Agents learn by trial and reward.
An agent learns a policy by interacting with an environment and receiving rewards, balancing exploration and exploitation. Methods: Q-learning, policy gradients, actor-critic, PPO. Applications: games (AlphaGo), robotics, resource management, RLHF. Sample efficiency and safe exploration are challenges.
A robot learns to walk by trial and error in simulation.
Reinforcement learning trains agents by reward, so it suits control problems where the best behaviour must be discovered by trial.
Agents learn by trial and reward.
Reinforcement learning trains agents by reward, so it suits control problems where the best behaviour must be discovered by trial.
A number the environment gives to the agent after an action, which the agent tries to maximise over time.
It needs many trials, which can be unsafe or expensive in the real world, so simulation and careful reward design are important.
Typical applications include training a robot to walk by rewarding forward progress, optimising the control of a process plant through simulation and learning a strategy for scheduling energy storage under changing prices.
Study "Computer vision" first (topic 99 of 104 in Software & Computing). It covers the groundwork this topic builds on.
Markov decision process. Discount factor. Reward design.
A robot learns to walk by trial and error in simulation.
The next topic is "MLOps and ethics", which builds directly on this one.
It is classed as Advanced in Software & Computing (topic 100 of 104). It is an advanced topic. It assumes you are comfortable with the intermediate topics in Software & Computing.
Want to practise? Open the interactive course for quick checks, quizzes, simulations and progress tracking.