4 ms·
Two months to advance the state of the art on a complex physics-based game with branching paths. Current approaches such as DQNs or god forbid DRL[1] barely rea
by mustdeparthasty 9y ago
Two months to advance the state of the art on a complex physics-based game with branching paths. Current approaches such as DQNs or god forbid DRL[1] barely reach the performance of my three year old cousin in atari game score maximization and are mostly non-transferable to new levels... Good luck.
[1] https://www.alexirpan.com/2018/02/14/rl-hard.html https://www.alexirpan.com/2018/02/14/rl-hard.html
- debt 9y agoThis should be required reading for anyone wanting to do this contest.
- hwoolery 9y agoOf course it's a difficult problem, but why try and dissuade people from attempting it? Nothing novel ever came from somebody who didn't fail first : )
- solarkraft 9y agoI don't consider it dissuasion, but a fair warning & proposition preparation. Seriously: Good luck!
- freyir 9y ago> Current approaches such as DQNs or god forbid DRL You say that like you're talking about two separate things. DQN is just one type of DRL.
- unityByFreedom 9y ago> Two months ... Good luck Well, it needs to be 2 months so Musk can transfer this learning to Autopilot... /s, I think
- backpropaganda 9y agoDo note that most RL methods including DQN were designed to be very general. While this contest doesn't really require generality. You could design a model specialized to the characteristics of the game. This is what the top winners of the Vizdoom challenge did.