Phillip C.S.R. Kilgore, Marjan Trutschl, and Urška Cvek
Machine Learning, SOM, Visualization, Parallel Architectures
Self-organizing maps have been noted as useful tools for augmenting scientific data visualizations, particularly in the case where visualization of multidimensional data is required. Previous work has introduced visualization methods centered around Kohonen’s self-organizing map (SOM). However, a chief disadvantage associated with the SOM in this capacity is its large runtime complexity, which may result in impractical execution times in real-world use cases. We tested the feasibility of applying parallel execution to the self-organizing map in an attempt to reduce execution time. Though the our algorithms did not exhibit linear speedup in general, we discovered a notable exception with SmartJitter.
Important Links:
Go Back
