2020/10/07 by Frank van der Meulen, Moritz Schauer, van der Meulen, Frank +1 · 1 citation
Computer Science · #60J25 #62M20 #Bayesian Modeling and Causal Inference #Computation (stat.CO) #FOS: Computer and information sciences #Gaussian Processes and Bayesian Inference #Machine Learning and Algorithms #Methodology (stat.ME) #Primary 62H22 #secondary 60J05
paper · pdf · doi:10.48550/arxiv.2010.03509
openalex publication_date 2020/10/07 · openalex created_date 2022/11/07 · openalex updated_date 2026/07/28
We incorporate discrete and continuous time Markov processes as building\nblocks into probabilistic graphical models with latent and observed variables.\nWe introduce the automatic Backward Filtering Forward Guiding (BFFG) paradigm\n(Mider et al., 2021) for programmable inference on latent states and model\nparameters. Our starting point is a generative model, a forward description of\nthe probabilistic process dynamics. We backpropagate the information provided\nby observations through the model to transform the generative (forward) model\ninto a pre-conditional model guided by the data. It approximates the actual\nconditional model with known likelihood-ratio between the two. The backward\nfilter and the forward change of measure are suitable to be incorporated into a\nprobabilistic programming context because they can be formulated as a set of\ntransformation rules.\n The guided generative model can be incorporated in different approaches to\nefficiently sample latent states and parameters conditional on observations. We\nshow applicability in a variety of settings, including Markov chains with\ndiscrete state space, interacting particle systems, state space models,\nbranching diffusions and Gamma processes.\n