Colombin

Foundation models are changing the landscape of artificial intelligence by enabling a single pretrained model to adapt to a wide range of tasks. This presentation explores whether a similar paradigm can be applied for human mobility, where data are inherently heterogeneous and irregular. I will first introduce the concept of human mobility foundation models and discuss the main approaches currently emerging in the field. I will then present an ongoing survey and comparative evaluation of existing models, focusing on the challenges of establishing a common and reproducible testing methodology. Finally, I will present my own research direction, focusing on space-invariant tokenization, and discuss future directions for developing general-purpose models for human mobility data.

Biography:

Dr. Jacopo Joy Colombin is a researcher from Scuola Normale Superiore di Pisa whose work focuses on foundation models for human mobility. His research explores how general-purpose pretrained models can be developed for heterogeneous and irregular mobility data, with particular interests in model evaluation, reproducible benchmarking, and space-invariant tokenization. His current work aims to advance more general and transferable modeling approaches for human mobility data.