Home Contact Us Site Map
 
 
 

Products

Dave Johson - Program Manager Blind Equalization Source Recovery - Michigan Aerospace Corporation
To learn more about our software products, contact:
David K. Johnson, Ph.D.
734-975-8777 x140
davidjohnson@michiganaerospace.com
Brochures


MAC has developed Pattern Recognition packages that may be licensed:

  • SPADE - (Special Processing Applied to Data Exploitation) a data mining solution for anomaly and fault detection. The SPADE approach distills input data into highly quantized features (byte- or bit-valued) and then uses MAC's novel techniques for constructing Ensembles of Decision Trees (EDTs) in order to develop extremely accurate diagnostic and prognostic models for classification, regression, clustering, anomaly detection and semi-supervised learning tasks. A key feature of EDTs is their inherent potential for explaining why a given decision/classification is made. There are many significant advantages to the SPADE approach:

    • Completely data-driven;
    • Training is extremely fast and per-sample, on-line evaluation is theoretically orders of magnitude faster than conventional methods;
    • Operates effectively on very large data tables (millions of samples by a million features, or larger); and,
    • Proven to be as accurate as state-of-the-art techniques, if not more so, in many significant real-world applications.

  • Taiga - Next Generation software currently being implemented as the follow-up to SPADE. It will be a sophisticated tool capable of producing extremely accurate diagnostic and prognostic models on supervised and unsupervised problems for feature-set reduction, classification, regression, clustering, anomaly detection and semi-supervised tasks. Major innovations include:

    • Automatic intelligent data imports
    • Highly optimized internal data representations that enable the handling of datasets with more than a billion records and a million features
    • Genetic methods for tree growing in conjunction with Pareto techniques for multi-objective learning
    • Methods for transforming trees into Hilbert Spaces for advanced analysis and redundancy elimination
    • Powerful interactive visualization and drill-down capabilities to provide explanations for observed behavior
    • Evaluation methods for efficient real-time deployment on platforms such as Field-Programmable Gate Arrays, Digital Signal Processing chips and Graphical Processing Units.
 


Copyright © 2000-2014 Michigan Aerospace Corporation. All rights reserved.
Send any questions or comments to
webmaster@michiganaerospace.com