Home Publications       Business Assistance       Sign Up for Free Services     
 Quick Search    Search Word/Phrase
Tech Profiles
Special Reports
Media Resources
Business Focus Workshops
Technology Applications Reviews
Media Resources
Business Focus Workshops
Tech App Reviews
  Need a technology solution?

Then you've come to the right place. Search MDA-funded technology abstracts and
find the technology that is right for you.

Bookmark and Share  

   Technology Profile#227    10/22/2001
Related TechUpdate Article(s):
Maximum Likelihood Adaptive Neural System (MLANS) Algorithm


To a computer, telling the difference between a missile and decoy is not that different from telling the difference between a useful and a useless document on the Internet. Using this idea, Torch Concepts, Inc. (Dulles, VA), has taken the Maximum Likelihood Adaptive Neural System (MLANS) algorithm, which was originally developed for BMDO missile discriminiation, and extended it to the analysis of symbolic (e.g., textual) information. This extended technology, called ACUMEN, is currently being used to develop content management software. More than just a search engine, this software automatically analyzes documents returned by a search and sorts them into categories, labels the categories, and summarizes each document. Two private companies have already licensed the software and are developing products for political and educational markets.

Technology Description:

Nichols Research Corporation (NRC) originally developed the Maximum Likelihood Adaptive Neural System (MLANS) algorithm to analyze sensor data streams and adaptively classify and track objects of interest while suppressing background clutter and noise. To perform this function, MLANS incorporates the advantages of other intelligent software systems while avoiding their drawbacks. Adaptive pattern recognition systems and neural networks use techniques to adaptively classify objects without using prior knowledge about what the processor is looking for. These systems require extensive amounts of accurate data to learn what to identify. Expert systems, on the other hand, use a lot of prior information to know what to identify, but can’t adapt to better detect objects or identify new ones. MLANS combines the use of prior information with adaptive learning to identify objects as they appear.

MLANS can use complicated models developed by human experts or it can construct statistical models of what objects look like. The system then refines the models as more data are received. The program analyzes an image, pixel by pixel, or a signal, element by element, classifying the information in each unit according to its attributes. Values for each attribute are used in a multidimensional model that combines statistical properties with shape and motion to provide a “fuzzy classification” to what class of objects the pixel or element belongs. The MLANS algorithm achieves the Cramer-Rao bound, indicating the fastest possible learning and accuracy. The classification results generated by the software also achieve the minimum theoretical error limit, the Bayes’ Error. MLANS has been encoded in a variety of programming languages, including C, and runs on a variety of systems, including personal computers. For greater speed, the software supports parallel processing.

MDA Origins:

NRC developed MLANS while working on a variety of technology R&D programs for BMDO, the U.S. Army, the U.S. Air Force, and the Defense Advanced Research

Projects Agency. MLANS addressed BMDO’s need for automatic target recognition (in clutter and noise), sensor fusion, and data analysis, compression, and archival.

Spinoff Applications:

Because MLANS works equally well in analyzing any type of data, it has the flexibility to work in a wide variety of commercial applications. Examples of applications investigated include: fingerprint identification, financial market prediction, hospital performance evaluations, oil exploration, medical imaging, search and rescue, speech recognition, and road navigation on the intelligent highway.


In 1999, NRC merged with Computer Science Corporation, after which Roy J. Nichols (Vice Chairman and founder of NRC) and others formed a private company named innoVerity. Using MLANS technology, innoVerity created ACUMEN, a high-performance, content classification algorithm to categorize and characterize data. In 2001, the company changed its name to Torch Concepts.

Torch Concepts has adapted ACUMEN for content management and information mining applications. The company has created four new software tools that help companies to automate the identification, organization, retrieval, and delivery of information.

•Torch Organizer: This tool finds and organizes relevant content based on the concepts inherent in each document. When you perform a search, the tool automatically analyzes the documents returned by the search and sorts them into categories based on their concepts. You select the category that suits your needs, and with a click you get subcategories or a short summary of each document in the category. You find the information that is relevant to your needs quickly, without opening irrelevant documents.

•Torch Sorter: This tool helps companies organize documents for their corporate portal. It organizes large collections of documents into categories based upon the concepts contained in each document. If a document contains multiple concepts, the tool will cross-reference the document in each appropriate category.

•Torch Associator: This tool associates documents to existing categories by matching their concepts to the correct categories. It is especially appropriate for corporations and content providers who need to routinely file and distribute massive numbers of documents.

•Torch Agent: This tool works like a personal assistant. It finds and delivers information relevant to each user’s individual needs. It employs a unique method that enables users to have continual access to relevant documents. When a user feeds several sample documents into Torch Agent, it analyzes their content for the concepts therein, conducts an ongoing search to find new documents that match these concepts, and then retrieves and delivers the documents to the user’s desktop.

Other products being marketed by Torch Concepts include Torch Search, a search engine and spider, and Torch Imager, software that can scan thousands of documents daily and convert them to electronic format. In addition, the company sells the Torch Pak, an integrated solution including Torch Search, Torch Organizer, and Torch Agent.

Two private companies have already licensed the software. Reveal Technologies is developing a product, called My Virtual TextBook, that helps educators programmatically recognize and organize the topics and content of online documents. Educators will appreciate this product because it save time by automatically designing educational plans using the most up-to-date and relevant information found on the Internet. True Aim Technologies is developing the ActionButton for browsers. This button allows one-click access to additional relevant documents from anywhere online, saving time spent digging through internal databases and external Web sites. Torch Concepts is actively seeking additional licensees for its software.

Company Profile:

With over 15 employees, Torch Concepts provides small to mid-size clients with intelligent software applications to manage critical content in internal corporate databases and on the World Wide Web.

Contact Information:

Bill Roark
Torch Concepts, Inc.
21351 Ridgetop Circle, Suite 300
Dulles, VA 20166
Tel: 703-654-6007
Fax: 703-832-8988
email: roarkb@torchconcepts.com
web: www.torchconcepts.com

 Accessibility  Privacy Statement  Disclaimer  Site sponsored by National Technology Transfer Center - Washington Operations
© 2008 NTTC-WO     All rights reserved.