# AI Major Techniques
- Unsupervised learning
- Clustering
- Identify groups/sub-groups (clusters) in a dataset
- The algorithm receives data, and has to find interesting "facts" about it
- Major issue: it requires a ton of labeled data (much more than human childs need to learn)
- Unsupervised learning has applications for various things, including NLP (e.g., improving Web search)
- Transfer Learning (TL)
- Learn from a task A (e.g., car detection), and help on another task B
- Very valuable technique, used for instance by computer vision systems
- Reinforcement Learning (RL)
- Similar to how you train a pet dog to behave
- Let the dog whatever it wants to do
- Whenever he does something good, you praise him
- Whenever he does something bad, you go "bad dog!"
- Over time the dog improves his behavior
- Use a "reward signal" when it's doing well or poorly. Large positive number when positive and large negative number when performing poorly. The goal being to maximize positive rewards
- Use for games such as Go, Chess, videogames, etc
- One downside is that it can require a huge amount of data
- Generative Adversarial Networks (GANs)
- Created by Ian Goodfellow, a former student of Andrew Ng
- Example: synthetize new images from scratch
- Knowledge Graph (KG)
- Huge economic potential and impact
- Capture data (different dimensions) about things/concepts/people
- Connect the data together to create knowledge graphs