Finance
Toss paper accepted to NeurIPS 2025 on federated learning optimisation
Toss said its federated learning paper was accepted to NeurIPS 2025, highlighting joint work with Seoul National University. The research proposes a method that addresses performance drops caused by differing or unseen data types across countries or user groups. It combines Infomap-based clustering with local prior alignment to improve learning stability and supports discovery of new categories in environments where data distribution is unknown. Toss said the acceptance marks its first formal recognition at a global AI conference.
