OpenAI Gym Retro Release
OpenAI has released the full version of Gym Retro, a reinforcement learning platform that expands the available game library to over 1,000 titles across multiple classic consoles to enable research into agent generalization.
OpenAI AI and Compute Analysis
OpenAI analysis reveals that compute used for the largest AI training runs has grown exponentially since 2012, doubling every 3.4 months and far outpacing Moore's Law.
AI Safety via Debate
OpenAI proposes a safety technique that trains AI agents to debate topics to help humans supervise systems that perform tasks beyond human cognitive capabilities.
OpenAI Evolved Policy Gradients (EPG)
OpenAI introduces Evolved Policy Gradients (EPG), a metalearning approach that evolves loss functions to help RL agents generalize to new tasks within a task family.
OpenAI Gotta Learn Fast Benchmark
OpenAI has introduced a new reinforcement learning benchmark based on the Sonic the Hedgehog video game franchise to measure transfer learning and few-shot learning performance.
OpenAI Retro Contest and Gym Retro Release
OpenAI launched the Retro Contest to measure reinforcement learning generalization across unseen video game levels and released Gym Retro, a platform integrating 30 SEGA Genesis and 62 Atari 2600 games into Gym.
OpenAI Variance Reduction for Policy Gradient with Action-Dependent Factorized Baselines
OpenAI researchers developed a bias-free action-dependent baseline for policy gradient methods to reduce gradient estimate variance, enabling faster learning in high-dimensional action spaces.
OpenAI OT-GAN: Improving GANs using optimal transport
OpenAI introduces Optimal Transport GAN (OT-GAN), a variant of generative adversarial nets that uses a new mini-batch energy distance metric to improve training stability and image generation quality.
OpenAI Hackathon Report March 2018
OpenAI hosted its first hackathon on March 3, 2018, bringing together 100 AI community members to develop projects spanning healthcare, safety, and reinforcement learning.
On First-Order Meta-Learning Algorithms
OpenAI introduces Reptile, a first-order meta-learning algorithm that enables agents to learn quickly on new tasks by optimizing parameter initialization without requiring second-order derivatives.
OpenAI Reptile: A Scalable Meta-Learning Algorithm
OpenAI introduced Reptile, a scalable meta-learning algorithm that enables models to generalize from a small number of examples by maximizing the inner product between gradients of different minibatches from the same task.
OpenAI Scholars Program
OpenAI launched the OpenAI Scholars program to provide stipends and mentorship to individuals from underrepresented groups to study deep learning and open-source a project.
OpenAI Research: Learning to Explore via Meta-Reinforcement Learning
OpenAI researchers introduced E-MAML and E-RL², two meta-reinforcement learning algorithms designed to improve exploration performance in complex environments.
OpenAI Ingredients for Robotics Research
OpenAI released eight simulated robotics environments and a Baselines implementation of Hindsight Experience Replay (HER) to enable reinforcement learning from sparse rewards in complex manipulation tasks.
OpenAI Multi-Goal Reinforcement Learning Robotics Environments
OpenAI has released a suite of challenging continuous control robotics tasks integrated with OpenAI Gym to advance research in Multi-Goal Reinforcement Learning.
OpenAI Hackathon March 2018
OpenAI hosted a community hackathon and series of talks on March 3, 2018, in San Francisco to encourage AI learning and project development.
OpenAI Supporters and Organizational Updates
OpenAI announced new donors, advisor appointments, and the departure of Elon Musk from its Board of Directors to avoid potential conflicts with Tesla's AI focus.
OpenAI Preparing for Malicious Uses of AI
OpenAI and partners released a research paper forecasting how AI could be misused by malicious actors and providing recommendations for mitigating these global security threats.
OpenAI Interpretable Machine Learning through Teaching
OpenAI introduces a machine teaching approach that uses a teacher-student neural network framework to identify the most illustrative examples of a concept, ensuring the resulting interpretations are human-interpretable rather than arbitrary.
OpenAI Discovering Types for Entity Disambiguation
OpenAI developed a system that uses a neural network to predict membership in automatically discovered types to improve entity disambiguation accuracy on CoNLL and TAC KBP 2010 datasets.
OpenAI Requests for Research 2.0
OpenAI released Requests for Research 2.0, a collection of seven unsolved technical problems in reinforcement learning and machine learning to encourage community contribution.
Scaling Kubernetes to 2,500 Nodes
OpenAI describes the technical optimizations required to scale a Kubernetes cluster to 2,500 nodes on Azure for deep learning research, addressing bottlenecks in etcd, networking, and image pulls.
OpenAI Block-sparse GPU Kernels
OpenAI has released highly-optimized GPU kernels for block-sparse neural network architectures, enabling models to be wider and deeper while running orders of magnitude faster than cuBLAS or cuSPARSE.
OpenAI Learning Sparse Neural Networks through L0 Regularization
OpenAI introduces a method for L0 norm regularization in neural networks to create sparse models that improve training speed, inference speed, and generalization by using stochastic gates to make weights exactly zero.
OpenAI Interpretable and Pedagogical Examples Research
OpenAI researchers demonstrate that training teacher and student neural networks iteratively, rather than jointly, produces interpretable teaching strategies that can effectively teach both AI and humans.
OpenAI Learning a Hierarchy Research
OpenAI has developed Meta-Learning Shared Hierarchies (MLSH), a reinforcement learning algorithm that automatically discovers high-level actions to solve complex, long-horizon tasks more efficiently than brute-force methods.
OpenAI Generalizing from Simulation
OpenAI developed techniques using dynamics and domain randomization to train robot controllers in simulation that can generalize to physical robots and react to unplanned environmental changes.
OpenAI Sim-to-Real Transfer of Robotic Control with Dynamics Randomization
OpenAI researchers developed a dynamics randomization method that allows robotic control policies trained exclusively in simulation to transfer to real-world hardware without further physical training.
OpenAI Asymmetric Actor Critic for Image-Based Robot Learning
OpenAI introduces an asymmetric actor-critic framework that improves robot learning by training the critic on full simulator states while the actor relies only on RGBD images, enabling efficient simulation-to-real transfer without real-world data.
OpenAI Domain Randomization and Generative Models for Robotic Grasping
OpenAI developed a data generation pipeline using domain randomization and an autoregressive model to achieve an 80% real-world grasping success rate on unseen objects using only simulated training data.
OpenAI Meta-learning for Wrestling
OpenAI demonstrated that meta-learning agents can quickly adapt their tactics to defeat stronger fixed-policy agents and recover from physical malfunctions in simulated robot wrestling.
OpenAI Competitive Self-Play Research
OpenAI demonstrated that competitive self-play allows simulated 3D robots to autonomously discover complex physical skills like tackling and diving without explicit human design.
Nonlinear Computation in Deep Linear Networks
OpenAI researchers discovered that deep linear networks can perform nonlinear computation by exploiting floating-point arithmetic underflow around zero.
OpenAI and Oxford Introduce LOLA for Multi-Agent Reinforcement Learning
OpenAI and the University of Oxford have developed Learning with Opponent-Learning Awareness (LOLA), an RL agent that shapes the learning of other agents to achieve mutually beneficial outcomes.
OpenAI Learning with Opponent-Learning Awareness (LOLA)
OpenAI introduces Learning with Opponent-Learning Awareness (LOLA), a multi-agent reinforcement learning method that enables agents to shape the learning of other agents to foster cooperation and reach Nash equilibrium.
OpenAI Baselines: ACKTR & A2C
OpenAI has released Baselines implementations of ACKTR and A2C, providing a more sample-efficient reinforcement learning algorithm (ACKTR) and a synchronous, deterministic variant of A3C (A2C).
OpenAI Dota 2 1v1 Bot Results
OpenAI developed a reinforcement learning agent that achieved superhuman performance in Dota 2 1v1, demonstrating that self-play can scale machine learning systems beyond human-level capability.
OpenAI Dota 2 Bot Announcement
OpenAI has developed a bot capable of defeating top professional Dota 2 players in 1v1 matches using self-play reinforcement learning without imitation learning or tree search.
OpenAI RL-Teacher Release
OpenAI has released RL-Teacher, an open-source interface that allows AI agents to be trained via occasional human feedback instead of hand-crafted reward functions.
OpenAI Better Exploration with Parameter Noise
OpenAI introduces a method of adding adaptive noise to neural network parameters rather than the action space, significantly boosting reinforcement learning performance and exploration consistency.
Proximal Policy Optimization (PPO) Release Notes – OpenAI Baselines
OpenAI released Proximal Policy Optimization (PPO), a simple-to-implement reinforcement learning algorithm that matches or exceeds state-of-the-art performance while becoming the default RL method at OpenAI.
OpenAI Robust Adversarial Inputs Research
OpenAI has developed images that remain adversarial across varied scales and perspectives, challenging the notion that multi-perspective image capture in systems like self-driving cars prevents malicious trickery.
OpenAI Hindsight Experience Replay
OpenAI introduced Hindsight Experience Replay, a technique that enables sample-efficient reinforcement learning from sparse, binary rewards by treating failed attempts as successful achievements of alternative goals.
OpenAI Teacher–Student Curriculum Learning
OpenAI introduces Teacher–Student Curriculum Learning (TSCL), an automatic framework that optimizes training by selecting subtasks based on the student's learning progress and performance decay.
mujoco-py 1.50.1.0 release notes / what's new
OpenAI has open-sourced mujoco-py 1.50.1.0, a high-performance Python 3 binding for the MuJoCo engine that introduces batched simulation and GPU-accelerated rendering.
OpenAI Learning from Human Preferences
OpenAI and DeepMind developed a reinforcement learning algorithm that infers goals from human preference comparisons, reducing the need for manually written reward functions.
OpenAI MADDPG: Learning to Cooperate, Compete, and Communicate
OpenAI introduced MADDPG, a reinforcement learning algorithm that uses centralized learning and decentralized execution to enable multiple agents to collaborate and compete in complex environments.
OpenAI UCB Exploration via Q-Ensembles
OpenAI researchers developed an exploration strategy using an ensemble of Q*-functions based on Upper-Confidence Bounds (UCB) to improve deep reinforcement learning performance on Atari benchmarks.
OpenAI Baselines: DQN release and reinforcement learning best practices
OpenAI has open-sourced OpenAI Baselines, starting with a high-performance implementation of DQN and three of its variants to improve reproducibility in reinforcement learning.
OpenAI Robots That Learn: One-Shot Imitation Learning from Simulation
OpenAI has developed a robotics system trained entirely in simulation that can learn new tasks from a single human demonstration provided via VR.