In this chapter we introduce Bayesian inference and use it to extend the frequentist models of the previous chapters. To do this, we describe the concept of model priors, informative priors, uninformative priors, and conjugate prior-likelihood pairs . We then discuss Bayesian updating rules for using priors and likelihoods to obtain posteriors. Building upon priors and posteriors, we then describe more advanced concepts including predictive distributions, Bayes factors, expectation maximization to obtain maximum posterior estimators, and model selection. Finally, we present hierarchical Bayesian models, Markov blankets, and graphical representations. We conclude with a case study on change point detection.

In this chapter we introduce the clustering problem and use it to motivate mixture models. We start by describing clustering in a frequentist paradigm and introduce the relevant likelihoods and latent variables. We then discuss properties of the likelihoods including invariance with respect to label swapping. Finally, we expand this discussion to describe clustering and mixture models more generally within a Bayesian paradigm. This allows us to introduce Dirichlet priors used in inferring the weight we ascribe to each cluster component from which the data are drawn. Finally, we describe the infinite mixture model and Dirichlet process priors within the Bayesian nonparametric paradigm, appropriate for the analysis of uncharacterized data that may contain an unspecified number of clusters.

In this chapter we present dynamical systems and their probabilistic description. We distinguish between system descriptions with discrete and continuous state-spaces as well as discrete and continuous time. We formulate examples of statistical models including Markov models, Markov jump processes, and stochastic differential equations. In doing so, we describe fundamental equations governing the evolution of the probability of dynamical systems. These equations include the master equation, Langevin equation, and Fokker–Plank equation. We also present sampling methods to simulate realizations of a stochastic dynamical process such as the Gillespie algorithm. We end with case studies relevant to chemistry and physics.

In this chapter we provide an overview of data modeling and describe the formulation of probabilistic models. We introduce random variables, their probability distributions, associated probability densities, examples of common densities, and the fundamental theorem of simulation to draw samples from discrete or continuous probability distributions. We then present the mathematical machinery required in describing and handling probabilistic models, including models with complex variable dependencies. In doing so, we introduce the concepts of joint, conditional, and marginal probability distributions, marginalization, and ancestral sampling.

In this chapter we formulate the general regression problem relevant to function estimation. We begin with simple frequentist methods and quickly move to regression within the Bayesian paradigm. We then present two complementary mathematical formulations: one that relies on Gaussian process priors, appropriate for the regression of continuous quantities, and one that relies on Beta–Bernoulli process priors, appropriate for the regression of discrete quantities. In the context of the Gaussian process, we discuss more advanced topics including various admissible kernel functions, inducing point methods, sampling methods for nonconjugate Gaussian process prior-likelihood pairs, and elliptical slice samplers. For Beta–Bernoulli processes, we address questions of posterior convergence in addition to applications. Taken together, both Gaussian processes and Beta–Bernoulli processes constitute our first foray into Bayesian nonparametrics. With end of chapter projects, we explore more advanced modeling questions relevant to optics and microscopy.

In this chapter we present computational Monte Carlo methods to sample from probability distributions, including Bayesian posteriors, that do not permit direct sampling. In doing so, we introduce the basis for Monte Carlo and Markov chain Monte Carlo sampling schemes and delve into specific methods. These include, at first, samplers such as the Metropolis–Hastings algorithms and Gibbs samplers and discuss the interpretation of the output of these samplers including the concept of burn-in and sample correlation. We also discuss more advanced sampling schemes including auxiliary variable samplers, multiplicative random walk samplers, and Hamiltonian Monte Carlo.

In this chapter we introduce the concept of likelihoods, how to incorporate measurement uncertainty into likelihoods, and the concept of latent variables that arise in describing measurements. We then show how the principle of maximum likelihood is applied to estimate values for unknown parameters and discuss a number of numerical optimization techniques used in obtaining maximum likelihood estimators. These techniques include a discussion of expectation maximization.