Articles Collectés

1,002

Total

941

Non traités

61

Traités

Non traitée 11/10/2017 09:00
Competitive self-play

RSS: OpenAI News

We’ve found that self-play allows simulated AIs to discover physical skills like tackling, ducking, faking, kicking, catching, and diving for the ball, without explicitly designing an environment wi...

Non traitée 11/10/2017 09:00
Meta-learning for wrestling

RSS: OpenAI News

We show that for the task of simulated robot wrestling, a meta-learning agent can learn to quickly defeat a stronger non-meta-learning agent, and also show that the meta-learning agent can adapt to ph...

Non traitée 29/09/2017 09:00
Non traitée 14/09/2017 09:00
Learning to model other minds

RSS: OpenAI News

We’re releasing an algorithm which accounts for the fact that other agents are learning too, and discovers self-interested yet collaborative strategies like tit-for-tat in the iterated prisoner’s ...

Non traitée 13/09/2017 09:00
Non traitée 18/08/2017 09:00
OpenAI Baselines: ACKTR & A2C

RSS: OpenAI News

We’re releasing two new OpenAI Baselines implementations: ACKTR and A2C. A2C is a synchronous, deterministic variant of Asynchronous Advantage Actor Critic (A3C) which we’ve found gives equal perf...

Non traitée 16/08/2017 09:00
More on Dota 2

RSS: OpenAI News

Our Dota 2 result shows that self-play can catapult the performance of machine learning systems from far below human level to superhuman, given sufficient compute. In the span of a month, our system w...

Non traitée 11/08/2017 09:00
Dota 2

RSS: OpenAI News

We’ve created a bot which beats the world’s top professionals at 1v1 matches of Dota 2 under standard tournament rules. The bot learned the game from scratch by self-play, and does not use imitati...

Non traitée 03/08/2017 09:00
Gathering human feedback

RSS: OpenAI News

RL-Teacher is an open-source implementation of our interface to train AIs via occasional human feedback rather than hand-crafted reward functions. The underlying technique was developed as a step towa...

Non traitée 27/07/2017 09:00
Better exploration with parameter noise

RSS: OpenAI News

We’ve found that adding adaptive noise to the parameters of reinforcement learning algorithms frequently boosts performance. This exploration method is simple to implement and very rarely decreases ...

Non traitée 20/07/2017 09:00
Proximal Policy Optimization

RSS: OpenAI News

We’re releasing a new class of reinforcement learning algorithms, Proximal Policy Optimization (PPO), which perform comparably or better than state-of-the-art approaches while being much simpler to ...

Non traitée 17/07/2017 09:00
Robust adversarial inputs

RSS: OpenAI News

We’ve created images that reliably fool neural network classifiers when viewed from varied scales and perspectives. This challenges a claim from last week that self-driving cars would be hard to tri...

Non traitée 05/07/2017 09:00
Non traitée 01/07/2017 09:00
Non traitée 28/06/2017 09:00
Faster physics in Python

RSS: OpenAI News

We’re open-sourcing a high-performance Python library for robotic simulation using the MuJoCo engine, developed over our past year of robotics research.

Non traitée 13/06/2017 09:00
Learning from human preferences

RSS: OpenAI News

One step towards building safe AI systems is to remove the need for humans to write goal functions, since using a simple proxy for a complex goal, or getting the complex goal a bit wrong, can lead to ...

Non traitée 08/06/2017 09:00
Learning to cooperate, compete, and communicate

RSS: OpenAI News

Multiagent environments where agents compete for resources are stepping stones on the path to AGI. Multiagent environments have two useful properties: first, there is a natural curriculum—the diffic...

Non traitée 05/06/2017 09:00
Non traitée 24/05/2017 09:00
OpenAI Baselines: DQN

RSS: OpenAI News

We’re open-sourcing OpenAI Baselines, our internal effort to reproduce reinforcement learning algorithms with performance on par with published results. We’ll release the algorithms over upcoming ...

Non traitée 16/05/2017 09:00
Robots that learn

RSS: OpenAI News

We’ve created a robotics system, trained entirely in simulation and deployed on a physical robot, which can learn a new task after seeing it done once.