Deep Learning for Time Series (DL4TS)

Overview

Machine learning models often assume that the data used for training and the data seen during deployment come from the same distribution. However, in the real world, this assumption rarely holds. The process of adapting models to perform well despite such shifts is known as domain adaptation. In the context of time series, these shifts can arise due to changes in environments, tasks, or sensor configurations, leading to challenges in generalization and robustness. This course explores methods for domain adaptation in time series, with a focus on how machine learning models can generalize across individuals, environments, tasks, and measurement conditions. We will investigate a range of approaches, ranging from classical methods to modern strategies such as parameter-efficient tuning, few-shot learning, and test-time adaptation. Special attention will be given to the unique challenges posed by time series data, such as distributional drift and heterogeneity across individuals, devices, or recording setups. Students will discuss recent literature and develop a hands-on research project applying these methods to real-world time series problems.

Instructional Team

Instructor: Eva Dyer (eva.dyer@seas.upenn.edu)

Teaching Assistants:
Divyansha Lachi (div11@seas.upenn.edu)
Shivashriganesh Mahato (smahato@seas.upenn.edu)

Schedule

This schedule will be updated throughout the semester.