Abstract: In a pattern recognition system, input signals are applied to a short term feature representation field of nodes. A pattern from the short term feature representation field selects at least one category node in a category representation field. The selected category then generates a template pattern. With an insufficient match between the input pattern and template pattern, the category selection is reset. Category selection is based on selection weights which are initially set equal to long term memory weights. After reset, however, selections weights are reduced. Reduction is greatest at those nodes where excitation in F.sub.2 was greater prior to reset. The category representation field is of the same form as the field which receives the input and may itself serve as an input to a higher level pattern recognition system.

Abstract: An A pattern recognition subsystem responds to an A feature representation input to select A-category-representation and predict a B-category-representation and its associated B feature representation input. During learning trials, a predicted B-category-representation is compared to that obtained through a B pattern recognition subsystem. With mismatch, a vigilance parameter of the A-pattern-recognition subsystem is increased to cause reset of the first-category-representation selection. Inputs to the pattern recognition subsystems may be preprocessed to complement code the inputs.

Abstract: An improved ART2 network provides fast and intermediate learning. The network combines analog and binary coding functions. The analog portion encodes the recent past while the binary portion retains the distant past. LTM weights that fall below a threshold remain below threshold at all future times. The suprathreshold LTM weights track a time average of recent input patterns. LTM weight adjustment (update) provides fast commitment and slow recoding. The network incorporates these coding features while achieving an increase in computational efficiency of two to three orders of magnitude over prior analog ART systems.

Abstract: A self-categorizing pattern recognition system includes an adaptive filter for selecting a category in response to an input pattern. A template is then generated in response to the selected category and a coincident pattern indicating the intersection between the expected pattern and the input pattern is generated. The ratio between the number of elements and the coincident pattern to the number of elements in the input pattern determines whether the category is reset. If the category is not reset, the adaptive filter and template may be modified in response to the coincident pattern. Reset of the selected category is inhibited if no expected pattern is generated. Weighting of the adaptive filter in response to a coincident pattern is inversely related to the number of elements in the input pattern. The selected categories reset where a reset function is less than a vigilance parameter which may be varied in response to teaching events.

Abstract: A neural network includes a feature representation field which receives input patterns. Signals from the feature representative field select a category from a category representation field through a first adaptive filter. Based on the selected category, a template pattern is applied to the feature representation field, and a match between the template and the input is determined. If the angle between the template vector and a vector within the representation field is too great, the selected category is reset. Otherwise the category selection and template pattern are adapted to the input pattern as well as the previously stored template. A complex representation field includes signals normalized relative to signals across the field and feedback for pattern contrast enhancement.

Abstract: A masking field network F.sub.2, is characterized through systematic computer simulations serves or a content addressable memory. Masking field network F.sub.2 receives input patterns from an adaptive filter F.sub.1 .fwdarw.F.sub.2 that is activated by a prior processing level F.sub.1. The network F.sub.2 activates compressed recognition close that are predictive with respect to the activation patterns flickering across F.sub.1, and competitively inhibits, or masks, codes which are unpredictive with respect to the F.sub.1 patterns. The masking field can simultaneously detect multiple groupings within its input patterns and assign activation weights to the recognition codes for these groupings which are predictive with respect to the contextual information embedded within the patterns and the prior learning of the network. Automatic rescaling of sensitivity of the masking field as the overall size of an input pattern changes, allows stronger activation of a code for the whole F.sub.

Abstract: A neural network includes a feature representation field which receives input patterns. Signals from the feature representation field select a category from a category representation field through a first adaptive filter. Based on the selected category, a template pattern is applied to the feature representation field, and a match between the template and the input is determined. If the angle between the template vector and a vector within the representation field is too great, the selected category is reset. Otherwise the category selection and template pattern are adapted to the input pattern as well as the previously stored template. A complex representation field includes signals normalized relative to signals across the field and feedback for pattern contrast enhancement.

Abstract: A real-time network enables robots to accurately learn sensory motor transformation and to self-train and self-calibrate operating parameters after accidents or with wear. Combinations of visual and present position signals are used to relearn a target position map. Target positions in body-centered. visually activated coordinates are mapped into target positions in motor coordinates which are compared with present positions in motor coordinates to generate motor commands. Feedback provides calibrated error signals for adjustment of learned gain with changes in the system due to aging, accidents and the like. A series of prestored motor commands may be performed with a later "go" command.

Abstract: Network interactions within a Boundary Contour (BC) System, a Feature Contour (FC) System, and an Object Recognition (OR) System are employed to provide a computer vision system capable of recognizing emerging segmentations. The BC System is defined by a hierarchy of orientationally tuned interactions, which can be divided into two successive subsystems called the OC filter and the CC loop. The OC filter contains oriented receptive fields or masks, which are sensitive to different properties of image contrasts. The OC filter generates inputs to the CC loop, which contains successive stages of spatially shore-range competitive interactions and spatially long-range cooperative interactions. Feedback between the competitive and cooperative stages synthesizes a global context-sensitive segmentation from among the many possible groupings of local featural elements.