OpenAI Retro Contest

Re:OpenAI Retro Contest by mustdeparthasty on Thursday April 05, 2018 @04:41PM (#16768627)

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

[ Parent | on HN ]

Re:OpenAI Retro Contest by hwoolery on Thursday April 05, 2018 @05:59PM (#16769152)

Of 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 : )

[ Parent | on HN ]

Re:OpenAI Retro Contest by solarkraft on Thursday April 05, 2018 @06:51PM (#16769490)

I don't consider it dissuasion, but a fair warning & proposition preparation. Seriously: Good luck!

[ Parent | on HN ]

Re:OpenAI Retro Contest by freyir on Friday April 06, 2018 @12:43AM (#16770927)

> 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.

[ Parent | on HN ]

Re:OpenAI Retro Contest by debt on Thursday April 05, 2018 @05:36PM (#16768977)

This should be required reading for anyone wanting to do this contest.

[ Parent | on HN ]

Re:OpenAI Retro Contest by backpropaganda on Friday April 06, 2018 @01:52AM (#16771182)

Do 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.

[ Parent | on HN ]

Re:OpenAI Retro Contest by unityByFreedom on Friday April 06, 2018 @01:41AM (#16771140)

> Two months ... Good luck

Well, it needs to be 2 months so Musk can transfer this learning to Autopilot... /s, I think

[ Parent | on HN ]