Patent US5805731A successfully refreshed from Google Patents.
US5805731A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Adaptive statistical classifier which provides reliable estimates or output classes having low probabilities
Assignee: Apple Inc
•
Filing: 1995-08-08
•
Granted: 1998-09-08
•
Expires: 2015-08-08
Abstract
Abstract A statistical classifier for pattern recognition, such as a neural network, produces a plurality of output signals corresponding to the probabilities that a given input pattern belongs in respective classes. The classifier is trained in a manner such that low probabilities which pertain to classes of interest are not suppressed too greatly. This is achieved by modifying the amount by which error signals, corresponding to classes which are incorrectly identified, are employed in the training process, relative to error signals corresponding to the correct class. As a result, output probabilities for incorrect classes are not forced to a low value as much as probabilities for correct classes are raised.
Inventors
- 👤 Larry S. Yaeger
- 👤 Richard F. Lyon
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-04 16:59
CPC Classifications
G; G06; G06F; G06F18/00; G06F18/20; G06F18/24; G06F18/241; G06F18/2415; G06F18/24155; G06F18/21; G06F18/214
Claims Analysis
Total Claims: 6
Claim 1
Independent Claim
181 words
A method for training a statistical classifier, comprising the following steps: selecting a training sample from a collection of training samples, each such training sample being associated with a label class from a predetermined set of distinct classes; providing data pertaining to said training sample as an input signal to the classifier; processing said data within the classifier in accordance with weight values to produce a plurality of output signals which respectively correspond to different classes in said predetermined set of distinct classes, providing a plurality of target signals which respectively correspond to different classes in said predetermined set of distinct classes, wherein the target signal corresponding to said label class is assigned a first predetermined signal value, and the others of said target signals are assigned a second predetermined signal value; determining error signals corresponding to each of said distinct classes, based on differences between said output signals and said target signals; multiplying said error signal which corresponds to said label class by a factor β, where β>1; and adjusting said weight values in accordance with said modified error signals.
Claim 2
Independent Claim
181 words
A method for training a statistical classifier, comprising the following steps: selecting a training sample from a collection of training samples, each such training sample being associated with a label class from a predetermined set of distinct classes; providing data pertaining to said training sample as an input signal to the classifier; processing said data within the classifier in accordance with weight values to produce a plurality of output signals which respectively correspond to different classes in said predetermined set of distinct classes, providing a plurality of target signals which respectively correspond to different classes in said predetermined set of distinct classes, wherein the target signal corresponding to said label class is assigned a first predetermined signal value, and the others of said target signals are assigned a second predetermined signal value; determining error signals corresponding to each of said distinct classes, based on differences between said output signals and said target signals; multiplying said error signal which corresponds to said label class by a factor β, where β>1; and adjusting said weight values in accordance with said modified error signals.
Claim 3
Independent Claim
206 words
A method for training a statistical classifier, comprising the following steps: selecting a training sample from a collection of training samples, each such training sample being associated with a label class from a predetermined set of distinct classes: providing data pertaining to said training sample as an input signal to the classifier; processing said data within the classifier in accordance with weight values to produce a plurality of output signals which respectively correspond to different classes in said predetermined set of distinct classes; providing a plurality of target signals which respectively correspond to different classes in said predetermined set of distinct classes, wherein the target signal corresponding to said label class is assigned a first predetermined signal value, and the others of said target signals are assigned a second predetermined signal value; determining error signals corresponding to each of said distinct classes, based on differences between said output signals and said target signals; adjusting said weight values in accordance with said error signals such that the adjustment of said weight values in accordance with the error signal corresponding to said label class has a greater effect than the adjustment of said weight values in accordance with error signals corresponding to classes other than said label class.
Claim 4
Independent Claim
206 words
A method for training a statistical classifier, comprising the following steps: selecting a training sample from a collection of training samples, each such training sample being associated with a label class from a predetermined set of distinct classes: providing data pertaining to said training sample as an input signal to the classifier; processing said data within the classifier in accordance with weight values to produce a plurality of output signals which respectively correspond to different classes in said predetermined set of distinct classes; providing a plurality of target signals which respectively correspond to different classes in said predetermined set of distinct classes, wherein the target signal corresponding to said label class is assigned a first predetermined signal value, and the others of said target signals are assigned a second predetermined signal value; determining error signals corresponding to each of said distinct classes, based on differences between said output signals and said target signals; adjusting said weight values in accordance with said error signals such that the adjustment of said weight values in accordance with the error signal corresponding to said label class has a greater effect than the adjustment of said weight values in accordance with error signals corresponding to classes other than said label class.
Claim 5
Independent Claim
118 words
A method for training a statistical classifier, comprising the steps of: providing data pertaining to a predetermined class as input data to the classifier; processing the input data within the classifier in accordance with weight values to produce a plurality of output signals which respectively correspond to different classes; determining error values corresponding to each of said different classes, based on said output signals; summing the error values corresponding to all of the classes other than said predetermined class; multiplying the summed error values by an attenuation factor less than 1; combining the attenuated, summed error values with the error value corresponding to said predetermined class; and adjusting said weight values in accordance with the combined error values.
Claim 6
Independent Claim
118 words
A method for training a statistical classifier, comprising the steps of: providing data pertaining to a predetermined class as input data to the classifier; processing the input data within the classifier in accordance with weight values to produce a plurality of output signals which respectively correspond to different classes; determining error values corresponding to each of said different classes, based on said output signals; summing the error values corresponding to all of the classes other than said predetermined class; multiplying the summed error values by an attenuation factor less than 1; combining the attenuated, summed error values with the error value corresponding to said predetermined class; and adjusting said weight values in accordance with the combined error values.
References Cited (Backward Citations)
US5161204A
Cited
US5577166A
Cited
US5337371A
Cited
US5204914A
Cited
US5555317A
Cited
US5392363A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2001-09-07 | 2002-03-09 | 2002-09-08 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2005-09-07 | 2006-03-09 | 2006-09-08 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2009-09-07 | 2010-03-09 | 2010-09-08 | $7700.00 | Window Closed / Lapsed |