Benchmarking the Continuous Gromov-Wasserstein Problem: Towards Reliable Solvers with NeuralGW
18:35
В этот четверг состоится очередной семинар, но он пройдет на английском языке. Также обращаем внимание: при входе в конференцию необходимо авторизоваться под своим именем и фамилией — участников с ником «Гость» мы не сможем пропустить.
What will the talk cover?
The Gromov-Wasserstein Optimal Transport (GWOT) problem aims to find structural correspondences between distributions without requiring point-to-point alignment. While continuous GWOT holds great potential for machine learning, most current solvers rely on discrete approximations. These methods often fail to generalize, restrict applicability to highly correlated datasets, and struggle with scalability.
In this seminar, we will:
- Review the challenges and limitations of existing GWOT solvers;
- Present a systematic benchmark across diverse experimental settings;
- Introduce NeuralGW, a continuous, gradient-based solver that avoids discrete constraints and better captures intrinsic distributional structure