Machine Learning: What is it?
- Updated
On this page
Learning patterns that must survive new cases
Established · Overview reviewed 2026-11-03
A learning algorithm fits a model using data. The real challenge is whether the model works on relevant observations it did not train on. Training success and practical usefulness are different measurements; data leakage can make a weak system appear strong.
Common pitfalls and recommended practices — scikit-learn developers · Checked
Early machines learn a simple distinction
Demonstrated result · July 1958 demonstration
Cornell’s historical account describes Rosenblatt’s perceptron learning to distinguish marked cards in a 1958 demonstration. It was an early example of adjusting a system from observations. The limited task and later setbacks show why the origins of machine learning should not be confused with the arrival of general intelligence.
Rosenblatt’s perceptron and its historical limits — Cornell Chronicle · Source date: 2019-09-25 · Checked
Accessible tools and larger learned representations
Established · Selected evolution: 2011–2017
The scikit-learn paper described a consistent Python interface for supervised and unsupervised methods. The ImageNet and Transformer papers then demonstrated influential neural architectures on defined tasks. This is a selected path through a much older field, not a complete origin story.
Scikit-learn: Machine Learning in Python — Pedregosa and colleagues / JMLR · Source date: 2011 · Checked
ImageNet Classification with Deep Convolutional Neural Networks — Krizhevsky, Sutskever and Hinton / NeurIPS · Source date: 2012 · Checked
Attention Is All You Need — Vaswani and colleagues / arXiv · Source date: 2017-06-12 · Checked