The talk is devoted to research in Yandex related to machine intelligence. We develop new technologies, which strongly improve products used in everyday life. We study key theoretical properties of new algorithms, their capabilities and constraints. The novelty of our research leads to papers at the major world conferences on Machine learning, Computer Vision, NLP, Web Mining, Computational Economics, and other topics.
As a particular example of our results, I will describe the key algorithmic techniques behind CatBoost, a new gradient boosting toolkit. Their combination leads to CatBoost outperforming other publicly available boosting implementations on a variety of datasets.