Bayesian experimental design without posterior calculations: an adversarial approach
From MaRDI portal
Abstract: Most computational approaches to Bayesian experimental design require making posterior calculations repeatedly for a large number of potential designs and/or simulated datasets. This can be expensive and prohibit scaling up these methods to models with many parameters, or designs with many unknowns to select. We introduce an efficient alternative approach without posterior calculations, based on optimising the expected trace of the Fisher information, as discussed by Walker (2016). We illustrate drawbacks of this approach, including lack of invariance to reparameterisation and encouraging designs in which one parameter combination is inferred accurately but not any others. We show these can be avoided by using an adversarial approach: the experimenter must select their design while a critic attempts to select the least favourable parameterisation. We present theoretical properties of this approach and show it can be used with gradient based optimisation methods to find designs efficiently in practice.
Cites work
- A course in game theory.
- A Review of Modern Computational Algorithms for Bayesian Optimal Design
- An induced natural selection heuristic for finding optimal Bayesian experimental designs
- Asymptotic Statistics
- Bayesian experimental design: A review
- Bayesian information in an experiment and the Fisher information distance
- Bayesian model comparison with the Hyvärinen score: computation and consistency
- Bayesian-Optimal Design via Interacting Particle Systems
- Efficient construction of Bayes optimal designs for stochastic process models
- Estimation of non-normalized statistical models by score matching
- Expected information as ecpected utility
- GRADIENT-BASED STOCHASTIC OPTIMIZATION METHODS IN BAYESIAN EXPERIMENTAL DESIGN
- scientific article; zbMATH DE number 1972910 (Why is no real title available?)
- scientific article; zbMATH DE number 1522713 (Why is no real title available?)
- scientific article; zbMATH DE number 6982909 (Why is no real title available?)
- Improving the efficiency of fully Bayesian optimal design of experiments using randomised quasi-Monte Carlo
- Information Geometry of the Gaussian Distribution in View of Stochastic Optimization
- On a Measure of the Information Provided by an Experiment
- Optimal Observation Times in Experimental Epidemic Processes
- Optimum experimental designs, with SAS
- Proper local scoring rules
- Proper local scoring rules on discrete sample spaces
- Properties of Fisher information gain for Bayesian design of experiments
- Robust experiment design via stochastic approximation
- Sequential Bayesian experimental design for implicit models via mutual information
- Stochastic approximation with two time scales
- Strictly Proper Scoring Rules, Prediction, and Estimation
- Towards Bayesian experimental design for nonlinear models that require a large number of sampling times
Cited in
(4)
This page was built for publication: Bayesian experimental design without posterior calculations: an adversarial approach
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6122019)
