Modelling human mobility is fundamental to understanding how people move in spaces and impacts several societal aspects such as urban planning, disease spreading, and others. Even if nowadays we are facing a proliferation of human mobility traces, it is not always possible to access such information. For instance, most open mobility datasets cover a limited number of US, European and, Chinese cities. Notably, multiple companies recently released mobility datasets to counteract the spread of COVID but still, we have relatively poor mobility data for many areas of the world. In this seminar, we show how deep learning can be used to leverage human mobility data and design models that are geographically transferable, i.e., trained on an area of th
e world in which data are available and test the model on areas with a scarcity of even absence of data. After, we describe known methods, we outline challenges and future directions for both individual and collective mobility tasks.
Hybrid seminar, presential in seminar room and streamed via Zoom at https://zoom.us/j/98286706234?pwd=bm1JUFVYcTJkaVl1VU55L0FiWDRIUT09
