Pattern classification using linear programming
AssigneeArizona Board of Regents acting on behalf of Arizona State University
InventorAsim Roy
Life of this patent
Estimating the term…
Loading patent…
AssigneeArizona Board of Regents acting on behalf of Arizona State University
InventorAsim Roy
Estimating the term…
A method and apparatus (system) is provided for the separation into and the identification of classes of events wherein each of the events is represented by a signal vector comprising the signals x1, x2, . . . , xj, . . . , xn. The system in principle is different from neural networks and statistical classifiers. The system comprises a plurality of assemblies. The training or adaptation module stores a set of training examples and has a set of procedures (linear programming, clustering and others) that operate on the training examples and determine a set of transfer function and threshold values. These transfer functions and threshold values are installed on a recognized module for use in the identification phase. The training module is extremely fast and can deal with very difficult pattern classification problems that arise in speech and image recognition, robotics, medical diagnosis, warfare systems and others. The systems also exploits parallelism in both the learning and recognition phases.
An information processing system for separating a plurality of patterns into a discrete set of classes, each of said patterns having a plurality of input signals disposed therewithin, said system comprising:
Dependencies are inferred from the claim text.