Being Bayesian in the 2020s: opportunities and challenges in the practice of modern applied Bayesian statistics

Joshua J. Bon, Adam Bretherton, Katie Buchhorn, Susanna Cramb, Christopher Drovandi, Conor Hassan, Adrianne L. Jenner, Helen J. Mayfield, James M. McGree, Kerrie Mengersen, Aiden Price, Robert Salomone, Edgar Santos–Fernández, Julie Vercelloni, Xiaoyu Wang

Abstract

Building on a strong foundation of philosophy, theory, methods and computation over the past three decades, Bayesian approaches are now an integral part of the toolkit for most statisticians and data scientists. Whether they are dedicated Bayesians or opportunistic users, applied professionals can now reap many of the benefits afforded by the Bayesian paradigm. In this paper, we touch on six modern opportunities and challenges in applied Bayesian statistics: intelligent data collection, new data sources, federated analysis, inference for implicit models, model transfer and purposeful software products.

Introduction

Modern Bayesian statistics has transitioned from a theoretical niche to a cornerstone of data science. However, the 2020s present unique practical hurdles: expensive data collection, "implicit models" (such as complex simulators) that lack explicit likelihood functions, and stringent privacy requirements that prevent data centralization. This paper explores how integrating computation, experimental design, and software engineering allows Bayesian methods to address these hurdles in fields ranging from ecology to oncology.

Methodology

The authors review several key techniques designed for modern data environments:

  • Bayesian Optimal Experimental Design (OED): Utilizes the Approximate Coordinate Exchange (ACE) algorithm and Laplace approximations to identify sampling locations that maximize "Expected Utility" (information gain).
  • Likelihood-Free Inference: Compares Sequential Monte Carlo-based Approximate Bayesian Computation (SMC-ABC) with Sequential Neural Likelihood (SNL) for models where simulation is possible but the likelihood is intractable.
  • Federated Learning: Enables distributed MCMC and Variational Inference (VI), allowing model training across decentralized data sources without sharing raw sensitive information.
  • Model Transfer: Employs Power Priors to leverage information from data-rich source domains to improve inference in data-poor target domains.

Case Studies and Data

The research draws on a diverse array of real-world datasets, including bathymetry and coral cover from the Great Barrier Reef, 360-degree VR imagery from the Antarctic, tumor growth measurements in oncology, and the Australian Cancer Atlas, which tracks incidence across 2,148 geographical areas.

Key Insights: Humans as Sensors

A major frontier is the use of "Humans as Sensors." By utilizing VR and citizen science platforms, researchers can elicit expert knowledge or process massive image sets. Bayesian models can explicitly account for human bias, sensitivity, and specificity, transforming "noisy" volunteer input into reliable data that matches expert-level accuracy.

Citizen science classification of coral images and the resulting latent coral proportion density.

Results and Impact

The application of OED in coral reef surveys demonstrated that focusing on physiologically likely areas (e.g., shallow, light-rich regions) significantly reduces wasted sampling effort. The introduction of "sampling windows" provides field teams with operational flexibility while maintaining 99% of optimal information.

Design efficiency and optimal transect region design for coral reef sampling.

Furthermore, "purposeful software products" like the Australian Cancer Atlas and the CoRiCAL COVID-19 risk calculator have successfully translated complex hierarchical models into interactive tools that inform national health policy and public perception.

Conclusion and Future Work

While promising, challenges remain. OED is highly dependent on the accuracy of the underlying model. Federated Bayesian analysis for spatial data remains an open research problem due to information "leakage" across geographical boundaries. Future work will focus on automating model transfer and developing robust algorithms for decentralized, complex hierarchical structures.

Summary updated

Abstract

Building on a strong foundation of philosophy, theory, methods and computation over the past three decades, Bayesian approaches are now an integral part of the toolkit for most statisticians and data scientists. Whether they are dedicated Bayesians or opportunistic users, applied professionals can now reap many of the benefits afforded by the Bayesian paradigm. In this paper, we touch on six modern opportunities and challenges in applied Bayesian statistics: intelligent data collection, new data sources, federated analysis, inference for implicit models, model transfer and purposeful software products.
BibTeX
@inproceedings{bon2023being,
  author = {Joshua J. Bon and Adam Bretherton and Katie Buchhorn and Susanna Cramb and Christopher Drovandi and Conor Hassan and Adrianne L. Jenner and Helen J. Mayfield and James M. McGree and Kerrie Mengersen and Aiden Price and Robert Salomone and Edgar Santos–Fernández and Julie Vercelloni and Xiaoyu Wang},
  title = {{Being Bayesian in the 2020s: opportunities and challenges in the practice of modern applied Bayesian statistics}},
  booktitle = {Philosophical},
  year = {2023},
  doi = {10.1098/rsta.2022.0156},
}
Figure