Bayesian Experimental Design: A Review

By K. Chaloner, Isabella Verdinelli. 1995

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Taking the Human Out of the Loop: A Review of Bayesian Optimization (2016, neutral): “See [37] for a detailed survey.”
On the Equivalence between Kernel Quadrature Rules and Random Feature Expansions (2015, background): “Moreover, if the decay is geometric as r i , then we get d ( λ ) 6 log( R 2 /λ ) , and thus n > (log m ) 2 , which is even more signiﬁcant.”
Coresets via Bilevel Optimization for Continual Learning and Streaming (2020, background): “For this special case there are natural connections to the literature on optimal experimental design, a well-studied topic in statistics [8].”
Gaussian Processes for Data-Efficient Learning in Robotics and Control (2015, neutral): “[10] and Bayesian optimization [9], [33], [41], [51] to avoid getting stuck in local minima.”
Information-Theoretic Regret Bounds for Gaussian Process Optimization in the Bandit Setting (2009, background): “…(Robbins, 1952), where the goal is to maximize cumulative reward by optimally balancing exploration and exploitation, and experimental design (Chaloner & Verdinelli, 1995), where the function is to be explored globally with as few evaluations as possible, for example by maximizing…”
Gaussian process dynamic programming (2009, background): “As described by Chaloner and Verdinelli (1995), the second term can be derived from the expected Shannon information (entropy) of the predictive distribution V ∗ k (˜ x ) or the Kullback-Leibler divergence between the predictive distribution of V ∗ k (˜ x )”
Foundations of Linear and Generalized Linear Models (2015, methodology): “The bootstrap is another possible way to determine standard errors for lasso estimates (Chatterjee and Lahiri 2011). Lockhart et al. (2014) proposed a significance test for the lasso, based on how much of the covariance between y and the model-fitted values can be attributed to a particular predicto”
Monte Carlo Gradient Estimation in Machine Learning (2019, background): “Designs are problem configurations, such as a hypothesis in an experiment, the locations of sensors on a device, or the hyperparameters of a statistical model (Chaloner and Verdinelli, 1995).”
Decision Theoretic Approaches to Experiment Design and External Validity (2016, unknown): “Even the proponents of Bayesian experimental design note that despite its strong normative appeal, it remains rarely, if ever, used (Chaloner and Verdinelli, 1995).”
Robust Submodular Observation Selection (2008, background): “, Chaloner and Verdinelli, 1995) is to linearize f around an initial parameter estimate θ0, that is, y = f (x,θ0)+V (x)(θ−θ0)+w, (8) where V (x) is the Jacobian of f with respect to the parameters θ, evaluated at θ0. Subsequently, a locally-optimal design is sought, which is optimal for the linear d”
A modern Bayesian look at the multi-armed bandit (2010, background): “Designs for nonlinear models, such as probit or logistic regression, are complicated by the fact that the information matrix depends on unknown parameters [9].”
Near-Optimal Sensor Placements in Gaussian Processes: Theory, Efficient Algorithms and Empirical Studies (2008, unknown): “By using different utility functions, Bayesian versions of A-, D-, and E- optimality can be developed (Chaloner and Verdinelli, 1995).”