Tuesday 1 June 2012 1:00 - 2:00 pm (Room 115/63)
Speaker: Hiran Ganegedara
Title: Exploratory data analysis using scalable self-organising maps

Abstract:
There is a significant growth in the amount of data available for anaysis and decision making purposes. The Self-Organising Map (SOM) and the Growing Self-Organising Map (GSOM) are widely used unsupervised techniques to visualises the data set and are useful in identifying patterns in data. Finding interesting patterns from massive volumes of data could be highly time consuming and the time requirement will grow with the increase in the data quantity when SOM/GSOMs are used. Processing high volumes of data is a challenging task, given the limited computing power available in most computers. Recent developments in parallel and distributed computing techniques as well as multi-core CPU architectures have opened up a new avenue for large scale data processing by providing high volumes of computing power. This presentation aims at introducing a new technique which enables the SOM algorithm to scale with the number of computing resources. The presented technique will improve SOM/GSOM’s performance by several orders while maintaining the same level of accuracy.