Yandex employees won the competition for the best search

In 2009, Yandex held another contest " Internet Mathematics 2009 ". The task proposed for the decision at the competition was to obtain the function of ranking documents based on the training set (queries and relevant documents with relevant relevance ratings).

Then we did not even realize that this competition would cause a wide response and get world renown, and that Yahoo! create your own competition - in the image and likeness of ours. There are two competitions (tracks), for which Yahoo! provided two data sets (large and small), which in the real system are used for ranking training.

The conditions of such contests certainly differ from actual conditions - for example, because the resulting ranking function may be unacceptably complex and calculated too long for the search engine that live people are looking for. Nevertheless, this competition is very interesting, because it uses the most advanced technology and the most powerful teams of researchers participate. It is also interesting because it makes it possible to compare different technologies, different ranking algorithms, including completely new and closed ones. And of course, as in any competition, there is a considerable element of luck in it.

We are pleased to see among the winners of the Yahoo! Yandex employees. In both lanes there are a lot of them. And Igor Kuralyonok and Andrey Gulin even won first place in the second track!
')

It seems to us that this confirms that, firstly, we have excellent search specialists, and secondly, the tools and algorithms that they use (Igor and Andrey, for example, used the MatrixNet research version, - the highest world level).

Igor, Andrey, congratulations to you!

Ilya Segalovich, fan

For those interested, lay out the presentations of our specialists at a seminar dedicated to this task at ICML - the main conference on machine learning, which has just ended in the city of Haifa, Israel:

BagBoo: Bagging the Gradient Boosting (Dmitry Pavlov and Cliff Brunk aka JOKER & team_404)

YetiRank: Everybody Lies (Igor Kuralenok and Andrey Gulin aka MN-U)
Application of Additive Groves to the Learning To Rank Challenge (Daria Sorokina aka AG)