Performance prediction of ocean color Monte Carlo simulations using multi-layer perceptron neural networks

A high‐performance computing framework for Monte Carlo ocean color simulations

“…Given a set of MLP input parameters, the prediction capability of trained MLPs further depends on the following: (1) the ranges of parameter values; (2) the intervals between values of each parameter; and (3) the total number of training samples. A previous publication has examined the effect of the size of training data, which is confirmed as a major MLP performance factor. The case study reported in the present work follows this research venue by taking the ranges and intervals of parameter values into consideration.…”

“…Second, the photon tracing time shows a nonlinear dependence on input simulation parameters, which is analytically intractable. Hence this component is modeled using MLP neural networks (see for more detail).…”

“…This section presents the JEM algorithm to minimize the total time for running MOX simulation jobs distributed to multiple execution environments . Let be a set of N s simulation jobs, be a set of N e execution environments, and f ℑ j = f ( s ℑ , e j ) represent the cost of running s ℑ on e j .…”

“…Second, the photon tracing time shows a nonlinear dependence on input simulation parameters, which is analytically intractable. Hence this component is modeled using MLP neural networks (see [19] for more detail). The expected execution time T of a production MOX run in a particular parallel computing environment is modeled as a function of a set of input simulation parameters, the number of CPU cores N c , and the number of photons N p as follows:…”

“…The present work advances reference research achievements by integrating new results and methods in a unifying framework of performance optimization techniques. Numerical results acknowledge an extended range of input simulation parameters to investigate the effects of environmental factors on run‐time performance optimization.…”