❤️ Apoiar projeto
PT
EN English UK Українська PL Polski ES Español DE Deutsch FR Français PT Português TR Türkçe JA 日本語 KO 한국어 RU Русский

Reinforcement Learning

Aprendizagem por Reforço (AR) é um paradigma de aprendizado de máquina onde um agente aprende a tomar decisões em um ambiente para maximizar um sinal de recompensa.

🌐 Termos em outros idiomas:
English Deutsch Español Français 日本語 한국어 Polski Português Русский Türkçe Українська

AR envolve um agente interagindo com um ambiente, tomando ações e recebendo feedback na forma de recompensas ou penalidades. O objetivo do agente é aprender uma política, que é uma estratégia para selecionar ações que maximizem a recompensa cumulativa ao longo do tempo. Este processo de aprendizado é frequentemente modelado como um Processo de Decisão de Markov (MDP), onde as ações do agente influenciam o estado do ambiente, e o ambiente fornece recompensas com base nessas transições de estado.

Algoritmos de AR exploram o ambiente por tentativa e erro, melhorando gradualmente sua política. Essa troca entre exploração e exploração é crucial, pois o agente deve equilibrar a tentativa de novas ações (exploração) com a exploração do conhecimento que já adquiriu (exploração). Vários algoritmos, como Q-learning, SARSA e gradientes de política, são usados para treinar agentes de AR. Esses algoritmos atualizam a política ou a função de valor do agente com base nas recompensas recebidas, guiando-o para um comportamento ideal.

        graph LR
  Center["Reinforcement Learning"]:::main
  Pre_machine_learning["machine-learning"]:::pre --> Center
  click Pre_machine_learning "/terms/machine-learning"
  Rel_deep_learning["deep-learning"]:::related -.-> Center
  click Rel_deep_learning "/terms/deep-learning"
  Rel_game_theory["game-theory"]:::related -.-> Center
  click Rel_game_theory "/terms/game-theory"
  Rel_logistic_regression["logistic-regression"]:::related -.-> Center
  click Rel_logistic_regression "/terms/logistic-regression"
  classDef main fill:#7c3aed,stroke:#8b5cf6,stroke-width:2px,color:white,font-weight:bold,rx:5,ry:5;
  classDef pre fill:#0f172a,stroke:#3b82f6,color:#94a3b8,rx:5,ry:5;
  classDef child fill:#0f172a,stroke:#10b981,color:#94a3b8,rx:5,ry:5;
  classDef related fill:#0f172a,stroke:#8b5cf6,stroke-dasharray: 5 5,color:#94a3b8,rx:5,ry:5;
  linkStyle default stroke:#4b5563,stroke-width:2px;
🕸️ Open in Universe

🧠 Teste de conhecimento

1 / 3

🧒 Explique como se eu tivesse 5 anos

🎮 Training a computer program like a puppy: rewarding good behavior and ignoring bad behavior until it learns to be helpful.

🤓 Expert Deep Dive

## RLHF: Aligning Human and Machine
Reinforcement Learning from Human Feedback (RLHF) is the secret sauce behind modern chatbots like ChatGPT. Since it's impossible to write a mathematical formula for 'a good, helpful answer,' we show the model pairs of answers and let humans rank them. An RL agent is then trained to predict these rankings, creating a 'Reward Model' that guides the LLM toward safe and helpful output.

🔗 Termos relacionados

Pré-requisitos:

📚 Fontes

1. Playing Atari with Deep Reinforcement Learning
2. Reinforcement Learning: An Introduction (2nd Edition)
3. Mastering the game of Go with deep neural networks and tree search
4. AlphaFold 2: Highly accurate protein structure prediction with deep learning
5. Decision Transformers: Reinforcement Learning via Sequence Modeling
6. Training language models to follow instructions with human feedback
7. Meta-Reinforcement Learning
8. Causal Reinforcement Learning
9. Sim-to-Real Transfer of Robotic Control with Dynamics Randomization
10. Graph Neural Networks for Deep Reinforcement Learning
11. Explainable Reinforcement Learning: A Survey
12. Hierarchical Reinforcement Learning: A Survey
13. Reinforcement Learning from Human Feedback
14. Graph Neural Networks for Deep Reinforcement Learning
15. Meta-Reinforcement Learning
16. Offline Reinforcement Learning: Tutorial, Review, and Perspectives
17. Decision Transformer: Reinforcement Learning via Sequence Modeling
18. Highly accurate protein structure prediction with AlphaFold
19. Asynchronous Methods for Deep Reinforcement Learning
20. Meta-Reinforcement Learning
21. Curiosity-Driven Exploration by Self-supervised Prediction
22. Asynchronous Methods for Deep Reinforcement Learning
23. Deep reinforcement learning for robotic manipulation
24. Deep Reinforcement Learning from Human Preferences
25. Deep Reinforcement Learning for Recommendation Systems
26. Reinforcement Learning from Human Feedback
27. Graph Neural Networks for Reinforcement Learning
28. Decision Transformers: Reinforcement Learning via Sequence Modeling
29. Meta-Reinforcement Learning
30. AlphaFold 2
31. Reinforcement Learning from Human Feedback
32. An open science platform for reproducible research in artificial intelligence
33. Deep Reinforcement Learning: An Introduction