Hands-on Tutorial on Tabular and Time Series Foundation Models

This tutorial introduces the emerging class of foundation models for structured data—spanning both tabular and time series domains. We will explore how large-scale pretraining on synthetic and real-world datasets enables models such as Mitra, TabPFN, and Chronos to generalize across tasks, achieving strong zero-shot and few-shot performance without traditional retraining. Through hands-on examples, participants will learn how to apply these models using the AutoGluon ecosystems, compare them with classical AutoML and deep learning baselines, and understand their design principles, limitations, and opportunities for research and real-world deployment.