Reinforcement Learning – Short Notes | TechAmbitionX

Reinforcement Learning (RL) – Exam Oriented Short Notes

Platform: TechAmbitionX

---

1. Definition

Reinforcement Learning is a type of machine learning in which an agent learns optimal behavior by interacting with an environment and maximizing cumulative reward.

2. Key Components

• Agent: Learner or decision-maker
• Environment: External system the agent interacts with
• State (S): Current situation of the agent
• Action (A): Possible moves by the agent
• Reward (R): Feedback from environment
• Policy (π): Strategy followed by the agent

3. Working of Reinforcement Learning

Observe State → Take Action → Receive Reward → Move to New State → Update Policy → Repeat

4. Reward Concept

• Positive reward → Encourages action
• Negative reward → Discourages action
• Goal is to maximize total reward over time

5. Exploration vs Exploitation

• Exploration: Trying new actions
• Exploitation: Using known best actions
• RL must balance both

6. Markov Decision Process (MDP)

Reinforcement Learning problems are modeled using MDP.

MDP = (S, A, R, P, γ)

• S → Set of states
• A → Set of actions
• R → Reward function
• P → Transition probability
• γ → Discount factor

7. Value Functions

• V(s): Value of being in state s
• Q(s,a): Value of taking action a in state s

8. Q-Learning

Q-Learning is a model-free reinforcement learning algorithm that learns the optimal action-value function.

Q-Learning Update Equation

Q(s, a) = Q(s, a) + α [ r + γ max Q(s′, a′) − Q(s, a) ]

Where:

• Q(s, a) → Current Q-value
• α (alpha) → Learning speed
• r → Immediate reward after taking an action
• γ (gamma) → Discount factor (The discount factor γ controls the importance of future rewards in reinforcement learning)
• max Q(s′, a′) → Maximum future Q-value
• s′ → Next state

9. Types of Reinforcement Learning

• Model-Free RL: Q-Learning, SARSA
• Model-Based RL: Learns environment model
• Deep RL: Uses neural networks (DQN)

10. Applications of Reinforcement Learning

• Some Games
• Robotics
• Self-driving cars
• Resource scheduling

11. Advantages

• No labeled data required
• Learns optimal behavior
• Suitable for dynamic environments

12. Limitations

• Requires large training time
• Reward design is difficult
• Computationally expensive

13. RL vs Other Learning Types

• Supervised: Uses labeled data
• Unsupervised: Finds hidden patterns
• Reinforcement: Uses reward signals

---

Focus: Exams • Concepts • Quick Revision