Hardwick, J. and Stout, Q.F., 1998. Flexible algorithms for creating and analyzing adaptive sampling procedures. New Developments and Applications in Experimental Design, IMS Lec. Notes--Mono. Series 34, 91--105.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....method is useful for unconstrained local optimization of scalar functions of a vector argument; PFTV92] contains an excellent description and code for implementation. The re indexing scenario may be thought of as a very large adaptive sampling procedure, or multi armed bandit problem (see [HS98] in which the cost of making an observation is balanced against the possible bene ts obtained by viewing the document. Given the size of the indexable web (320 million pages [LG98] in 1998, and 800 million pages [LG99] in 1999) it is probably not tractable to treat the search engine s problem ....

J. P. Hardwick and Q. F. Stout. Flexible algorithms for creating and analyzing adaptive sampling procedures, 1998. To appear. Also see http://www.eecs.umich.edu/~qstout/abs/Seattle97.html. 142

....prior to their observation, so planning must involve building models from previous observations. This is a tricky problem, since we will be basing observation schedules on these dynamic models. Similar problems in the experimental design literature are referred to as allocation problems [HS98] classic examples of which are bandit problems [Ber87] The terms stems from the plight of a casino gambler, having the option of playing some number of slot machines (colloquially known as one armed bandits ) The payoff from each machine is governed by some unknown probability distribution. ....

J. P. Hardwick and Q. F. Stout. Flexible algorithms for creating and analyzing adaptive sampling procedures, 1998. To appear. Also see http://www.eecs.umich.edu/~qstout/abs/Seattle97.html.

No context found.

Hardwick, J. and Stout, Q.F., 1998. Flexible algorithms for creating and analyzing adaptive sampling procedures. New Developments and Applications in Experimental Design, IMS Lec. Notes--Mono. Series 34, 91--105.

No context found.

Hardwick, J. and Stout, Q.F., 1998. Flexible algorithms for creating and analyzing adaptive sampling procedures. New Developments and Applications in Experimental Design, IMS Lec. Notes--Mono. Series 34, 91--105.

No context found.

Hardwick, J. and Stout, Q.F. (1998), "Flexible algorithms for creating and analyzing adaptive sampling procedures", New Developments and Applications in Experimental Design, IMS Lec. Notes--Mono. Series 34: 91--105.

No context found.

J Hardwick, QF Stout. Flexible algorithms for creating and analyzing adaptive sampling procedures. In: New Developments and Applications in Experimental Design. IMS Lec Notes-- Mono Series 34:91--105, 1998. 27

....of a trial it has been observed that one treatment is performing better than the others, then more patients may be assigned to the apparently better treatment. Thus adaptive designs can provide significant ethical benefits, and in industrial settings can have significant cost and time advantages [6]. However, adaptive designs are rarely used, largely because they are far more difficult to analyze. Analytical solutions are impossible in all but the most trivial cases, and computational approaches are often considered infeasible. We are developing new algorithms, and optimized ....

....impeding their use. Our goal is to reduce computational concerns to the point where they are not a key issue in the selection of appropriate designs. This paper has concentrated on the parallel computational aspects of this work, while other papers analyze the statistical and application impact [4, 5, 6]. Unfortunately, the recurrences involved have attributes that make it difficult to achieve high performance and scalability. Space tends to be the limiting factor, and trying to ameliorate this causes overhead and a significant increase in program complexity. As noted in Section 4, increases in ....

Hardwick, J. and Stout, Q.F. (1998), "Flexible algorithms for creating and analyzing adaptive sampling procedures", New Developments and Applications in Experimental Design, IMS Lec. Notes--Mono. Series 34, pp. 91--105.

....of a trial it has been observed that one treatment is performing better than the others, then more patients may be assigned to the apparently better treatment. Thus adaptive designs can provide significant ethical benefits, and in industrial settings can have significant cost and time advantages [6]. However, adaptive designs are rarely used, largely because they are far more difficult to analyze. Analytical solutions are impossible in all but the most trivial cases, and computational approaches are often considered infeasible. We are developing new algorithms, and optimized ....

....impeding their use. Our goal is to reduce computational concerns to the point where they are not a key issue in the selection of appropriate designs. This paper has concentrated on the parallel computational aspects of this work, while other papers analyze the statistical and application impact [4, 5, 6]. Unfortunately, the recurrences involved have attributes that make it difficult to achieve high 13 n uncompressed initial scalable 100 100 1 1 200 1 21 16 300 1 1 173 Max problem solvable: uncompressed: 105; initial: 231; scalable: 1. Table 9: Min. processors (p) needed to solve problem of ....

Hardwick, J. and Stout, Q.F. (1998), "Flexible algorithms for creating and analyzing adaptive sampling procedures", New Developments and Applications in Experimental Design, IMS Lec. Notes--Mono. Series 34, pp. 91--105.

....approach. 3 Computations Due to their high dimensional nature, implementing the recurrences in a computer program can be somewhat challenging. The state space is 6 dimensional, so the most straightforward implementation would use an array of size n 6 . By using the well known techniques (see [7]) of doing calculations level by level and overwriting old values this array space can be reduced to n 5 , and by utilizing the constraint that s 1 f 1 u 1 s 2 f 2 u 2 n and mapping to a 1 dimensional array, this can be further reduced to approximately n 5 =5 . For sample sizes of ....

Hardwick, J. and Stout, Q.F. (1998), "Flexible algorithms for creating and analyzing adaptive sampling procedures", New Developments and Applications in Experimental Design, IMS Lec. Notes--Mono. Series 34, pp. 91--105.

No context found.

J.P. Hardwick and Q.F. Stout. Flexible algorithms for creating and analyzing adaptive sampling procedures. In N. Flourny, W.F. Rosenberger, and W.K. Wong, editors, New Developments and Applications in Experimental Design: Selected Proceedings of a 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC