[abstract] A PARAMETER ESTIMATION SYSTEM WITH COMPUTATIONAL ADVANTAGES FOR FITTING PROBABILISTIC DECOMPRESSION MODELS TO EMPIRICAL DATA.

Rubicon Research Repository/Manakin Repository

[abstract] A PARAMETER ESTIMATION SYSTEM WITH COMPUTATIONAL ADVANTAGES FOR FITTING PROBABILISTIC DECOMPRESSION MODELS TO EMPIRICAL DATA.

Show full item record


Title: [abstract] A PARAMETER ESTIMATION SYSTEM WITH COMPUTATIONAL ADVANTAGES FOR FITTING PROBABILISTIC DECOMPRESSION MODELS TO EMPIRICAL DATA.
Author: Howle, LE; Weber, PW; Vann, RD
Abstract: BACKGROUND: Parameter values for probabilistic decompression models can be estimated directly from empirical calibration data. A modular parameter estimation system was created with computational benefits including object-oriented programming methods, use of analytic instead of numerical techniques, a library of optimization methods including hybrid gradient ascent algorithms and parallel processing. Reliability of the approach was tested by reproducing the results of earlier work that fit “exponential-exponential” (EE) and “linear-exponential” (LE) models to a standard calibration database (1). MATERIALS AND METHODS: Code was developed with an engine written in C#.NET. Exact integrations of risk functions for the EE and LE models were derived and employed to reduce computation time. A baseline parameter optimization method was developed and compared with several techniques from the C#.NET library. Parallel processing techniques were also developed to apportion tasks to multiple networked computers and on stand-alone multiprocessor computers. Previously published calibration data was used in this system to estimate previously published parameters, and these parameters were used as a basis of comparison (1, 2). RESULTS: Excellent agreement with previously published results was obtained with this system. Re-optimizing the parameters with the system also produced improved results. The non-normal, non-analytic behavior of one variant of the LE model caused problems not previously reported. Complete optimization for a 9-parameter LE model required approximately ten minutes with exact integration and approximately one month with numerical integration. Optimization techniques from a commercial C#.NET library produced optimization times which were 30-fold less than the baseline. Homogeneous parallel processing reduced model optimization time by 40%, and grid computing offered performance improvement even with network communication delays. CONCLUSIONS: The modular parameter estimation system developed produced results that were in agreement with those of previously published systems and offered advantages in computation speed and portability. REFERENCES: 1. Thalmann. UHM 24:255. 1997. 2. Gerth. UHM 24:275. 1997.
Description: Abstract of the Undersea and Hyperbaric Medical Society, Inc. Annual Scientific Meeting held June 14-16, 2007. Ritz-Carlton Kapalua Maui, Hawaii (http:www.uhms.org)
URI: http://archive.rubicon-foundation.org/5066
Date: 2007

Files in this item

Files Size Format View
abstract.txt 259bytes Text file View/Open

This item appears in the following Collection(s)

  • UHMS Meeting Abstracts
    This is a collection of the published abstracts from the Undersea and Hyperbaric Medical Society (UHMS) annual meetings.

Show full item record

Browse

My Account