Совместный семинар HDI Lab & TFAIM Lab "Towards Better Tabular Representations: Generative Modeling and Distribution Shift"

This talk focuses on tabular data and examines it through the lens of different representation paradigms. We first consider subset-based representations and meta-features, which capture structural properties of datasets, and then move to distribution-level representations based on generative modeling, including Schrödinger bridges. These approaches are illustrated through key applications such as synthetic data generation, out-of-distribution (OOD) testing, and OOD detection. The talk highlights how choosing appropriate representations enables more robust, interpretable, and effective tabular machine learning systems. The presentation is based on the following papers:

On the Use of Schrödinger Bridges for Tabular Data Generation
Meta-Features Informed WGAN for Tabular Data
Evaluating Robustness of Tabular Models under Meta-Features Based Shifts

Семинар будет проводиться на английском языке

По всем вопросам обращайтесь к Зеленовой Карине Михайловне kzelenova@hse.ru или к Горностаевой Екатерине Дмитриевне egornostaeva@hse.ru