ManySpace
Türkçe Sign in Sign up

AI & MLSpecializedAIML-RL-021

Offline Reinforcement Learning

Learning policies from fixed logged data without further interaction, and the out-of-distribution action problem.

Difficulty
5 / 5
Importance
3 / 5
Study time
1–2 h
Kind of work
Mixed

Why it matters

The learner can explain why naive off-policy Q-learning overestimates values for actions absent from the dataset and name one conservative fix.

The route to this topic

Route AIML-RL-021
Natural Numbers What Is a Model? Rational Agents and Task Environments Variables & Constants Tasks: Regression, Classification, and Beyond Agent and Environment Solving Linear Equations Function Notation & Composition Classification Framing Return and Discounting Policies Complex Numbers Scores and Logits Exploration vs Exploitation Markov Decision Processes Basic Ratios The Artificial Neuron Value Functions Polynomial Functions Activation Functions Bellman Intuition Monte Carlo Prediction and Control Definition of the Derivative Differentiation Rules The Multilayer Perceptron Temporal-Difference Learning Chain Rule Forward Propagation Q-Learning Computational Graphs The Chain Rule in Neural Nets Backpropagation Deep Q-Networks Offline Reinforcement Learning