Research

OpenAI conducts fundamental, long-term research toward the creation of safe AGI.

OpenAI Baselines DQN

OpenAI Baselines

OpenAI Baselines

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

OpenAI Baselines

Robots that Learn

Robots that Learn

Robots that Learn

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.

Robots that Learn

Unsupervised sentiment neuron

Unsupervised<br> Sentiment Neuron

Unsupervised
Sentiment Neuron

We’ve developed an unsupervised system which learns an excellent representation of sentiment, despite being trained only to predict the next character in the text of Amazon reviews.

Unsupervised
Sentiment Neuron

Evolution Strategies as a Scalable Alternative to Reinforcement Learning

Evolution Strategies

Evolution Strategies

We've discovered that evolution strategies (ES), an optimization technique that's been known for decades, rivals the performance of standard reinforcement learning (RL) techniques on modern RL benchmarks (e.g. Atari/MuJoCo), while overcoming many of RL's inconveniences.

Evolution Strategies

Generative Models

Generative Models

Generative Models

Four projects that share a common theme of enhancing or using generative models.

Generative Models

Publications

Research by OpenAI

We regularly publish our research, which focuses on reinforcement learning, robotics, and generative models.

Deep reinforcement learning from human preferences

Safety June 12, 2017

Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments

Language June 7, 2017

Parameter Space Noise for Exploration

Reinforcement Learning June 6, 2017

UCB and InfoGain Exploration via Q-Ensembles

Reinforcement Learning June 5, 2017

Equivalence Between Policy Gradients and Soft Q-Learning

Reinforcement Learning April 21, 2017

Learning to Generate Reviews and Discovering Sentiment

Language April 5, 2017

One-shot Imitation Learning

Robotics March 21, 2017

Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real World

Robotics March 20, 2017

Emergence of Grounded Compositional Language in Multi-Agent Populations

Language March 15, 2017

Prediction and Control with Temporal Segment Models

Generative Models March 12, 2017

Evolution Strategies as a Scalable Alternative to Reinforcement Learning

Evolution March 10, 2017

Third Person Imitation Learning

Robotics March 6, 2017

Adversarial Attacks on Neural Network Policies

Safety February 8, 2017

PixelCNN++: Improving the PixelCNN with Discretized Logistic Mixture Likelihood and Other Modifications

Generative Models January 19, 2017

Universe

Reinforcement Learning December 5, 2016

#Exploration: A Study of Count-Based Exploration for Deep Reinforcement Learning

Reinforcement Learning November 15, 2016

On the Quantitative Analysis of Decoder-Based Generative Models

Generative Models November 14, 2016

Stochastic Neural Networks for Hierarchical Reinforcement Learning

Reinforcement Learning November 12, 2016

A Connection between Generative Adversarial Networks, Inverse Reinforcement Learning, and Energy-Based Models,

Generative Models November 11, 2016

RL2: Fast Reinforcement Learning via Slow Reinforcement Learning

Reinforcement Learning November 9, 2016

Variational Lossy Autoencoder

Generative Models November 8, 2016

Adversarial Training Methods for Semi-Supervised Text Classification

Safety November 7, 2016

Extensions and Limitations of the Neural GPU

Memory November 2, 2016

Semi-supervised Knowledge Transfer for Deep Learning from Private Training Data

Safety October 18, 2016

Transfer from Simulation to Real World through Learning Deep Inverse Dynamics Model

Robotics October 11, 2016

Infrastructure for Deep Learning

Infrastructure August 29, 2016

Concrete Problems in AI Safety

Safety June 21, 2016

Improving Variational Inference with Inverse Autoregressive Flow

Generative Models June 15, 2016

InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets

Generative Models June 12, 2016

Improved Techniques for Training GANS

Generative Models June 10, 2016

OpenAI Gym

Reinforcement Learning June 5, 2016

Weight Normalization: A Simple Reparameterization to Accelerate Training of Deep Neural Networks

Optimization June 4, 2016

VIME: Variational Information Maximizing Exploration

Generative Models May 31, 2016