Reinforcement learning

lastmod 2026-03-06created 2016-11-167 edits

Reinforcement learning is one of the problems dealt with in machine learning: an agent defined within an environment perceives its current state and, among the available actions, selects the action or sequence of actions that maximizes reward.

Summery

The environment can generally be expressed as a finite-state Markov decision process (MDP). From this perspective, reinforcement learning algorithms are closely related to dynamic programming. In an MDP, the state transition probabilities and rewards may be stochastic values or values that are already determined.

Reinforcement learning differs from supervised learning, which explicitly corrects unwanted actions, in that it focuses on online performance. Reinforcement learning is about balancing the exploration of uncharted territory against the exploitation of existing knowledge. This trade-off between exploration and exploitation can be seen in problems such as the Multi-armed bandit.

Categories

• Fully Attentional Network for Semantic Segmentation

Documentation

Fundamental of Reinforcement Learning

https://www.gitbook.com/book/dnddnjs/rl/details

MODULABS - Basics of reinforcement learning

Learning Monocular Reactive UAV Control in. Cluttered Natural Environments.

https://arxiv.org/abs/1211.1690

See also

• Deep learning
• Markov decision process
• Unsloth - Fine-tuning & Reinforcement Learning for LLMs

Favorite site

• Wikipedia (en) Reinforcement learning
• Reinforcement Learning and Artificial Intelligence (reinforcement learning and artificial intelligence)
• Mathematical foundations of reinforcement learning (RL): book and YouTube lectures | GeekNews
  • [Original] Github - MathFoundationRL/Book-Mathematical-Foundation-of-Reinforcement-Learning - This is the homepage of a new book entitled "Mathematical Foundations of Reinforcement Learning."
  • Mathematical Foundations of Reinforcement Learning - YouTube - English YouTube lecture playlist