![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
cageymaru
Fully [H]
- Joined
- Apr 10, 2003
- Messages
- 22,080
So you may have seen humans get destroyed in chess by machine learning and thought that will never happen to me. Then machine learning took over the Go strategy game scene and you probably yawned because you've never thought of playing it. Well a small research team from MIT decided to use deep reinforcement learning to teach a computer how to defeat humans on a Nintendo game console. The computer was taught to play Super Smash Bros with a combination of off the shelf A.I. algorithms. These aren't video game algorithms, but can learn to play if given enough data. Then the research team set the computer to play against itself to learn how to master the game. This was the same strategy used to teach DeepMind AlphaGo to defeat Go players.
Interesting tidbits was that the A.I. reacted within 2 frames or 33ms. The average human has a 200ms reaction time. The A.I. never learned how to use projectiles so they limited the amount of character that it could choose from. The characters that professional players say were the most advanced and hardest to master; turned out to be the same characters that the A.I. was slowest to learn. The A.I. responded so fast to situations that the human players couldn't follow the logic and thought it was weird. The research team is ready to even the playing field by limiting the computer to human reaction times.
Are you ready to lose to a computer playing your favorite game? This is well above an aimbot; this is a competitor that reacts much faster to every situation. What do you think?
One interesting finding was that transfer learning, a hot topic in deep learning, applied across characters. This means that an AI agent trained on, say, a character like Fox McCloud, found its skills also applied to characters like Captain Falcon and Peach as well.
"I suspect transfer learning works because many of the fundamentals (how to move, attacking in the direction of the opponent when they are close) are broadly applicable to any character," Firoiu said.
Interesting tidbits was that the A.I. reacted within 2 frames or 33ms. The average human has a 200ms reaction time. The A.I. never learned how to use projectiles so they limited the amount of character that it could choose from. The characters that professional players say were the most advanced and hardest to master; turned out to be the same characters that the A.I. was slowest to learn. The A.I. responded so fast to situations that the human players couldn't follow the logic and thought it was weird. The research team is ready to even the playing field by limiting the computer to human reaction times.
Are you ready to lose to a computer playing your favorite game? This is well above an aimbot; this is a competitor that reacts much faster to every situation. What do you think?
One interesting finding was that transfer learning, a hot topic in deep learning, applied across characters. This means that an AI agent trained on, say, a character like Fox McCloud, found its skills also applied to characters like Captain Falcon and Peach as well.
"I suspect transfer learning works because many of the fundamentals (how to move, attacking in the direction of the opponent when they are close) are broadly applicable to any character," Firoiu said.
