#15 · ML / AI methods

Simulation-based inference tackles intractable likelihoods in epidemic and phylodynamic…

The dominant thread is parameter inference for mechanistic epidemic models whose likelihoods are difficult to evaluate. Simulation-based routes appear in several items: an ABC-SMC tutorial in PyMC for ecological and epidemiological models, and neural posterior estimation applied to Ebola epidemiological and phylodynamic models with online tutorials. A comparison pits particle MCMC against conditional normalizing flows for stochastic SIS, SIR and SEIR models. Other items apply learning or statistical frameworks to new data structures: a semi-supervised deep network estimating COVID-19 case counts from wastewater viral RNA, a phylogeny-free framework profiling mutation dynamics across 6.6M SARS-CoV-2 genomes, and a perturbation-augmented neural ODE for gut microbiome dynamics. What is new is the pairing of specific data structures with specific inference or modeling tools, a pattern not shared by every item.