Sberbished first place in the SHIFTS CHALLENGE International Competition for Machine Training Specialists in the Nomination “VEHICLE MOTION PREDICTION). The results of the competition were voiced at the prestigious conference Neurips.
For comparison of the solutions of the contestants, a dataset of 1600 hours of movement was used, divided into 600,000 marked trail fragments on the roads of Russia, Israel and the United States in good weather, snow and rain. Based on this data, it is possible to evaluate how machine learning models are coping with unfamiliar situations. Participants in Shifts Challenge trained their algorithms on the data provided, and then checked the quality of their work on new data. The decision from the team of Sberbank – Alexander Gamayunov and Alexey Postnikov – was recognized as the best.
“Specialists of the Laboratory of Robotics Solve developed a solution that with a large margin ahead of competitors’ solutions for all major metrics. And this work is important for the development of self-governing road transport in Russia as a whole. Now we are already working on the possibility of checking this algorithm on our subsidiary’s data. Sberautotech. Congratulations to colleagues with this victory, which, I am sure, will speed up the launch of the safe, mobile and environmentally friendly autonomous transport of the future in our country, “said Stanislav Kuznetsov, Deputy Chairman of the Board of Sberbank.
Shifts Challenge Competition is organized by Yandex together with scientists from Oxford and Cambridge to draw attention to the problem of “data shift” in various machine learning tasks. Participants competed in three nominations: weather prediction, machine translation and prediction of vehicle movement.