Ekaterina Filimoshina presented a talk at the International Conference on Machine Learning (ICML 2025) in Vancouver, Canada
Research intern of the Laboratory for Geometric Algebra and Applications Ekaterina Filimoshina took part in the International Conference on Machine Learning (ICML 2025). This is one of the largest and most significant conferences in computer science. At HSE University, the ICML conference is included in the list of A* conferences (A-CONF), participation in which is equivalent to publication in Q1-journals. The ICML conference has been held annually since 1980 (ICML 1980 Pittsburgh, United States).
Conference website: https://icml.cc/
The ICML 2025 conference was held at the Vancouver Convention Center in Vancouver, Canada, from July 13 to 19. This year, 12,107 applications were submitted to ICML, which is 28% more than last year. Of these applications, 3,260 were accepted, i.e. the acceptance rate was 26.9%. The ICML conference includes several formats: oral presentations, workshops, tutorials, and poster sessions. The conference included 6 large 2.5-hour poster sessions, during each of which about 500 posters were presented.
Conference program: https://icml.cc/virtual/2025/calendar
Ekaterina presented a poster talk at ICML 2025:
Talk: Ekaterina Filimoshina & Dmitry Shirokov, 'GLGENN: A Novel Parameter-Light Equivariant Neural Networks Architecture Based on Clifford Geometric Algebras', July 16, 2025.
Abstract: We propose, implement, and compare with competitors a new architecture of equivariant neural networks based on geometric (Clifford) algebras: Generalized Lipschitz Group Equivariant Neural Networks (GLGENN). These networks are equivariant to all pseudo-orthogonal transformations, including rotations and reflections, of a vector space with any non-degenerate or degenerate symmetric bilinear form. We propose a weight-sharing parametrization technique that takes into account the fundamental structures and operations of geometric algebras. Due to this technique, GLGENN architecture is parameter-light and has less tendency to overfitting than baseline equivariant models. GLGENN outperforms or matches competitors on several benchmarking equivariant tasks, including estimation of an equivariant function and a convex hull experiment, while using significantly fewer optimizable parameters.
Text of the work: https://openreview.net/forum?id=H0ySAzwu8k¬eId=8xBQcGC7zC
The work will be published in the Proceedings of the 42nd International Conference on Machine Learning, Vancouver, Canada, PMLR 267, 2025. The work was carried out as part of the project 'HSE University Mirror Laboratories: 'Quaternions, Geometric Algebras and Applications'.
