Constrained Fisher Scoring for a Mixture of Factor Analyzers

Report No. ARL-TR-7836
Authors: Gene T Whipps, Emre Ertin, and Randolph L Moses
Date/Pages: September 2016; 44 pages
Abstract: This report considers the problem of learning an object appearance manifold using a spatially distributed network of sensors. Sensor nodes observe an object from different aspects and then learn a joint statistical model for the object manifold. We employ a mixture of factor analyzers model and derive a Fisher scoring method for maximum-likelihood estimation of the model parameters. We analyze convergence of the scoring method and derive stopping conditions for exiting the iterative algorithm. Simulation examples demonstrate that the proposed approach provides faster model learning over the popular expectation-maximization algorithm with similar computational requirements. Lastly, we demonstrate the efficacy of the proposed method for learning a global appearance model across the entire sensor network.
Distribution: Approved for public release
  Download Report ( 0.795 MBytes )
If you are visually impaired or need a physical copy of this report, please visit and contact DTIC.
 

Last Update / Reviewed: September 1, 2016