| Environmentally adaptive design of multistatic sonar systems in coastal environments is important because of the propagation medium impact on system performance. Complex underwater acoustic models allowing computation of sonar performance prediction have been developed. In this work, we develop intelligent optimization algorithms for solving the placement problems. The objective functions, derived from the sonar performance prediction model, are not analytically tractable. Therefore, standard mathematical programming techniques can not be applied without simplifying assumptions. Instead, we resort to stochastic approximate methods, especially a novel population-based metaheuristic algorithm, the particle swarm optimization (PSO).; Population-based techniques are adept at finding near-optimal solutions for combinatorial optimization problems. Thanks to advances in computational power, they have been successfully applied to some challenging real-world optimization problems. Their wide acceptance is limited however because most real-world optimization problems present objective functions that are computationally intensive. In addition, the thousands of function evaluations required by this class of optimizers render their application impractical. Surrogate models and parallelization are commonly used to address that problem, but even the evaluation of these models can be time-consuming. Another limitation is the lack of guarantee in finding the optimal solution.; The problem at hand, sonar sensor placement, is in the category of real-world optimization problems with expensive objective functions. Here we propose and develop computationally frugal methods based on PSO for solving coverage maximization problem. We then apply it to multi-objective placement problems.; In the second part of our work, we present important contributions to the area of genomic signal processing and analysis. Specifically, we looked at the problem of quantitation, or estimation of initial concentration, using real-time Polymerase Chain Reaction (PCR). Quantitative estimation of genetic material is an important problem with applications in multiple fields from homeland security to clinical diagnostic or genomic research. Real-time PCR is one of the tools for solving this problem. Because of its simple basic principles, there has been a huge development of instrumentation to perform this reaction. On the analytical side, there exist multiple models to explain the reaction but none are accurate enough to completely model it. Using data generated by a high-throughput real-time PCR instrument we developed; the first of its kind to operate on single-digit microliter capillaries, we modeled reaction using artificial neural networks. |
