture 3 and 4 be up at all? Thx.

陈健-bq
Автор

This was a great lecture. I wish there was more explanation of how the neural network precisely represents the policy simulating by Q in Q-learning tho..basically, I don't get how the neural network representing Q poses an advantage without having a crap-load of layers

rahulnath
Автор

Not a great lecture compared to bay area lectures. He presents dqns as the latest fashion while not talking about pg a3c methods at all.

randywelt
Автор

less "um" and "like" would be nice, and sounds is in and out.

arainboldt