Last reviewed:
What is machine learning? Definition and difference from deep learning
Machine learning is the branch of artificial intelligence in which a program learns to perform a task from examples, instead of following rules written by a programmer. It relies on a model whose parameters are automatically adjusted on data until it produces reliable results.
In July 1959, IBM researcher Arthur Samuel published one of the founding papers of the field in the IBM Journal of Research and Development: his checkers program learned to play better than its author, from the rules and its own games alone. France's official definition (Journal officiel, December 9, 2018) describes a “process by which an algorithm evaluates and improves its performance without the intervention of a programmer”, by repeating its execution on datasets. In practice, a model has parameters, often initialised at random, which training adjusts to reduce the gap between its predictions and the expected results. Once trained, the model is put into production and applied to new cases (CNIL). There are three learning modes. Supervised: each example carries the right answer (this customer cancelled, this invoice is fraudulent). Unsupervised: the algorithm finds groupings on its own, for example customer segments. Reinforcement: an agent learns through trial and reward. So-called classic methods (regression, decision trees, random forests, gradient boosting) excel on a company's tabular data: sales, stock, customer history. Deep learning is a subfamily of machine learning, based on neural networks with many layers, better suited to images, sound and text. Generative AI is a recent application of it.
Concrete example
Illustrative case: a French regional garden centre chain with 25 stores wants better sales forecasts to cut unsold plants. It has three years of till data, the promotions calendar and weather records. A gradient boosting model, trained on this tabular data, forecasts sales by store and by week. After two seasons, department managers order based on the forecast and adjust at the margin. No deep learning is needed: the data is structured and modest in volume, and the team wants to understand which variables matter (weather, public holidays, promotions), which a classic model makes readable.
Comparison
| Classic machine learning | Deep learning | Generative AI | |
|---|---|---|---|
| Place within AI | Family of AI that learns from data | Subfamily of machine learning | Application of deep learning |
| Techniques | Regression, decision trees, random forests, gradient boosting | Deep neural networks (CNNs, RNNs, transformers) | Large language models, diffusion models |
| Typical data | Structured tables (sales, customers, sensors) | Images, audio, text, large volumes | Text, images, code; pre-training on massive corpora |
| Data preparation | Manual choice of useful variables | The network extracts features itself | Model already trained, adapted through prompts or documents |
| Output | A prediction or a class | A prediction, a detection, a transcription | New content (text, image, code) |
| Computing cost | Low, a standard server is enough | High, graphics cards (GPUs) | Very high to train; billed per use (tokens) |
FAQ
What is machine learning, simply put?
It is a way of programming by example. Instead of writing every rule, you show an algorithm thousands of cases with the right answer, and it adjusts its own model to reproduce those answers, then applies them to new cases.
What is the difference between machine learning and deep learning?
Deep learning is part of machine learning. It uses neural networks with many layers, which extract useful features from raw data (pixels, sounds, words) on their own. It needs much more data and computing power. Classic machine learning is often enough for tabular data.
What is the difference between machine learning and artificial intelligence?
Artificial intelligence is the overall field. Machine learning is its most widely used family today: the one that learns from data. An AI can also rely on rules written by experts, with no learning at all.
What are examples of machine learning in business?
Sales and stock forecasting, fraud detection, credit scoring, customer churn prediction, product recommendation, predictive maintenance of machines, automatic sorting of emails or invoices.
What are the types of machine learning?
Three main types. Supervised learning, with labelled examples (predicting a price, detecting fraud). Unsupervised learning, with no labels (segmenting customers). Reinforcement learning, through trial and reward (optimising a strategy, controlling a robot).
Does machine learning need a lot of data?
Less than people think for classic methods: a few thousand well-labelled examples are often enough for tabular data. Quality and representativeness matter more than volume. Deep learning, on the other hand, generally needs far more examples.
See also
Further reading
Machine Learning Crash Course, Google for Developers (free introductory course)
Sources
- Some Studies in Machine Learning Using the Game of Checkers, Arthur L. Samuel, IBM Journal of Research and Development, vol. 3, no. 3, July 1959. https://ieeexplore.ieee.org/document/5392560
- Vocabulaire de l'intelligence artificielle, official French definition of “apprentissage automatique”, Journal officiel No. 0285, December 9, 2018. https://www.legifrance.gouv.fr/jorf/id/JORFTEXT000037783813
- Apprentissage automatique (definition), CNIL (French data protection authority). https://www.cnil.fr/fr/definition/apprentissage-automatique