US6505181B1
Expired / Lapsed
US
⏱️ 0.0 years remaining
Recognition system
Assignee: Xylon LLC
•
Filing: 1997-10-16
•
Granted: 2003-01-07
•
Expires: 2017-10-16
Abstract
Abstract A recognition system of the self-organizing artificial neural network type is arranged to classify input data according to stored categories which have been determined by a training process. In the training process the initial category representations are selectively iteratively updated in response to a series of training patterns and in accordance with a competitive learning routine. this routine uses measures of category utilization based on the proportion of all inputs received over a representative period, particularly long term utilisation and short term utilization, to ensure that all available categories will be used and that the system is stable. The training rate which determines the amount of modification to a category representation at an up-date is local to each category and is based upon the maturity of the category and on the similarity measure between the internal representative pattern and the training input so that the training duration can be minimized. A user-operated selectively-operable suggestion learning input is provided to each category to modify the training process or to enable secondary training to proceed during classification of input data using that input data as the training patterns. The categories are represented by multiple reference patterns with respective importance values from which the degree of compatibility between an input and a category is computed taking into account the importance values.
Inventors
- 👤 Robert Bartholemew Lambert
- 👤 Richard John Fryer
- 👤 William Paul Cockshott
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-25 00:38
CPC Classifications
G; G06; G06V; G06V10/00; G06V10/70; G06V10/764; G06N; G06N3/00; G06N3/02; G06N3/08; G06N3/088; G06V10/40; G06V10/46; G06V10/469; G06V10/82; G06T; G06T2207/00; G06T2207/20; G06T2207/20084
Claims Analysis
Total Claims: 11
Claim 1
Independent Claim
229 words
A system for classifying data vectors into at least one of a plurality of categories respectively defined by stored vector representations which have been determined by a training process, the system comprising a plurality of interconnected processing elements which, for implementing the training process, include: input means for receiving input training vectors; storage means for storing vector representations of the categories; calculating means for calculating a measure of similarity between an input vector and the stored representations for each category; and means for selecting stored vector representations and then modifying the selected stored vector representations so as to re-define the categories on an iterative basis during the training process; wherein the selecting and modifying means comprises for the processing elements of each category: means for recording a plurality of different measures of learning for the category based on previous modifications to the stored vector representation of that category with respect to the input vectors which have been received by the system; means for combining said recorded measures of learning to form an activation value for the category; means for evaluating a category strength value from said activation value and the respective measure of similarity; and means for receiving the category strength values from the processing elements of other categories in the system and deciding, based upon comparative category strengths, whether or not the selected category representation should be modified.
Claim 2
Independent Claim
229 words
A system for classifying data vectors into at least one of a plurality of categories respectively defined by stored vector representations which have been determined by a training process, the system comprising a plurality of interconnected processing elements which, for implementing the training process, include: input means for receiving input training vectors; storage means for storing vector representations of the categories; calculating means for calculating a measure of similarity between an input vector and the stored representations for each category; and means for selecting stored vector representations and then modifying the selected stored vector representations so as to re-define the categories on an iterative basis during the training process; wherein the selecting and modifying means comprises for the processing elements of each category: means for recording a plurality of different measures of learning for the category based on previous modifications to the stored vector representation of that category with respect to the input vectors which have been received by the system; means for combining said recorded measures of learning to form an activation value for the category; means for evaluating a category strength value from said activation value and the respective measure of similarity; and means for receiving the category strength values from the processing elements of other categories in the system and deciding, based upon comparative category strengths, whether or not the selected category representation should be modified.
Claim 3
Dependent Claim
58 words
A system as claimed in claim 1 , wherein the recorded measures of learning monitor long term category learning and short term category learning, the long term measure recording how frequently the category has previously had its stored vector representation modified, and the short term measure recording how recently the category has had its stored vector representation modified.
Claim 4
Dependent Claim
20 words
A system as claimed in claim 2 , wherein the short term measure augments non-linearly with each unused input vector.
Claim 5
Dependent Claim
22 words
A system as claimed in claim 2 or claim 3 , wherein the combination of said recorded measures is a product combination.
Claim 6
Dependent Claim
41 words
A system as claimed in claim 1 , wherein the amount by which categories may be modified is local to each category and is based both on the maturity of the category and on the measure of similarity of the category.
Claim 7
Dependent Claim
41 words
A system as claimed in claim 5 , wherein each category is represented by multiple reference vector representations each having an associated important value corresponding to the degree to which the representation can be regarded as typical of the overall category.
Claim 8
Dependent Claim
34 words
A system as claimed in claim 6 , wherein each internal reference representation has a local training rate based upon its own importance value and upon the maturity of the category as a whole.
Claim 9
Dependent Claim
38 words
A system as claimed in claim 1 , wherein the processing elements are arranged in a multi-layered arrangement and a suggestive learning input is provided to each category, the suggestive learning input being selectively operable by a user.
Claim 10
Independent Claim
191 words
A method for training a classification system wherein each of the input vectors in a sequence of input training vectors is classified into at least one of a plurality of categories, respectively defined by stored vector representations, by comparing each input vector with the plurality of stored vector representations and selectively re-defining the categories by modifying at least some of the stored vector representations after the comparison and prior to comparison of the next input vector in the sequence, wherein in order to effect said modification of stored vector representation, the system (a) records a plurality of different measures of learning for each category based on previous modifications to the stored vector representation of that category with respect to the input vectors which have been received by the system; (b) quantifies the similarity in the current comparison for each category; (c) combines the measures from steps (a) and (b) into a strength factor for each category; (d) correlates and compares the numerous strength factors for all the categories; (e) identifies those categories having a strength factor which fall into a pre-determined class; and (f) modifies the categories identified by step (e).
Claim 11
Independent Claim
191 words
A method for training a classification system wherein each of the input vectors in a sequence of input training vectors is classified into at least one of a plurality of categories, respectively defined by stored vector representations, by comparing each input vector with the plurality of stored vector representations and selectively re-defining the categories by modifying at least some of the stored vector representations after the comparison and prior to comparison of the next input vector in the sequence, wherein in order to effect said modification of stored vector representation, the system (a) records a plurality of different measures of learning for each category based on previous modifications to the stored vector representation of that category with respect to the input vectors which have been received by the system; (b) quantifies the similarity in the current comparison for each category; (c) combines the measures from steps (a) and (b) into a strength factor for each category; (d) correlates and compares the numerous strength factors for all the categories; (e) identifies those categories having a strength factor which fall into a pre-determined class; and (f) modifies the categories identified by step (e).
References Cited (Backward Citations)
US5265224A
Cited
US5517597A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2006-01-06 | 2006-07-08 | 2007-01-07 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2010-01-06 | 2010-07-08 | 2011-01-07 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2014-01-06 | 2014-07-08 | 2015-01-07 | $7700.00 | Window Closed / Lapsed |