Reinforcement Leaning

Learning what to do to maximise rewards

Difference with Machine learning

Supervised learning
- Uses labelled data set to Generalisation responses to act correctly in new situations that are not in training set
- Not always practical to obtain examples of optimal action which are both correct and representative
Unsupervised learning
- Finding hidden structure in collection of unlabelled data
- Reinforcement learning is about maximising rewards and not discovering structure
- Although knowledge of that could be useful for maximising reward

Links: