7 articles
OpenAI publishes a study on count-based exploration for deep reinforcement learning. Researchers analyze how agents can efficiently explore environments using counting mechanisms to identify rarely-visited states.
OpenAI and Microsoft expand their partnership: OpenAI will now run most of its large-scale experiments on Microsoft's Azure infrastructure.
OpenAI publishes quantitative analysis of decoder-based generative models. The study examines mathematical mechanisms and performance properties of these architectures, with no specific results or benchmarks detailed in the excerpt.
OpenAI establishes a theoretical connection between generative adversarial networks (GANs), inverse reinforcement learning, and energy-based models. This mathematical unification provides a new perspective on three fundamental machine learning paradigms.
OpenAI introduces RL², a reinforcement learning method that leverages slow learning to enable fast adaptation of agents. The technique trains models to learn efficiently from limited experience, improving generalization and convergence speed on new tasks.
OpenAI introduces a variational lossy autoencoder, an unsupervised compression architecture that learns compact latent representations. The method combines VAE and lossy compression to reduce dimensionality of complex data.
OpenAI analyzes capabilities and limitations of neural GPUs for AI model acceleration. The study examines current performance and technical obstacles to overcome for improved computational efficiency.