On April 16, the poster session “Scientific Breakthrough of the Year in AI“ was held at the Data Fusion 2025 conference. Skoltech was represented by five researchers with four papers:
— Alexander Tyurin (Senior Lecturer at the AI Center, Head of the AI Group) — On the Optimal Time Complexes in Decentralized Stochastic Asynchronous Optimization. The article discusses a class of algorithms for decentralized asynchronous optimization. The authors theoretically established the best possible estimates of the running time of such algorithms and proposed methods that reach these theoretical limits. Work on the same day was marked the Data Fusion 2025 Award.
— Stepan Vasiliev (4th year postgraduate student, scientific supervisor - Oleg Khamisov, Senior Lecturer at the Center for Energy Technologies) — A Data‑Driven Approach for Managing Renewable Energy Systems During Short‑Term Voltage Instability. The method based on the Lyapunov exponent and clusterization diagnoses and mitigates short-term voltage surges in renewable energy systems without installing additional equipment.
— Alexander Korotin (Senior Lecturer, Group Leader at the AI Center) — Light Schrodinger Bridge Solver A “lightweight“ CPU‑friendly algorithm for the Schrodinger‑Bridge problem is proposed: one simple optimization goal, a minimum of hyperparameters, and theoretical versatility.
— Valery Shevchenko and Vladimir Zholobov (research engineers at the AI Center) — From Variability to Stability: Advancing RecSys Benchmarking Practices. A new benchmark for recommendation systems: 30 datasets (two new samples), 11 algorithms, 9 metrics — gives comparable results regardless of the domain and data preprocessing.
The session brought together research on optimization, grid sustainability, generative models, and an objective assessment of RecSys, concluding the first day of the conference.
