Optimisation of parallel k-d trees using heuristics for neuron touch detection task

Sammanfattning

Neuroscience has benefited from neuronal network simulation and an important task in the simulation is finding points in space where two neurites approach each other so a synapse could be formed. The task of finding the touching points could be seen as similar to the ray collision in ray tracing in computer graphics. This thesis aimed to investigate if the heuristics used in computer graphics and self-defined to speed up ray tracing can be used in the neurite touchpoint task. For analysis, we measured the time used for building the k-d trees (one per neuron), the time for querying and the memory usage. The tests were made using one specific neuron type called interneuron and realistic densities. This was made for simplicity, but the only difference with other types of neurons is the conditions for generating a touching point. It was found that due to their density, the interneurons do not benefit from these heuristics.