Output details
13 - Electrical and Electronic Engineering, Metallurgy and Materials
University of Surrey
Incremental Linear Discriminant Analysis Using Sufficient Spanning Sets and Its Applications
The work was motivated by the need in industry to enhance the performance of deployed decision making systems, as more training data becomes available. Many commercial systems (such as OmniPerception face recognition technology) use Linear Discriminant Analysis (LDA) for decision-making, and a redesign of the decision making subsystem by batch learning is a time consuming hurdle. The solution developed allows incremental update of the LDA parameters with incoming training data. This is a major step forward, facilitating a continuous enhancement of system performance. The innovation has been integrated into OmniPerception's (Dr Messer on k.messer@omniperception.com) Collossus face retrieval engine.