Glossary
Who was Andrey Kolmogorov
In short
Andrey Kolmogorov was a mathematician whose 1957 theorem on representing functions of many variables became the base of KAN neural networks, introduced in 2024.
People · Andrey Kolmogorov
MMOOZZGG.com editorsChecked: 1 min read
For a child
Imagine a hard pie recipe, where the taste depends at once on flour, sugar, eggs and the heat of the oven. Andrey Kolmogorov proved that even a very tangled rule can be split into small simple steps, and each step looks at only one thing. Then the steps are put together, and you get the same rule. Many years later, people used this idea to build new programs that learn from examples.
In plain words
Kolmogorov worked in probability theory, mathematical analysis and information theory. In 1957 he proved a surprising thing: any continuous function of many variables can be built from functions of one variable and ordinary addition. How an apartment’s price depends on its area, floor and district can, in principle, be split into simple curves for each feature, their sums and one more layer of simple curves. This result is known as the Kolmogorov-Arnold theorem, and in 2024 KAN neural networks were built on it. In 1965 Kolmogorov also proposed measuring the information in an object by the length of the program that describes it.
For business
For business, Kolmogorov’s theorem matters through KAN networks, which promise models you can read like a formula. An ordinary neural network gives a forecast of demand or customer churn but does not explain it. A KAN learns a separate simple function on each link, and an analyst can see exactly how price, season or region affects the result. This matters where a decision must be justified to a bank, an auditor or a board of directors. The risk is that the architecture is young: there are fewer proven industrial tools than for familiar models, and the gain on your data is unknown in advance. So such models are compared with ordinary ones on one forecasting task before the workflow changes.
For a specialist
A. N. Kolmogorov’s paper “On the representation of continuous functions of several variables as superpositions of continuous functions of one variable and addition” appeared in Doklady Akademii Nauk SSSR (vol. 114, no. 5, pp. 953–956, 1957). The result, known as the Kolmogorov-Arnold representation theorem, writes a continuous function of many variables on a cube as a sum of outer functions of one argument applied to sums of inner functions of one argument. The paper by Liu, Wang, Vaidya, Ruehle, Halverson, Soljačić, Hou and Tegmark, “KAN: Kolmogorov-Arnold Networks” (arXiv 2404.19756, 30 April 2024), carries this scheme into neural networks. In a multilayer perceptron (MLP), fixed activation functions sit in the nodes; in a KAN, trainable functions sit on the edges, there are no linear weights at all, and each function is a spline. The authors report gains in accuracy and interpretability on approximation tasks and on solving partial differential equations. The 1965 paper “Three approaches to the definition of the concept ‘quantity of information’” in Problemy Peredachi Informatsii introduces the algorithmic approach, the base of Kolmogorov complexity.
Example
An insurance company builds a model of the probability of an insured event. Instead of a black box, the analysts try a KAN network based on Kolmogorov’s idea. The model shows a separate curve for age, driving experience and region, and these curves show why a client got a given rate.
Related terms
Questions and answers
Sources
- A. N. Kolmogorov, On the representation of continuous functions of several variables as superpositions of continuous functions of one variable and addition, Doklady Akademii Nauk SSSR, 1957mathnet.ru
- A. N. Kolmogorov, Three approaches to the definition of the concept "quantity of information", Problemy Peredachi Informatsii, 1965mathnet.ru
- Liu, Wang, Vaidya et al., KAN: Kolmogorov-Arnold Networks, arXiv 2404.19756, 2024arxiv.org
Not sure whether your company needs this?
AI solutions map