An Automatic Multi-dimensional Global Optimisation Implementation With Dynamic Task Mapping Model
Xiaotian Xu · 2010
The problem addressed here is about continuous global optimisation which frequently arises in many cases such as minimizing the energy function for protein structures, chemistry, econometrics, data analysis etc. A method is implemented for the optimisation of a complex multi-dimensional biocomputational problem. The quality of optimising results of these methods may depend on the pre-unknown passed in initial values. This implies that a sorting algorithm to search for some decent initial values is necessary. Meanwhile the computation is load imbalance, and the computation of the optimisation for each initial value is highly independent which means the computations are inherently parallel, so the employing of the dynamic task scheduling by task farm can balance the work load on a massive parallel many-processor machine in order to save the computational time. The work is a small part of my supervisor Dr. Kevin Stratford’s project named Multiple Light Input Signals to the Gene Network of the Circadian Clock which is implemented in C++ with MPI, utilized libSBML[1] and CVODES[2] libraries to set up the biology model and solve the ODEs respectively. The project originally comes from Prof. Andrew Millar of the centre for Systems Biology in Millar Laboratory, Edinburgh[3].