The head of the laboratory, Alexander Gasnikov, spoke at the seminar "Mathematical foundations of Artificial Intelligence" at the V.A. Steklov Mathematical Institute of the Russian Academy of Sciences. The seminar was held within the framework of the cycle «Federal training. Fundamental aspects. Open tasks» in the MIAN conference hall. The meeting was attended by key academics and young specialists in the AI field.

Federated Learning (FL) allows you to train algorithms on multiple devices containing local data samples without exchanging them. Currently, FL is a full—fledged section of stochastic optimization with a significant mathematical component. Despite active research, even in the most basic formulations of federated learning, there are many open questions.

In his speech, Alexander Gasnikov reviews the results of the Woodworth-Srebro group, one of the most active FL groups in the world. The scientist also demonstrates some new results related to the latest achievements. At the end, the fundamental tasks of federal education that still remain unresolved are highlighted.

Photo source: website of the Steklov Mathematical Institute of the Russian Academy of Sciences.