TY - GEN
T1 - Estimation of parameters sensitivity for scientific workflows
AU - Khan, Fakhri Alam
AU - Han, Yuzhang
AU - Pllana, Sabri
AU - Brezany, Peter
PY - 2009
Y1 - 2009
N2 - Usually workflow activities in the scientific domain depend on a collection of parameters. These parameters determine the output of the activity, and consequently the output of the whole workflow. In the scientific domain, workflows have exploratory nature and are used to understand a scientific phenomenon or answer scientific questions. In the process of a scientific experiment a workflow is executed multiple times using various values of the parameters of activities. It is relevant to identify (1) which parameter strongly affects the overall result of the workflow and (2) for which combination of parameter values we obtain the expected result. Foreseeing these issues, in this paper we present our methodology to estimate the significance of all scientific workflow parameters as well as to estimate the most significant parameter to the workflow. The estimation of parameter significance will enable the scientist to fine tune, and optimize his results efficiently. Furthermore, we empirically validate our methodology on Non-Invasive Glucose Measurement workflow and discuss our results. The NIGM workflow uses the neural network model to calculate the glucose level in patient blood. The neural network model has a set of parameters, which affect the result of the workflow significantly. But, unfortunately the impact significance of these parameters is commonly unknown to the user. We present our approach for estimating and quantifying impact significance of neural network parameters.
AB - Usually workflow activities in the scientific domain depend on a collection of parameters. These parameters determine the output of the activity, and consequently the output of the whole workflow. In the scientific domain, workflows have exploratory nature and are used to understand a scientific phenomenon or answer scientific questions. In the process of a scientific experiment a workflow is executed multiple times using various values of the parameters of activities. It is relevant to identify (1) which parameter strongly affects the overall result of the workflow and (2) for which combination of parameter values we obtain the expected result. Foreseeing these issues, in this paper we present our methodology to estimate the significance of all scientific workflow parameters as well as to estimate the most significant parameter to the workflow. The estimation of parameter significance will enable the scientist to fine tune, and optimize his results efficiently. Furthermore, we empirically validate our methodology on Non-Invasive Glucose Measurement workflow and discuss our results. The NIGM workflow uses the neural network model to calculate the glucose level in patient blood. The neural network model has a set of parameters, which affect the result of the workflow significantly. But, unfortunately the impact significance of these parameters is commonly unknown to the user. We present our approach for estimating and quantifying impact significance of neural network parameters.
UR - http://www.scopus.com/inward/record.url?scp=77949518223&partnerID=8YFLogxK
U2 - 10.1109/ICPPW.2009.9
DO - 10.1109/ICPPW.2009.9
M3 - Conference contribution
AN - SCOPUS:77949518223
SN - 9780769538037
T3 - Proceedings of the International Conference on Parallel Processing Workshops
SP - 457
EP - 462
BT - ICPPW 2009 - The 38th International Conference Parallel Processing Workshops
ER -