Reinforcement Learning Masterclass

A 24-hour reinforcement-learning masterclass — from Markov decision processes and Q-learning to deep Q-networks, trained in OpenAI Gym and the CARLA self-driving simulator. You learn how agents make sequential decisions and how to train them safely in simulation.

From Markov decision processes and Q-learning to deep Q-networks, you train agents in simulated environments and apply them to control and autonomous-driving scenarios. The capstone trains a control or driving agent.

• ML practitioners exploring decision-making and control.
• Robotics, autonomous-systems and simulation engineers.

• Model decision problems as Markov decision processes.
• Train agents with Q-learning and deep Q-networks.
• Use OpenAI Gym to build and test environments.
• Apply RL to control and self-driving with CARLA.
• Evaluate agent policies and reward design.

• Hands-on labs throughout the program (70% project-based).
• A capstone project demonstrating end-to-end skills.

• Train agents with Q-learning and DQN in simulated environments.
• Apply RL to control and autonomous-driving scenarios.

Globally accredited certificate issued by Epsilon AI Learning — USA, with a unique Certificate ID and EPSILON ID. Awarded on 80% attendance, 80% final exam, and a capstone project.

• Agents, environments and rewards
• Exploration vs. exploitation
• Markov decision processes (MDP)

• Q-learning and value iteration
• Policies and the Bellman equation
• Tuning learning parameters

• Function approximation with DQN
• Experience replay and target networks
• Stability and common pitfalls

• Building environments in OpenAI Gym
• Reward shaping
• Evaluating agent policies

• The CARLA simulator
• Control and driving tasks
• Safety in simulation

• Defining a control problem
• Training and evaluating the agent
• Analyzing behavior