Illuminating the Hidden Universe with Physics and AI

Katie Bouman explains how researchers reconstruct scientific images from incomplete and indirect measurements by combining physical models, prior knowledge and computation. Black-hole imaging illustrates how confidence comes from features that remain consistent across many reconstructions of the same observations. Similar inference problems arise in familiar imaging technologies. The lecture examines the usefulness and limitations of simple assumptions and shows how artificial intelligence can explore a wider range of reconstructions while respecting physical constraints. It considers why understanding image formation is essential when interpreting observations at the limits of what instruments can measure.