This seminar will explore how to learn meaningful, generalizable, and transferable representations from physical sensing and scientific measurement modalities, including radar, sonar, LiDAR, cryo-EM, and beyond. We will examine how these modalities acquire and encode information, the physical principles governing their measurements, and the challenges they pose for representation learning. We will critically examine the assumptions of foundational representation learning methods underlying assumptions and investigate when and why techniques developed for natural images, language, and conventional audio may fail to generalize.We will take a deep dive into how measurement physics can inform the design of representations, tokenization strategies, model architectures, data augmentations, and learning objectives.