Machine learning is a branch of artificial intelligence that allows computers to learn from data and improve their performance on a task without being explicitly programmed for each particular case. In other words: instead of writing manual rules («if the temperature > 30°C and humidity > 80% then probable storm»), the system is given many past examples and it discovers itself the patterns that allow to predict or decide. This is the approach that has contributed to the explosion of modern AIs since ~2012.
Machine learning consists of:
- analyze data,
- detect patterns,
- build predictive models,
- improve its performance with experience.
Example:
- Netflix recommendation
- Detection of bank fraud
- Image recognition
There are 4 main families of Machine Learning.
- Supervised learning
In this type of learning, the model learns from data labeled (with a known response). The objective is to predict and classify.
Example of application: spam detection, real estate price prediction,...
- Unsupervised learning
Here the data has no labels, the model discovers structures. It is often used in problems of segmentation and pattern detection.
Example of application: customer segmentation in marketing, fraud detection, anomaly detection (bank fraud), ...
- Semi-supervised learning
Here we make a mix between using few labeled data and many unlabeled data. It is used when labeling is expensive or if data is scarce.
- Reinforcement learning
In this type of learning, an agent learns through a system of rewards / punishments
Example of application: Humanoid robots that learn to walk, real-time logistics optimization, high frequency trading