Theory Of Optimal Experiments

By V. Fedorov, W. J. Studden, E. Klimko. 1972

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

Coresets via Bilevel Optimization for Continual Learning and Streaming (2020, neutral): “Note that by imposing a bound on the weights as commonly done in experimental design [18], our algorithm would be equivalent to Frank-Wolfe [22].”
PowerPlay: Training an Increasingly General Problem Solver by Continually Searching for the Simplest Still Unsolvable Problem (2011, neutral): “CONNECTION TO TRADITIONAL ACTIVE LEARNING Traditional active learning methods (Fedorov, 1972) such as AdaBoost (Freund and Schapire, 1997) have a totally different set-up and purpose: there the user provides a set of samples to be learned, then each new classifier in a series of classifiers focuses ”
Active Learning with Statistical Models (1996, methodology): “2 Example: Active Learning with a Neural Network In this section we review the use of techniques from Optimal Experiment Design (OED) to minimize the estimated variance of a neural network (Fedorov, 1972; MacKay, 1992; Cohn, 1994).”
First Experiments with PowerPlay (2012, methodology): “Note that active learning methods [5] such as AdaBoost [6] have a totally different set-up and purpose: there the user provides a set of samples to be learned, then each new classiﬁer in a series of classiﬁers focuses on samples badly classiﬁed by previous classiﬁers.”
Learning with Submodular Functions: A Convex Optimization Perspective (2011, background): “, [64]), and is thus equivalent to maximizing a submodular function (for which forward selection has theoretical guarantees, see §11.”
Optimal discrimination designs (2009, background): “However, a robust version of the two optimality criteria can easily be obtained applying a sequential, Bayesian or (standardized) maximin approach (see, e.g., Atkinson and Fedorov [3], Müller and Pázman [25], Dette and Neugebauer [13, 14] or Dette [11], among many others).”
Near-Optimal Sensor Placements in Gaussian Processes: Theory, Efficient Algorithms and Empirical Studies (2008, unknown): “Approaches to the difficult combinatorial optimization include simulated annealing (Meyer and Nachtsheim, 1988), pairwise exchange (Fedorov, 1972; Mitchell, 1974a,b; Cook and Nachtsheim, 1980; Nguyen and Miller, 1992), forward and backward greedy heuristics (MacKay, 1992; Caselton and Zidek, 1984).”
Bayesian Gaussian process models : PAC-Bayesian generalisation error bounds and sparse approximations (2003, background): “It is also somewhat simpler than the latter query-based one which has been studied in [53, 108, 175, 58, 31].”
Sensor Selection via Convex Optimization (2009, methodology): “The sensor selection problem is closely related to the D-optimal experiment design problem [28], [29].”
An Efficient Algorithm for Constructing Bayesian Optimal Choice Designs (2009, methodology): “Second, the algorithm is an adaptation of Meyer and Nachtsheim’s (1995) coordinate-exchange algorithm, which is much faster than the modified Fedorov algorithm. Third, it involves a revised formula for the VB-optimality criterion so that its computation is even more efficient. Last, it uses an updat”
Bayesian Experimental Design: A Review (1995, background): “Apart from Flournoy (1993) there are no true \case studies" that we know of, where Bayesian ideas have been formally applied to the design of an actual scienti c experiment before it is done.”
A survey of adaptive sampling for global metamodeling in support of simulation-based complex engineering design (2017, methodology): “To improve optimization efficiency, we often employ the modified exchange algorithm (Fedorov 1972) that replaces a point by the one that improves the sampling criterion at each iteration.”
Logistic regression with an auxiliary data source (2005, methodology): “In this section we describe how D p l can be actively selected from D p , based on the Fisher information matrix [Fedorov, 1972, MacKay, 1992].”
Optimal designs for inverse prediction in univariate nonlinear calibration models (2004, methodology): “The first one follows the principle of the modified Fedorov exchange algorithm (Fedorov [14], Miller and Nguyen [15]).”
Curiosity driven reinforcement learning for motion planning on humanoids (2014, methodology): “Subsequently, a probabilistic AC version (Storck et al., 1995) used the well known Kullback-Leibler (KL) divergence (Lindley, 1956; Fedorov, 1972) to deﬁne non-stationary, intrinsic rewards reﬂecting the changes of a probabilistic model of the environment after new experiences.”
Active learning for semi-supervised multi-task learning (2009, methodology): “The true value of θ required in calculating z and γ is replaced with the most recent update of the parameters, following the strategy taken in [10, 12].”