Free Information Center
Exploration by random network distillation (RND) is a technique in reinforcement learning used to encourage agents to explore unfamiliar states by providing intrinsic rewards based on the novelty of observations. It leverages a fixed random neural network and a trainable network to measure prediction error as a proxy for novelty.
OpenAI Five is an artificial intelligence system developed by OpenAI to play the complex multiplayer online battle arena game Dota 2. It demonstrated advanced capabilities in strategic gameplay by competing against and defeating professional human players.
RL² is a framework that combines fast reinforcement learning with slow reinforcement learning to optimize learning efficiency in complex environments.
Stochastic value gradients (SVG) refer to a method used in optimization and machine learning, particularly for enhancing reinforcement learning algorithms.
Multitask reinforcement learning (MT-RL) is a subfield of machine learning that focuses on training agents to perform multiple tasks simultaneously using shared knowledge.
MuZero is a reinforcement learning algorithm developed by DeepMind that achieves high performance in games without prior knowledge of their rules by combining model-based and model-free learning.
Go-Explore is an exploration algorithm that enhances the efficiency of reinforcement learning in environments with sparse rewards by focusing on effective exploration strategies.
Dreamer is a model-based reinforcement learning algorithm that combines planning and learning to improve decision-making in complex environments.
AlphaStar is an artificial intelligence program developed by DeepMind to play the real-time strategy game StarCraft II at a professional level. It utilizes deep reinforcement learning and neural networks to master complex strategies and tactics within the game.
Distributional reinforcement learning (DRL) is a paradigm in machine learning that focuses on predicting the distribution of potential future rewards rather than a single expected value.