Forecasting big time series: theory and practice

During The Web Conference in April, Amazon scientists and scholars joined external researchers, policy makers, developers and others for an all-virtual conference to discuss the evolution of the Web, the standardization of its associated technologies, and the impact of those technologies on society and culture.

One component of the event: a tutorial of time series forecasting, a key ingredient in the automation and optimization of business processes, by scientists Yuyang (Bernie) Wang, Valentin Flunkert, Jan Gasthaus, and Tim Januschowski, and Amazon Scholar Christos Faloutsos.

”Some of the world’s most challenging forecasting problems can be found inside Amazon or are posed to us by AWS customers,” says Januschowski. “Forecasting can estimate future demand in different regions to help retail businesses decide which products to order and where to store them; in cloud computing, the estimated future usage of services and infrastructure components guides capacity planning; and workforce scheduling in warehouses and factories requires forecasts of the future workload. Amazon scientists are constantly inventing new methods on customers’ behalf using deep learning and probabilistic methods to address these problems better than before.”

Watch the video below for an overview of the most important methods and tools available for solving large-scale forecasting problems. The presenters review the state of the art in three related fields:

1. Classical modeling of time series;
2. Modern methods including tensor analysis and deep learning for forecasting; and
3. The tools and practical aspects of building a large scale forecasting system.

Part 1: Forecasting fundamentals

Part 2: Deep learning for forecasting

Part 3: Forecasting in practice

System, practice, and the AWS forecasting stack | Forecasting big time-series | Amazon Science