Parallel Programs for Adaptive Designs

We discuss the role of parallel computing in the design and analysis of adaptive sampling procedures, and show how some efficient parallel programs we re developed to allow one to analyze useful sample sizes. Response adaptive designs are an important class of learning algorithms for a stochastic environment and apply in a large number of situations. As an illustrative example, we focus on the problem of optimally assigning patients to treatments in clinical trials. While response adaptive designs have significant ethical and cost advantages, they are rarely utilized because of the complexity of optimizing and analyzing them. Computational challenges include massive memory requirements, few calculations per memory access, and multiply-nested loops with dynamic indices. We analyze the effects of various parallelization options, showing that, while standard approaches do not necessarily work well, with effort an efficient, highly scalable program can be developed. This allows us to solve problems thousands of times more complex than those solved previously, which helps make adaptive designs practical.

[1]  D Tang,et al.  An efficient parallel dynamic programming algorithm , 1995 .

[2]  Jonathan C. Hardwick.  Computational problems associated with minimizing the risk in a simple clinical trial , 1989 .

[3]  J. A. Bather,et al.  On the Allocation of Treatments in Sequential Medical Trials , 1985 .

[4]  R. Bellman.  A PROBLEM IN THE SEQUENTIAL DESIGN OF EXPERIMENTS , 1954 .

[5]  Art Lew,et al.  Dynamic programming, decision tables, and the Hawaii Parallel Computer , 1994 .

[6]  L. J. Wei,et al.  The Randomized Play-the-Winner Rule in Medical Trials , 1978 .

[7]  Quentin F. Stout,et al.  Scalable Algorithms for Adaptive Statistical Designs , 2000 .

[8]  Quentin F. Stout,et al.  Using Path Induction to Evaluate Sequential Allocation Procedures , 1999, SIAM J. Sci. Comput..

[9]  Wojciech Rytter,et al.  On Efficient Parallel Computations for some Dynamic Programming Problems , 1988, Theor. Comput. Sci..

[10]  W F Rosenberger,et al.  A COMPARISON OF URN DESIGNS FOR RANDOMIZED CLINICAL TRIALS OF K > 2 TREATMENTS , 2000, Journal of biopharmaceutical statistics.

[11]  Xikui Wang.  A bandit process with delayed responses , 2000 .

[12]  P. Armitage.  The search for optimality in clinical trials , 1985 .

[13]  Uttam Bandyopadhyay,et al.  Delayed Response in Randomized Play-The-Winner Rule; A Decision Theoretic Outlook , 1996 .

[14]  Maurice Tchuente,et al.  Dynamic Programming on Two-Dimensional Systolic Arrays , 1988, Inf. Process. Lett..

[15]  V. G. Kulkarni,et al.  Optimal Bayes procedures for selecting the better of two Bernoulli populations , 1986 .

[16]  Quentin F. Stout,et al.  A program for sequential allocation of Bernoulli populations , 1999 .

[17]  R. C. Oehmke,et al.  High performance dynamic array structures , 2004 .

[18]  You-Gan Wang.  Sequential allocation in clinical trials , 1991 .

[19]  Raghu Sastry,et al.  A systolic array for approximate string matching , 1993, Proceedings of 1993 IEEE International Conference on Computer Design ICCD'93.

[20]  Tao Jiang,et al.  On Efficient Parallel Algorithms for Solving Set Recurrence Equations , 1993, J. Algorithms.

[21]  Roger L. Wainwright,et al.  Load balancing techniques for dynamic programming algorithms on hypercube multiprocessors , 1993, SAC '93.

[22]  Roderick M. Duchesne.  The Use of Computers in British Libraries and Information Services: an Analysis , 1974 .

[23]  Quentin F. Stout,et al.  Optimal Adaptive Designs for Delayed Response Models: Exponential Case , 2001 .

[24]  Sartaj Sahni,et al.  String Editing on an SIMD Hypercube Multicomputer , 1990, J. Parallel Distributed Comput..

[25]  Janis Hardwick.  A modified bandit as an approach to ethical allocation in clinical trials , 1995 .

[26]  R. Simon,et al.  Adaptive treatment assignment methods and clinical trials. , 1977, Biometrics.

[27]  Quentin F. Stout,et al.  Flexible Algorithms for Creating and Analyzing Adaptive Sampling Procedures , 1998 .

[28]  R. N. Bradt.  On the Design and Comparison of Certain Dichotomous Experiments , 1954 .

[29]  Hideyuki Douke.  On sequential design based on Markov chains for selecting one of two treatments in clinical trials w , 1994 .

[30]  Eric Bach,et al.  Asynchronous Analysis of Parallel Dynamic Programming Algorithms , 1996, IEEE Trans. Parallel Distributed Syst..

[31]  Quentin F. Stout,et al.  Optimal few-stage designs , 2002 .

[32]  Stephen G. Eick,et al.  The two-armed bandit with delayed responses , 1988 .