Yandex company blog

In Memory of Alexey Yakovlevich Chervonenkis

29 September 2014, 18:31

Alexey Yakovlevich Chervonenkis tragically died on September 22. A professor of the Moscow Institute of Physics and Technology as well as Royal Holloway, University of London, and a lecturer at the Yandex School of Data Analysis, he made a huge contribution to the theory of machine learning.

chervonenkis.jpg

 

So far there have been three periods in the science of machine learning: pre-computer, computational, and the contemporary period of big data.

The first great work of Chervonenkis and Vapnik was this article from 1971. The theory of the uniform convergence of the frequencies of occurrence of events to their probabilities set the course for the development of this field of science for several decades ahead.

This was the period of the “theoretical” development of machine learning. At that time, only some kind of M-200 or, at best, a BESM was available for computing so there was not yet even any talk of widespread 'practical application in the nation's industry'. But even then it could already be used to find targets in the air, for example, or to help detect abnormalities in echocardiograms.

Then came the second period in the history of machine learning – the computational stage. In the 1990s people learned, for example, how to quite effectively recognise and digitise texts (including handwritten documents) and keep e-mail free of spam. Half of these methods worked on the renowned SVM (Support Vector Machine) method conceived in the early 1990s by Chervonenkis and Vapnik (Vapnik-Chervonenkis dimension). In the mid-2000s all the well-known companies worked on SVMs – including us, and Yahoo!, and Google, and Amazon. SVM is described in any textbook on the subject.

And then came the third era in the development of machine learning, with the appearance of big data and methods for working with it. Now it appears that everything around us, all objects and services, will become a bit smarter and learn to help us in every detail, anticipating our desires to some extent. This is similar to how various mechanical and chemical inventions have changed our lives, only now in a slightly different sphere.

In this third era, Chervonenkis taught at the Yandex School of Data Analysis, and presented the development of his fundamental 1971 work at our conference.

Alexey Chervonenkis loved to walk. He would walk 20 kilometres a day – around Moscow, or London, or forests – that’s how he thought. In summer he had an operation and he couldn’t walk for three weeks. Then one day he started walking again – first a kilometre, then two, then three. And last week he set off on a 20-kilometre walk along a familiar route through Losiny Ostrov National Park.

Arkady Volozh