Event Date
Abstract
Time series forecasting plays a pivotal role in automating and optimizing decision-making across business and scientific domains. In recent years, the field has undergone a paradigm shift—from model- and assumption-driven approaches toward data-driven, scalable, and automated learning systems. This transformation is fueled by the growing availability of large, heterogeneous time series data and the computational advances that make it possible to learn directly from them. In this talk, we will trace the evolution of probabilistic forecasting of the last decade, beginning with deep generative models that unify statistical rigor with neural expressivity. We will then discuss automated machine learning (AutoML) frameworks for time series, which democratize model development and enable consistent performance across diverse datasets. Finally, we will explore the emerging wave of time-series foundation models — large pretrained models that generalize across domains, tasks, and horizons—highlighting their architectural principles, training paradigms, and their potential to redefine how forecasting systems are built and deployed at scale.
Bio
Bernie Wang is a Principal Machine Learning Scientist at AWS, where he leads research on automated machine learning (AutoML), large-scale time series forecasting, and foundation models for structured and unstructured data (LLMs). His work focuses on democratizing advanced AI/ML capabilities and developing scalable, efficient algorithms that empower practitioners across diverse domains. Bernie leads the team behind several widely adopted open-source libraries, including AutoGluon, GluonTS, and Chronos, which have become foundational tools for automated ML and time series analysis. Bernie holds a PhD in Computer Science from Tufts University and a master's degree from Tsinghua University. His research interests span statistical machine learning, numerical linear algebra, and random matrix theory.