Summary
Highlights
Machine learning is about making a robot or machine behave like a human, learning from experience, and even excelling at tasks. An example is a robot learning to cross the road only when a traffic light is green, avoiding cars.
Instead of direct experience, a machine can learn from collected data. If someone documents when a duck survived crossing the road based on the traffic light color, this data can be used to train the machine. This trained machine is called a 'model'.
Unsupervised learning is when the machine learns without explicit guidance on what's right or wrong. 'Clustering' groups similar data points (like ducks surviving red vs. green lights). 'Association' finds relationships between data points, such as pregnancy and having a partner.
In supervised learning, the machine makes a decision, and then it is told whether that decision was correct or incorrect. This feedback helps the machine learn over time, similar to a child learning from a parent.
Semi-supervised learning combines elements of both, using both labeled and unlabeled data. For example, a machine categorizing flowers might use some photos with known colors and infer the rest.
There are generally two types of models. 'Classification' models differentiate between categories (e.g., 'car' or 'not car'). 'Regression' models predict continuous values, like estimating how many clicks an email might get based on the number sent.
Besides classification and regression, and supervised and unsupervised learning, other machine learning algorithms exist, such as reinforcement learning, decision trees, and random forests, expanding the possibilities of AI.