Machine Learning

Machine learning is the art of building algorithms that improve with experience. Instead of hand-coding every rule, you give the algorithm examples and let it derive the rules itself.

The three families

• Supervised learning — you give the model labelled examples ("this is a cat", "this is a dog") and it learns the mapping. The bulk of modern ML, including image classifiers and spam filters.
• Unsupervised learning — no labels. The model finds structure on its own — clusters of customers, anomalies, latent topics in text.
• Reinforcement learning — the model takes actions, sees a reward signal, and learns the strategy that maximises long-term reward. See reinforcement learning.

ML inside robotics

• Object detection in computer-vision pipelines
• Speech recognition (Pepper, Mitra)
• Anomaly detection (industrial fault prediction)
• Manipulation policies (Tesla Optimus, DeepMind's robotic arm)
• Path planning under uncertainty

ML is built on neural networks — particularly deep neural networks (see deep learning) — for the most powerful applications. Image-focused work uses computer vision techniques.