Modeling air pollution at a city scale
Please login to view abstract download link
To get accurate predictions of the level of pollution at every point in a city one has to efficiently exploit different data sources: sensors measuring pollutant concentrations, car emissions (estimated from Google traffic images) or the graph structure of the streets of the city. In this talk several models that can produce reliable predictions by leveraging all these different types of information will be presented. Particular focus will be put on effective models of pollution based on reaction-diffusion equations on graphs. It will also be shown how Graph Convolutional Neural Networks can arise as a competing Machine Learning approach to this problem. Finally, a way of combining them all to better profit from each model's advantage will be presented.