Machine Learning — How Machines Learn From Data, Explained

Overview

This video explains machine learning as the part of AI that enables software to learn patterns from data rather than hand-written rules. It covers how machines learn, the three main families of machine learning, the role of neural networks, and the common pitfall of overfitting.

In this video

• Machine learning allows software to learn patterns from data instead of following rules written by humans.
• At its core, machine learning is a program that improves at a task through experience, where experience means data, not extra code.
• The term machine learning dates to 1959, when a researcher taught a computer to play checkers better than he could.
• Machine learning systems discover patterns by being shown thousands of examples, mapping an input to an output.
• Data is split into a training set for the model to learn from and a test set for honest evaluation of its performance.
• Learning is an optimization process where adjustable numbers called weights are nudged by algorithms like gradient descent to shrink the model's error.
• A humbling rule of thumb in machine learning is that more and cleaner data usually beats a cleverer algorithm.
• Machine learning splits into three families: supervised learning, unsupervised learning, and reinforcement learning.
• Neural networks are the engine of the deep-learning era, consisting of layers of simple units with weighted connections, trained using backpropagation.
• Overfitting is a common pitfall where a model memorizes its training data but performs poorly on new, unseen data, which is fought with more data and simpler models.

Frequently asked questions

What is machine learning?

Machine learning is the part of AI that allows software to learn patterns directly from data instead of relying on rules coded by humans. It's a program that gets better at a task through experience, using data as its learning source.

What are the three main types of machine learning?

Machine learning divides into three families: supervised learning, unsupervised learning, and reinforcement learning. Most everyday business applications of machine learning fall under supervised learning.

How do neural networks contribute to machine learning?

Neural networks are the engine of the deep-learning era, consisting of stacked layers of simple units with weighted connections. They transform data incrementally and are trained using backpropagation to adjust weights and reduce error.

What is overfitting in machine learning and how is it avoided?

Overfitting is a common pitfall where a model memorizes its training data perfectly but fails to perform well on new, unseen data. It is fought with more data, simpler models, a technique called regularization, and judging performance on a held-out test set.

Where is machine learning used today?

Machine learning is used almost everywhere, including streaming app recommendations, bank fraud detection, medical imaging support for doctors, demand forecasting, language translation, and the perception systems in self-driving cars.