For a typical galaxy like the Milky Way, roughly one third of all starlight is re-processed through

cosmic dust. The only way to directly observe and measure the interstellar dust content of galaxies

uses observations at far-infrared and sub-millimeter (FIR/submm) wavelengths. Unfortunately,

there are now no FIR/submm missions operational or approved. In principle, however, it should be

possible to predict the FIR/submm emission from a galaxy, if we have UV, optical and nearinfrared

imaging data at hand.

We propose to develop a framework based on supervised machine learning techniques to predict

the FIR/submm emission of galaxies from UV to NIR fluxes. We will train the algorithm with

available state-of-the-art multi-wavelength data sets, both on global and on local (~100 pc) scales.

We will use our framework to investigate the physical properties that drive the shape of the

FIR/submm spectral energy distribution, and to investigate the influence of environment and

morphology on the FIR/submm properties of galaxies. We will also construct a FIR/submm atlas of

about 100 large nearby galaxies, at an angular resolution that cannot be obtained observationally,

and use these images to test more complex radiative transfer modelling techniques.