Cognitive radio researchers at Trinity College Dublin implemented a machine learning algorithm to dynamically determine the best network channel allocation in decentralised wireless network. Every network node applies this algorithm independently to establish links to other nodes in a dynamic spectrum setup. They wanted to test this algorithm in a Monte-Carlo simulation, applying game theory to determine the effect on the global network state for all possible network topologies. Using their sequential implementation, this simulation could take several days to complete.
The researchers used Xcelerit in order to boost the simulation performance. This case study shows how tremendous speedups were achieved using GPU accelerators. It enabled the researchers to optimise and tune their machine learning algorithm and get fast feedback about how it performs.