Parasom: An Efficient Self-Organizing Map for Parallel Multidimensional Input Processing and Clustering


This work utilizes a novel Self-Organizing Map, called ParaSOM, that is an advancement of the traditional SOM. The ParaSOM is different from existing architectures in that it processes the entire input space in parallel. It also utilizes a feature called a cover region, in which individual neurons “cover” whole regions of the input space, and not just a single vector. The effectiveness of the cover region is represented by its cover value. Neurons in the ParaSOM network have an age attribute. This attribute is used in a fashion similar to the Fritzke networks, from which it is borrowed. The age is a counter, indicating how long a neuron has existed in the vicinity of either sufficiently dense or sparse inputs. In dense input, the age is incremented after each epoch; in sparse input, the age is decremented. The architecture also incorporates a growth component. When a neuron is situated in an area that is dense with inputs for a sufficient period of time, as determined by the age threshold, it is understood that the neuron is well placed, and that this region is a good place to add another neuron. Conversely, neurons existing in sufficiently sparse input areas can be removed from the network. These features, along with others, allow the ParaSOM to execute epochs at a faster rate than other architectures, as well as converge in fewer epochs, and provide better metrics for input space coverage.

  • Abstract
  • Introduction
  • Parasom - Architecture and Features
  • Parasom - the Algorithm
  • Implementation
  • Tests and Results
  • Conclusions
  • References

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In