Neural Variational Inference for Markov Jump Processes

Masterarbeit, Bachelorarbeit

This thesis investigates the latent state inference and parameter learning ofMarkov Jump Processes from noisy observations. Markov Jump Processes arecontinuous-time stochastic processes with applications in many fields such asbiology, queuing theory and finance. However, for large Markov Jump Processesexact inference of the latent state and parameter learning is intractable. The goalof this thesis is to apply variational inference, a deterministic approximation tothe inference problem. Therefore, we want to approximate the posterior processwith neural networks and learn a good parametrization with stochastic gradientmethods.