Support vector machine (SVM)
Máquina de vectores de soporte (SVM)
- Date
- 1992 · confidence medium
- Place
- AT&T Bell Laboratories, Holmdel, New Jersey, United States
- Credited to
- Bernhard Boser, Isabelle Guyon and Vladimir Vapnik
- Field
- Knowledge & symbols
⚠️ Disputed: the sources disagree.
What it is
Boser, Guyon and Vapnik applied the "kernel trick" to the maximum-margin linear classifier that Vapnik and Chervonenkis had devised in Moscow in 1964, producing a machine able to draw nonlinear decision boundaries; Cortes and Vapnik's "soft margin" version (1995) made it practical for noisy or non-separable data.
Wikidata dates the item to 1963, the year of Vapnik and Chervonenkis's original linear algorithm (the "generalized portrait") in the USSR; what is understood today as an SVM, with nonlinear kernels, is the 1992 version, extended with the soft margin in 1995.
What it led to
- Widespread adoption of SVMs in text classification, bioinformatics and computer vision through the 1990s-2000s
- Systematic study of kernel functions in machine learning