The field of phylodynamics hypothesises that the way viruses spread leaves footprints in their genetic information. This opens great potential to provide insights into complex epidemiological dynamics, but in practice, there are many hurdles to be overcome. One of them is that the number of parameters to infer from raw DNA or RNA sequence data is huge, making likelihood-based methods challenging for non-trivial models. A promising alternative is a simulation-based approach called ABC for Approximate Bayesian Computation. Its downside is that it is computationally intensive, since thousands (if not millions) of simulation runs to obtain a good quality model fit. To avoid this, people have proposed all sorts of optimization schemes, such as regression ABC, which relies on state-of-the-art machine learning techniques.