Computational Approaches For The Estimation Of Extensive Biomedical Complex Networks

Complex network approaches find an increasing application for the investigation of fundamental problems in neuroscience, medicine and biology. Depending on the dimension and structure, the complexity of the network computation could be extremely resource intensive, long-winded and could last several days. This study investigates the performance gain that can be achieved through parallelism and compares different optimization approaches for (multi-core) CPU or CPU calculations. Benchmarks were performed simulating scenarios, typical for exploration of real brain networks on distributed source level. The results show that the calculations benefit mainly from the combination of standard optimization techniques and parallelization on multi-core CPUs. OpenCL is an alternative, if the measure-algorithm is inherently parallelizable with a high number of processes.