Particle Swarm Optimization on HPC applied to Music Similarity
This talk will summarize our research for learning distance measures using Swarm Intelligence(SI) methods for music similarity. The use of SI can be quite time-consuming, as the likelihood of better results rises with the number of simulated particles in the swarm. To allow for simulations of 100000 and more particles, we parallelised an existing Particle Swarm Optimisation algorithm using the IPython python package. The code is now run — testing a new HPC cluster — via the PBS queuing system.
UPDATE: You can download the slides at: https://mirg.city.ac.uk/blog/wp-content/uploads/2014/08/Talk_PSO_Cluster_PDF.pdf