AD-802 (B) RGPV B.Tech
Reinforcement Learning (RL)
Previous year papers and the official syllabus are free to read, no account needed.
Previous Year Papers
1 paper, latest Jun 2025
Syllabus, notes and important questions for AD-802 (B)
Syllabus units
Topics as the official RGPV syllabus lists them.
Unit 1 Introduction to Reinforcement Learning 6 topics
- Definition of reinforcement learning framework and Markov Decision Process ($MDP$)
- Policies
- Value functions and Bellman equations
- Exploration vs. exploitation strategies
- Components and architecture of an RL agent
- Common challenges and problems in reinforcement learning
Unit 2 Bandit Algorithms and Dynamic Programming 9 topics
- Multi‑armed bandit algorithms: Upper Confidence Bound (UCB), Probably Approximately Correct (PAC), Median Elimination
- Policy gradient methods for bandits
- Full reinforcement learning and $MDP$ formulation
- Bellman optimality principle
- Dynamic programming techniques: value iteration, policy iteration, and policy evaluation
- Extensions of DP and convergence analysis using contraction mapping
- Monte‑Carlo (MC) learning
- Temporal‑Difference (TD) learning
- TD‑$\lambda$ and eligibility traces
Unit 3 Q‑Learning, Function Approximation and Deep Q‑Networks 7 topics
- Q‑learning algorithm and its relation to temporal‑difference methods
- Eligibility traces in Q‑learning
- Function approximation for value functions: linear, non‑linear, and least‑squares methods
- Incremental vs. batch update methods
- Deep Q‑Learning analysis and stability considerations
- Deep Q‑Network (DQN) architecture
- Experience replay mechanisms
Unit 4 Advanced Q‑Learning, Policy Gradient, Hierarchical and POMDPs 7 topics
- Fitted Q‑iteration and its applications
- Advanced Q‑learning variants and handling deep Q‑learning problems
- Imitation learning from optimal controllers
- Integration of DQN with policy‑gradient approaches
- Policy‑gradient algorithms for full reinforcement learning
- Hierarchical reinforcement learning concepts
- Partially observable Markov decision processes (POMDPs)
Unit 5 Actor‑Critic, Inverse RL and Current Trends 5 topics
- Actor‑Critic methods and their variants
- Inverse reinforcement learning (IRL)
- Maximum entropy deep inverse reinforcement learning
- Generative adversarial imitation learning (GAIL)
- Recent trends in reinforcement learning architectures