Self-organizing neural network for pattern classification
First Claim
1. A neural network system for classifying an input pattern, comprising:
- a) means for comparing the input pattern to templates of already learned patterns, each template having a class associated with said template;
b) means for selecting a class for the input pattern based on results of the comparison by the means for comparing;
c) means for determining whether a correct class was selected by the means for selecting for the input pattern;
d) means, operative in response to a determination that a correct class was selected for the input pattern, for determining whether the correct class has been stably classified by comparing a template of the selected class which is most similar to the input pattern to a template of another class which is most similar to the input pattern among templates of said another class to determine whether the compared templates are similar to a degree that classification may not be accurate; and
e) means, operative in response to a determination that the correct class has not been stably classified, for adjusting the compared templates to ensure stable classification.
0 Assignments
0 Petitions
Accused Products
Abstract
A neural network includes a plurality of input nodes for receiving the respective elements of the input vector. A copy of all of the elements of the input vector is sent to the next level of nodes in the neural network denoted as intermediate nodes. The intermediate nodes each encode a separate template pattern. They compare the actual input pattern with the template and generate a signal indicative of the difference between the input pattern and the template pattern. Each of the templates encoded in the intermediate nodes has a class associated with it. The difference calculated by the intermediate nodes is passed to an output node for each of the intermediate nodes at a given class. The output node then selects the minimum difference amongst the values sent from the intermediate nodes. This lowest difference for the class represented by the output node is then forwarded to a selector. The selector receives such values from each of the output nodes of all of the classes and then selects that to output value which is a minimum difference. The selector in turn, generates a signal indicative of the class of the intermediate node that sent the smallest difference value.
37 Citations
63 Claims
-
1. A neural network system for classifying an input pattern, comprising:
-
a) means for comparing the input pattern to templates of already learned patterns, each template having a class associated with said template; b) means for selecting a class for the input pattern based on results of the comparison by the means for comparing; c) means for determining whether a correct class was selected by the means for selecting for the input pattern; d) means, operative in response to a determination that a correct class was selected for the input pattern, for determining whether the correct class has been stably classified by comparing a template of the selected class which is most similar to the input pattern to a template of another class which is most similar to the input pattern among templates of said another class to determine whether the compared templates are similar to a degree that classification may not be accurate; and e) means, operative in response to a determination that the correct class has not been stably classified, for adjusting the compared templates to ensure stable classification. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22)
-
-
23. A neural network system for classifying an input pattern, comprising:
-
a) means for comparing the input pattern to templates of already learned patterns, each template having a class associated with said template; b) means for selecting a class for the input pattern based on results of the comparison by the means for comparing; c) means for determining whether a correct class is selected for the input pattern; d) means, responsive to a signal indicative of the correct class and of classes of the templates of already learned patterns, for determining whether a template of a correct class exists; and e) means, operative in response to a determination that the correct class was not selected for the input pattern and a determination that a template of the correct class does not exist, for creating a new template based on the input pattern for a new class to be used by the means for comparing. - View Dependent Claims (24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34)
-
-
35. A neural network for classifying an input vector into a class comprising:
-
a. an input buffer for storing the input vector; b. means for storing templates of already learned patterns of respective classes, each template holding a pattern associated with a particular class; c. means for comparing the input vector Which is stored in the input buffer with each of the templates which is stored in said means for storing templates and for generating an output signal for each comparison with a template, wherein said output signal indicates a difference between the input signal vector and the template; d. means for selecting for each class an output signal from output signals that are generated for a respective class, said means for selecting receiving all of the output signals from the means for comparing which indicate results of comparisons performed by the means for comparing with templates of the class and selecting the output signal having a smallest difference; e. means for receiving the selected output signals for each class, and for determining among the selected output signals which output signal has the lowest value, and for determining the class of said means for selecting that selected the lowest value output signal, and for generating an output indicative of the class of said means for selecting that sent the lowest value output signal; f. means for receiving a signal indicative of a correct class of the input vector; and g. means for learning for iteratively updating the template which is stored in the means for storing templates so that the neural network learns to correctly classify the input vector, said means for leaning including means for determining whether the output of the means for receiving the selected output signal indicated the correct class, and means, responsive to a determination that the correct class was selected for the input vector, for determining whether the correct class has been stably classified by comparing a template of the correct class which is most similar to the input vector to a template of another class which is most similar to the input vector among templates of said another class, to determine whether the compared templates are similar to a degree that classification may not be accurate, responsive to a determination that the correct class has not been stably classified, for adjusting the compared templates. - View Dependent Claims (36, 37, 38, 39)
-
-
40. In a neural network, a method of classifying an input pattern of physical data, comprising the steps of:
-
a. comparing the input pattern of physical data with templates of patterns for each class of input to yield a value indicative of the difference between the template and the input pattern; b. selecting a minimum value yielded by the comparisons with templates of patterns of each class as a class minimum value for the class; c. selecting a smallest value among the class minimum values; d. classifying an input signal as having a class identical to the class of a template that yielded the smallest value among the class minimum values; e. determining whether the input signal was correctly classified; and f. comparing the template that yielded the smallest value among the class minimum values and adjusting the compared templates when the compared templates are similar to a degree that classification may not be accurate. - View Dependent Claims (41, 42, 43, 44, 45, 46, 47, 48, 49, 50)
-
-
51. A neural network system for classifying an input pattern, comprising:
-
a) means for comparing the input pattern to templates of already learned patterns, each template having a class associated with said template; b) means for selecting a class for the input pattern based on results of the comparison by the means for comparing; c) means for determining whether a correct class was selected by the means for selecting for the input pattern; d) means, operative in response to a determination that the correct class was selected for the input pattern, for determining whether the correct class has been stably classified; e) means, operative in response to a determination that the correct class has not been stably classified, for adjusting templates of already learned pattern so as to ensure stable classification; f) means, operative in response to a determination that the correct class was not selected for the input pattern, for determining a reason for incorrect classification; g) means, operative in response to a determination of the reason for incorrect classification, for adjusting templates according to the determined reason to improve a likelihood of correct classification; and h) means, operative in response to a failure to determine the reason for incorrect classification, for adjusting the template of only the correct class which is most similar to the input pattern. - View Dependent Claims (52, 53)
-
-
54. A neural network system for classifying an input pattern, comprising:
-
a) means for comparing the input pattern to templates of already learned patterns, each template having a class associated with said template; b) means for selecting a class for the input pattern based on results of the comparison by the means for comparing; c) means for determining whether a correct class is selected for the input pattern; d) means, operative in response to a determination that the correct class was not selected for the input pattern, for determining whether the correct class has a template sufficiently close to the input pattern to permit correct classification; and e) means, operative in response to a determination that the correct class does not have a template sufficiently close to the input pattern to permit correct classification, for creating a new template for the correct class based on the input pattern. - View Dependent Claims (55)
-
-
56. A neural network system for classifying an input pattern, comprising:
-
a) means for comparing the input pattern to templates of already learned patterns, each template having a class associated with said template; b) means for selecting a class for the input pattern based on results of the comparison by the means for comparing; c) means for determining whether a correct class was selected by the means for selecting for the input pattern; d) means, operative in response to a determination that the correct class was selected for the input pattern, for determining whether the correct class has been stably classified by comparing a template of the selected class which is most similar to the input pattern to a template of another class which is most similar among templates of said another class to the input pattern to determine whether the templates are similar to a degree that classification may not be accurate; and e) means, operative in response to a determination that the correct class was not selected for the input pattern, for determining whether the correct class has been stably classified by comparing a template of the selected class which is most similar to the input pattern to a template of the correct class which is most similar to the input pattern among templates of the correct class to determine whether the templates are similar to a degree that classification may not be accurate; and f) means, operative in response to a determination that the correct class has not been stably classified, for adjusting the compared templates so as to ensure stable classification.
-
-
57. In a neural network, a method of classifying an input pattern of physical data, comprising the steps of:
-
a. comparing the input pattern of physical data with templates of patterns for each class of input to yield a value indicative of the difference between the template and the input pattern; b. selecting a minimum value yielded by the comparisons with templates of each class as a class minimum value for the class; c. selecting a smallest value among the class minimum values; d. classifying an input signal as having a class identical to the class of a template that yielded the smallest value among the class minimum values; e. determining whether the input signal was correctly classified; f. determining whether a template of a correct class exists; and g. creating a new template based on the input pattern for a new class to be used in the step of comparing when a template of the correct class does not exist and the input pattern is incorrectly classified. - View Dependent Claims (58, 59, 60)
-
-
61. In a neural network, a method of classifying an input pattern of physical data, comprising the steps of:
-
a. comparing the input pattern of physical data with templates of patterns for each class of input to yield a value indicative of the difference between the template and the input pattern; b. selecting a minimum value yielded by the comparisons with templates of each class as a class minimum value for the class; c. selecting a smallest value among the class minimum values; d. classifying an input signal as having a class identical to the class of a template that yielded the smallest value among the class minimum values; e. determining whether the input signal was correctly classified; and f. determining whether the correct class has a template sufficiently close to the input pattern to permit correct classification; and g. creating a new template for the correct class based on the input pattern when the correct class does not have a template sufficiently close to the input pattern to permit correct classification. - View Dependent Claims (62)
-
-
63. In a neural network, a method of classifying an input pattern of physical data, comprising the steps of:
-
a. comparing the input pattern of physical data with templates of patterns for each class of input to yield a value indicative of the difference between the template and the input pattern; b. selecting a minimum value yielded by the comparisons with templates of each class as a class minimum value for the class; c. selecting a smallest value among the class minimum values; d. classifying an input signal as having a class identical to the class of a template that yielded the smallest value among the class minimum values; e. determining whether the input signal was correctly classified; and f. determining whether the correct class has been stably classified; g. adjusting templates of already learned patterns so as to ensure stable classification when the correct class has not been stably classified; h. determining a reason for any incorrect classification; i. adjusting templates according to any determined reason for incorrect classification to improve a likelihood of correct classification; and j. adjusting the template of only the correct class which is most similar to the input pattern when no reason can be determined for any incorrect classification.
-
Specification