This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| Approximate Bayesian Computation | |
|---|---|
| Name | Approximate Bayesian Computation |
| Introduced | 1990s |
| Field | Statistics, Computational Biology |
| Notable examples | Markov chain Monte Carlo, Sequential Monte Carlo |
Approximate Bayesian Computation
Approximate Bayesian Computation is a class of computational techniques for performing Bayesian inference when likelihood functions are intractable or expensive to evaluate, relying on simulation and comparison to observed data. It originated in population genetics and has since been adopted across fields that require complex stochastic modeling, often interfacing with computational frameworks and high-performance computing resources. ABC methods trade exact likelihood evaluations for approximate summaries and distance-based acceptance, enabling parameter estimation and model comparison in settings where classical Bayesian machinery is impractical.
ABC emerged in the 1990s from work in population genetics and phylogeography where researchers struggled with intractable likelihoods from coalescent models. Early practitioners included figures associated with the development of coalescent theory and collaborators within institutions such as the University of Oxford and the University of Cambridge. The approach parallels developments in simulation-based inference explored by researchers in applied statistics and computational biology, and it intersects with algorithmic advances championed by groups at institutions like Imperial College London and Princeton University.
The core ABC workflow simulates datasets from a prior distribution over parameters using a generative model, computes summary statistics, and accepts parameter draws if the simulated summaries are sufficiently close to observed summaries under a chosen distance metric. Implementations often combine simulation engines developed in laboratories at institutions such as Harvard University, Stanford University, and the Max Planck Institute with sampling strategies inspired by Monte Carlo methods from researchers linked to CNRS and ETH Zurich. Practical pipelines integrate software paradigms that echo contributions from teams at Microsoft Research, Google Research, and the R community for reproducible computation.
Selecting informative summary statistics is critical: practitioners have borrowed techniques from sufficiency theory and dimensionality reduction, with contributions from statisticians affiliated with Columbia University, University of California Berkeley, and University of Chicago. Distance metrics and tolerance thresholds are tuned using cross-validation and calibration strategies developed by groups at the University of Toronto and the University of Washington, often incorporating ideas from optimization work at the Massachusetts Institute of Technology and Carnegie Mellon University. Methods for automatic summary selection draw on machine learning innovations from researchers at DeepMind, the Allen Institute, and the Broad Institute.
ABC encompasses rejection ABC, ABC-MCMC, ABC-SMC, and regression-adjusted ABC, each influenced by algorithmic advances across computational statistics. Rejection samplers reflect early Monte Carlo ideas linked to the lineage of work at Los Alamos National Laboratory and Sandia National Laboratories. ABC-MCMC adaptations trace conceptual roots to developments at Bell Labs and Los Alamos, while sequential Monte Carlo variants have been advanced by teams at the University of Oxford and the Max Planck Institute. Recent variants incorporate neural density estimators and amortized inference techniques driven by labs at Google Brain, OpenAI, and Facebook AI Research.
ABC supports parameter estimation and model comparison by approximating posterior probabilities and Bayes factors via accepted simulations and weighted summaries; debates on adequacy and bias involve contributions from scholars at Princeton University, Yale University, and Columbia University. Cross-validation and posterior predictive checks for ABC have been championed by researchers associated with the London School of Economics and political scientists employing simulation studies in the University of Michigan and Duke University. Hierarchical modeling and latent variable structures in ABC draw on methodological work from the University of Pennsylvania and New York University.
The theoretical foundations address convergence of ABC posteriors, asymptotic behavior under shrinking tolerances, and the impact of non-sufficient summaries; formal results have been developed by theoreticians linked to ETH Zurich, Université Paris-Saclay, and the University of Warwick. Studies of identifiability, information loss, and coverage properties cite statistical theory contributions emanating from Princeton University and the University of Cambridge. Connections to approximate inference in graphical models reflect intersections with research from the Santa Fe Institute and Los Alamos National Laboratory.
ABC has been applied extensively in population genetics, epidemiology, systems biology, ecology, and cosmology, with landmark case studies produced by researchers at the Wellcome Sanger Institute, Institut Pasteur, the Centers for Disease Control and Prevention, and the European Space Agency. Notable applied works include demographic inference in human populations comparable to studies involving the Max Planck Institute for Evolutionary Anthropology, outbreak reconstruction in collaboration with Public Health England, and ecological modeling by teams at the Smithsonian Institution and CSIRO. In engineering and finance, practitioners at NASA and the Federal Reserve have explored ABC adaptations alongside machine learning research from the Vector Institute and the Alan Turing Institute.
Category:Statistical methods