Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1-2, 5, 10, 13-15, 18, and 22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Shinvu, Improving misclassification for one class in a multi-class classification task, 2021 in view of Ratnam, Multioutput -Multiclass Classification, pp. 1-9, April 23 (Year: 2022).
With respect to claim 1, Shinvu teaches 1. A computer implemented method comprising: providing a multi-attribute classifier trained to classify a plurality of attributes” on pp. 1-2 (“Here I am trying to use 3 convolution layer neural network to classify a set of images. . . Black rot: 1180 Esca: 1383healthy: 423 leaf blight: 1076”); (Examiner finds 3 layer neural network (NN) teaches the multi attribute classifier; Examiner finds “Black rot, Esca, healthy, and leaf blight” teach the attributes);
“evaluating a performance of the multi-attribute classifier for classifying each attribute of the plurality of attributes” on p. 1 (“I have one class which is misclassified and I cannot understand what to do next. I have tried to reduce the value of dropout layer, but results got worst. [sic]”); (Examiner finds this passage teaches an evaluation that all but one of the attributes are misclassified);
“determining that the performance of the multi-attribute classifier for at least a particular attribute of the plurality of attributes falls below a defined standard” on p. 1 (“I have one class which is misclassified and I cannot understand what to do next. I have tried to reduce the value of dropout layer, but results got worst. [sic]”);(Examiner finds “misclassified” teaches falling below a standard; Examiner finds the particular attribute is the one class (i.e. black rot)); p. 2 heat map
PNG
media_image1.png
515
734
media_image1.png
Greyscale
(Examiner finds heat map/confusion matrix indicates black rot falls below threshold (i.e. because the shading is lighter));
“and responsive to determining that the performance of the multi-attribute classifier for at least the particular attribute of the plurality of attributes falls below the defined standard, causing training and generating of a single attribute classifier for classifying the particular attribute” on p. 3 (“Split into two networks; the first differentiates between Leaf Blight-Healthy-Black Rot/Esca, and the second differentiates between Black Rot-Esca”); (Examiner finds “the second differentiates between Black Rot-Esca” teaches training and generating the “single attribute classifier”);
“wherein the single attribute classifier is subsequently used in combination with the multi-attribute classifier for classifying the particular attribute of the plurality of attributes” on p. 3 (“Split into two networks; the first differentiates between Leaf Blight-Healthy-Black Rot/Esca, and the second differentiates between Black Rot-Esca”); (Examiner finds “the second differentiates between Black Rot-Esca” teaches training and generating the “single attribute classifier”); (Examiner finds “split into two networks” teaches using each network in combination).
It appears Shinvu fails to explicitly teach “wherein the multi-attribute classifier is a multi-output classifier configured to simultaneously produce a plurality of classification outputs, each classification output corresponding to a respective attribute of the plurality of attributes.”
However, Ratnam teaches “wherein the multi-attribute classifier is a multi-output classifier configured to simultaneously produce a plurality of classification outputs, each classification output corresponding to a respective attribute of the plurality of attributes” on p. 2
Multi- output : Yes, there will be multiple outputs (2 or more) for a single feature set( a set of independent values)
One data point can belong to one or more labels. For example: when building movie genre prediction model, the model can classify one movie into more than one label, since a movie can be action, thriller ,can be both action and thriller.
Ratnam and Shinvu are analogous art because they are from the same field of endeavor as the claimed invention.
It would have been obvious to one skilled in the art before the effective filing date of the invention to modify the multi-attribute classifier in Shinvu to include “wherein the multi-attribute classifier is a multi-output classifier configured to simultaneously produce a plurality of classification outputs, each classification output corresponding to a respective attribute of the plurality of attributes” as taught by Ratnam.
The motivation would have been to quickly classify data that can be in more than one category (e.g. movie that is both action and thriller). See id.
Claim 14 and claim 22 are rejected for the same reason as claim 1 above.
With respect to claim 2, Shinvu teaches “2. The method of claim 1 wherein the defined standard is a threshold associated with training criteria for the multi-attribute classifier and evaluating the performance includes:
applying the multi-attribute classifier to a test data set to determine a performance score for each said attribute” on
p. 1
Here I am trying to use 3 convolution layer neural network to classify a set of images (train data: (3249) , validation data: (487), test data: (326))
I have one class which is misclassified and I cannot understand what to do next. I have tried to reduce the value of dropout layer, but results got worst.
p. 2, heat map/confusion matrix (Examiner finds the heat map at top of page 2 teaches a performance score for each attribute):
PNG
media_image1.png
515
734
media_image1.png
Greyscale
“and comparing the performance score for at least the particular attribute to the threshold for the multi-attribute classifier in classifying attributes” on p. 1
Here I am trying to use 3 convolution layer neural network to classify a set of images (train data: (3249) , validation data: (487), test data: (326))
I have one class which is misclassified and I cannot understand what to do next. I have tried to reduce the value of dropout layer, but results got worst.
p. 2, heat map (Examiner finds predicted black rot falls below threshold of other attributes based on shading, for example);
PNG
media_image1.png
515
734
media_image1.png
Greyscale
(Examiner finds the shading teaches a comparison of performance between the attributes).
Claim 15 is rejected for the same reason as claim 2 above.
With respect to claim 5, Shinvu teaches “5. The method of claim 3, wherein the new data set is a subset of a larger data set, and the original training data set was generated by filtering that larger data set based on a criterion” on p. 2
I had split the two datasets as follow[s]:
PNG
media_image2.png
229
1315
media_image2.png
Greyscale
(Examiner finds the original training data set (x_train, y_train) is filtered from the larger, initial data set by definition; that is, training data, test data, and validation data are all subsets of a larger initial data set by definition; new data set is any one of validation or test sets (x_valid, x_test, etc.). Claim 18 rejected for this same reason.
With respect to claim 10 Shinvu teaches “10. The method of claim 1, the method further comprising: providing a system comprising the multi-attribute classifier and the single attribute classifier; and in response to the particular attribute having the performance below the defined standard for the multi-attribute classifier, the system configured for applying the single attribute classifier instead of the multi-attribute classifier for classifying the particular attribute” on p. 3 (emphasis added):
Split into two networks; the first differentiates between Leaf Blight-Healthy-Black Rot/Esca, and the second differentiates between Black Rot-Esca.
(In order to determine the classification of “Black Rot or Esca” the single NN classifier is used).
With respect to claim 13, Shinvu teaches “13. The method of claim 1, wherein the multi-attribute classifier is a multi-class multi- output classifier wherein each said attribute of the plurality of attributes input to the multi- attribute classifier is associated with a separate classification output. “ on p. 3
Split into two networks; the first differentiates between Leaf Blight-Healthy-Black Rot/Esca, and the second differentiates between Black Rot-Esca.
(multi-output is (1)Leaf Blight, (2) Healthy, or (3) Black Rot/Esca; (1),(2),and (3) are attributes of the NN multi classifier).
Claim(s) 3, 7, 16, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Shinvu in view of Ratnam, as applied to claim 1 and claim 14 above and further in view of Fernandez-Navarro, A dynamic over-sampling procedure based on sensitivity for multi-class problems, 2011.
With respect to claim 3, it appears Shinvu fails to explicitly teach
“3. The method of claim 1, wherein the single attribute classifier is caused to be trained with a new data set different from an original training data set for the multi-attribute classifier”
However, Fernandez-Navarro, A dynamic over-sampling procedure based on sensitivity for multi-class problems, teaches “3. The method of claim 1, wherein the single attribute classifier is caused to be trained with a new data set different from an original training data set for the multi-attribute classifier” in the abstract (Examiner finds the new, different data set is the dataset that results from the oversampled minority (minimum sized) class; Examiner finds this minority class teaches the single attribute; Examiner further finds that the minority class in Fernandez-Navarro is at least one); p. 1826 left column first full paragraph (“. . .selects the minimum size class. . . “)
Fernandez-Navarro, and Shinvu are analogous art because they are from the same field of endeavor as the claimed invention.
It would have been obvious to one skilled in the art before the effective filing date of the invention to modify the single attribute classifier in Shinvu “to be trained with a new data set different from an original training data set for the multi-attribute classifier” as suggested by Fernandez-Navarro,
The motivation would have been to increase the accuracy of the single classifier. See Fernandez-Navarro, abstract.
Claim 16 is rejected for the same reason as claim 3 above.
With respect to claim 7, it appears Shinvu fails to explicitly teaches“7. The method of claim 3, further comprising: determining that the particular attribute is related to a particular category of object and causing the new data set to contain samples related to the particular attribute for categories of available objects other than the particular category of object in addition to samples related to the particular category of object.”
However, Fernandez-Navarro teaches “determining that the particular attribute is related to a particular category of object and causing the new data set to contain samples related to the particular attribute for categories of available objects other than the particular category of object in addition to samples related to the particular category of object” in the abstract (new minority data set has both attributes of available object and “other” objects).
Shinvu and Fernandez-Navarro are analogous art because they are from the same field of endeavor as the claimed invention.
It would have been obvious to one skilled in the art before the effective filing date of the invention to modify the particular attribute in Shinvu et al. to include “determining that the particular attribute is related to a particular category of object and causing the new data set to contain samples related to the particular attribute for categories of available objects other than the particular category of object in addition to samples related to the particular category of object” as taught by Fernandez-Navarro. The motivation would have been to increase the accuracy of the single classifier. See Fernandez-Navarro, abstract.
Claim 20 is rejected for this same reason.
Claim(s) 8, 9, 11, and 21 is/are rejected under 35 U.S.C. 103 as being unpatentable over Shinvu in view of Ratnam as applied to claim 1 and claim 14 above and further in view of Rodriguez US 20230237369 A1.
With respect to claim 8, it appears Shinvu fails to explicitly teach “The method of claim 1 wherein the defined standard is a precision constraint value.”
However, Rodriguez US 20230237369 A1 teaches “The method of claim 1 wherein the defined standard is a precision constraint value” in para. 52 (“Indeed, during their experiments (e.g., using a “testing” dataset and a separate “production” dataset), the present inventors implemented an embodiment described herein: after initial training, the machine learning classifier exhibited a recall rate of 32%, a precision rate of 29%, and an area-under-curve score of 0.78; however, after deletion of weakly-predictive feature categories and retraining, the machine learning classifier exhibited a recall rate of 96%, a precision rate of 49%, and an area-under-curve score of 0.99”).
Shinvu and Rodriguez are analogous art because they are from the same field of endeavor as the claimed invention.
It would have been obvious to one skilled in the art before the effective filing date of the invention to modify the defined standard in Shinvu to include “wherein the defined standard is a precision constraint value.
The motivation would have been to increase the accuracy of the model. See Rodriguez para. 52.
Claim 21 is rejected for the same reason as claim 8 above.
With respect to claim 9, it appears Shinvu fails to explicitly teach “9. The method of claim 1 wherein the defined standard includes a precision constraint value and a recall value; and wherein determining that the performance for the particular attribute falls below the defined standard includes: determining that the performance for the particular attribute meets the precision constraint value; and responsive to determining that the performance for the particular attribute meets the precision constraint value.
However, Rodriguez teaches “wherein the defined standard includes a precision constraint value and a recall value” in para. 52
the present inventors implemented an embodiment described herein: after initial training, the machine learning classifier exhibited a recall rate of 32%, a precision rate of 29%, and an area-under-curve score of 0.78; however, after deletion of weakly-predictive feature categories and retraining, the machine learning classifier exhibited a recall rate of 96%, a precision rate of 49%, and an area-under-curve score of 0.99. Clearly, various embodiments described herein constitute concrete and tangible technical improvements in the field of machine learning classification, and thus such embodiments certainly qualify as useful and practical applications of computers.
(Examiner finds the recall value is anything over 32% and the precision constraint value is anything above 29%).
“and wherein determining that the performance for the particular attribute falls below the defined standard includes: determining that the performance for the particular attribute meets the precision constraint value”
the present inventors implemented an embodiment described herein: after initial training, the machine learning classifier exhibited a recall rate of 32%, a precision rate of 29%, and an area-under-curve score of 0.78; however, after deletion of weakly-predictive feature categories and retraining, the machine learning classifier exhibited a recall rate of 96%, a precision rate of 49%, and an area-under-curve score of 0.99. Clearly, various embodiments described herein constitute concrete and tangible technical improvements in the field of machine learning classification, and thus such embodiments certainly qualify as useful and practical applications of computers.
“and responsive to determining that the performance for the particular attribute meets the precision constraint value.”
the present inventors implemented an embodiment described herein: after initial training, the machine learning classifier exhibited a recall rate of 32%, a precision rate of 29%, and an area-under-curve score of 0.78; however, after deletion of weakly-predictive feature categories and retraining, the machine learning classifier exhibited a recall rate of 96%, a precision rate of 49%, and an area-under-curve score of 0.99. Clearly, various embodiments described herein constitute concrete and tangible technical improvements in the field of machine learning classification, and thus such embodiments certainly qualify as useful and practical applications of computers.
(Examiner finds the precision constraint value is anything above 29%).
Rodriguez and Shinvu are analogous art because they are from the same field of endeavor as the claimed invention.
It would have been obvious to one skilled in the art before the effective filing date of the invention to modify the defined standard in Shinvu to include “wherein the defined standard includes a precision constraint value and a recall value; and wherein determining that the performance for the particular attribute falls below the defined standard includes: determining that the performance for the particular attribute meets the precision constraint value; and responsive to determining that the performance for the particular attribute meets the precision constraint value” as taught by Rodriguez.
The motivation would have been to increase the accuracy and improve the performance of the classifier. See Rodriguez para. 52.
With respect to claim 11, it appears Shinvu fails to explicitly te4ach “11. The method of claim 2, wherein training of the single attribute classifier additionally comprises evaluating performance for the single attribute classifier to determine whether meets the defined standard and responsive to evaluating that the performance fails to meet the defined standard causing re-training of the single attribute classifier. “
However, Rodriguez teaches “11. The method of claim 2, wherein training of the single attribute classifier additionally comprises evaluating performance for the single attribute classifier to determine whether meets the defined standard and responsive to evaluating that the performance fails to meet the defined standard causing re-training of the single attribute classifier” in para. 52:
[0052] Moreover, various embodiments of the subject innovation can integrate into a practical application various teachings described herein relating to automated training of machine learning classification for patient missed care opportunities or late arrivals. As explained above, existing techniques can cause machine learning classifiers to achieve quite low performance metrics. The present inventors realized that such low performance metrics are often caused because existing techniques train machine learning classifiers on irrelevant and/or weakly-predictive features (e.g., such machine learning classifiers can become distracted by irrelevant and/or weakly-predictive features). Accordingly, the present inventors devised various embodiments described herein. Specifically, various embodiments described herein can train a machine learning classifier on a set of annotated data candidates. Furthermore, after such training, various embodiments described herein can analyze (e.g., analytically and/or via artificial intelligence) the trained, updated, and/or optimized internal parameters of the machine learning classifier, so as to rank the feature categories that define the set of annotated data candidates in order of classification importance. In various cases, various embodiments described herein can delete from the set of annotated data candidates any feature categories whose ranks (e.g., whose classification importance scores) fail to satisfy any suitable threshold value. Accordingly, various embodiments described herein can retrain the machine learning classifier on the set of annotated data candidates after such deletion, which can cause the machine learning classifier to achieve significantly improved performance (e.g., the weakly-predictive features can be no longer present to distract and/or bog down the machine learning classifier). Indeed, during their experiments (e.g., using a “testing” dataset and a separate “production” dataset), the present inventors implemented an embodiment described herein: after initial training, the machine learning classifier exhibited a recall rate of 32%, a precision rate of 29%, and an area-under-curve score of 0.78; however, after deletion of weakly-predictive feature categories and retraining, the machine learning classifier exhibited a recall rate of 96%, a precision rate of 49%, and an area-under-curve score of 0.99. Clearly, various embodiments described herein constitute concrete and tangible technical improvements in the field of machine learning classification, and thus such embodiments certainly qualify as useful and practical applications of computers.
Rodriguez and Shinvu are analogous art because they are from the same field of endeavor as the claimed invention.
It would have been obvious to one skilled in the art before the effective filing date of the invention to modify the “training of the single attribute classifier” Shinvu to include “wherein training of the single attribute classifier additionally comprises evaluating performance for the single attribute classifier to determine whether meets the defined standard and responsive to evaluating that the performance fails to meet the defined standard causing re-training of the single attribute classifier”” as taught by Rodriguez.
The motivation would have been to increase the accuracy and improve the performance of the classifier. See Rodriguez para. 52.
Claim(s) 12 is/are rejected under 35 U.S.C. 103 as being unpatentable over Shinvu in view of Ratnam.
With respect to claim 12, Shinvu fails to explicitly teach “12. The method of claim 1, further comprising: tracking performance of the multi-attribute classifier together with one or more single attribute classifiers to determine whether to generate additional single attribute classifiers based on the tracking, wherein each single attribute classifier is used for classifying a respective associated attribute instead of the multi-attribute classifier.
However, Examiner finds “tracking performance of the multi-attribute classifier together with one or more single attribute classifiers to determine whether to generate additional single attribute classifiers based on the tracking, wherein each single attribute classifier is used for classifying a respective associated attribute instead of the multi-attribute classifier” would have been obvious to one skilled in the art based on KSR rationale E. See MPEP 2143(I)(E).
Examiner finds (1) before the effective filing date of the invention, there was a recognized design need to solve a problem. That is, there was a need to solve the problem of a multiclass classifier failing to classify a particular attribute. See Shinvu p. 1.
Examiner further finds (2) before the effective filing date of the invention there were a finite number of identified, predictable potential solutions to the recognized need or problem. Examiner finds that the predicable potential solutions were that each of the classifiers in a neural network could fall below a particular threshold based on a number of factors including data drift. Examiner finds the finite number of solutions were to improve any classifier that fell below a certain threshold by retraining the model. See generally Shinvu pages 1-2.
Examiner further finds (3) one of ordinary skill in the art could have pursued the known potential solutions with a reasonable expectation of success. That is, one skilled in the art would have pursued “tracking performance of the multi-attribute classifier together with one or more single attribute classifiers to determine whether to generate additional single attribute classifiers based on the tracking, wherein each single attribute classifier is used for classifying a respective associated attribute instead of the multi-attribute classifier.” This would have improved performance of any of the classifications in a neural network when new training data was available to the neural network and any one of the classifiers failed to classify new attribute of the new data.
Allowable Subject Matter
Claims 4, 6, 17, and 19 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Response to Argument
Applicant argues
1. Shinvu Does Not Disclose a Multi-Attribute Classifier
Independent claim 1 recites "providing a multi-attribute classifier trained to classify a plurality of attributes", which is not taught or suggested by Shinvu.The Examiner maps the "multi-attribute classifier" recited in claim 1 to Shinvu's three-layer convolutional neural network that classifies leaf images into one of four disease classes: Black Rot, Esca, Healthy, and Leaf Blight. Office Action at 2. Applicant does not agree. The classifier in Shinvu produces a single classification output selecting one label from among multiple candidate classes.
The quoted claim language "providing a multi-attribute classifier trained to classify a plurality of attributes” does not require assigning two labels (attributes) to one sample. Applicant’s argument is therefore not persuasive.
Applicant further argues the following:
Although Shinvu may output scores for multiple disease classes, those scores correspond to alternative labels for a single classification task. Accordingly, Shinvu discloses, at most, a multi-class classifier. It does not disclose the claimed multi-attribute classifier, which is a multi-output classifier configured to simultaneously produce a plurality of classification outputs, each classification output corresponding to a respective attribute of the plurality of attributes.
As described in paragraph [2] of the present specification "a multi-attribute classification model will have n outputs, where n is the number of classes associated with that model ". By way of illustration, and as described in paragraph [88] of the specification, a multi- attribute classifier applied to clothing images may simultaneously predict a color attribute (e.g., red, blue, green), a size attribute (e.g., small, medium, large), and a product type attribute (e.g., shoes, shirts, hats), all as classification outputs from a single model. No such multi-output, multi-attribute architecture is disclosed or suggested by Shinvu. As amended, independent claims 1, 14, and 22 now explicitly recite that the multi-attribute classifier is "a multi-output classifier configured to simultaneously produce a plurality of classification outputs, each classification output corresponding to a respective attribute of the plurality of attributes." This limitation should unambiguously distinguish the claimed invention from Shinvu's model.
Examiner agrees that Shinvu does not explicitly teach “wherein the multi-attribute classifier is a multi-output classifier configured to simultaneously produce a plurality of classification outputs, each classification output corresponding to a respective attribute of the plurality of attributes” However, this argument is rendered moot by the new ground of rejection above.
Applicant further argues:
Accordingly, Applicant submits that the feature "providing a multi-attribute classifier trained to classify a plurality of attributes" within the meaning of claim 1 is not anticipated by Shinvu.
Examiner disagrees that Shinvu does not teach “providing a multi-attribute classifier trained to classify a plurality of attributes.” Shinvu teaches classifying a plurality of attributes such as Black Rot, Esca, Healthy using a trained classifier. Examiner agrees Shinvu fails to explicitly teach “wherein the multi-attribute classifier is a multi-output classifier configured to simultaneously produce a plurality of classification outputs, each classification output corresponding to a respective attribute of the plurality of attributes.” However, this argument is rendered moot by the new ground of rejection above.
Applicant further argues
2. Shinvu Does Not Disclose Training and Generating a Single Attribute Classifier
Used "In Combination With" the Multi-Attribute Classifier
Independent claim 1 also recites "causing training and generating of a single attribute classifier for classifying the particular attribute, wherein the single attribute classifier is subsequently used in combination with the multi-attribute classifier for classifying the particular attribute of the plurality of attributes", which is not taught or suggested by Shinvu. The Office Action maps the "single attribute classifier" and its use "in combination with the multi-attribute classifier" to a suggestion on Shinvu's page 3 to "[s]plit into two networks; the first differentiates between Leaf Blight-Healthy-Black Rot/Esca, and the second differentiates between Black Rot-Esca." Office Action at pp. 3-4. This mapping is deficient for several independent reasons.
First, the Shinvu passage is a suggestion by a community respondent on a question-and- answer website. It is not a description of any implemented system or method.
The prior art is not required to teach “an implemented system” in order to anticipate or render obvious the claim language. This argument is not persuasive.
Applicant further argues
The suggestion proposes entirely replacing the original single classifier with two new, different classifiers that collectively cover the original class space. This is fundamentally different from the claimed approach of using a targeted single attribute classifier in combination with a multi-attribute classifier to address specific underperforming attribute(s).
Applicant concludes this “suggestion” is fundamentally different from the claimed approach without quoting any specific claim language that the prior art purportedly fails to teach.
Applicant further argues
Second, neither of the "two networks" described in the Shinvu suggestion constitutes a "single attribute classifier" within the meaning of the claims. The first network (which differentiates among Leaf Blight, Healthy, and Black Rot/Esca) is itself a multi-class classifier. The second network (which differentiates between Black Rot and Esca) is also a multi-class classifier that distinguishes between two classes. None of them is a single attribute classifier for classifying the particular attribute.
Applicant is arguing that the prior art does not use the same terminology as the claimed invention. This argument is not persuasive. Here, the classifier that distinguishes between Black Rot and Esca is a single attribute classifier in that it classifies a sample into one attribute—Black Rot or Esca.
Applicant further argues
Third, the Shinvu suggestion does not teach using the second network "in combination with" the original classifier. Rather, the original classifier is discarded entirely and replaced by the two classifiers.
Shinvu does not teach discarding the classifiers entirely. See p. 3 (“If you can’t augment your data to have a better distribution across the 4 classes, here are some ideas. . . “). Also, it is clear the two networks used “in combination” is a modification of the first classifier (i.e. Leaf-Bligh-Healthy-BlackRot/Esca) and a second classifier (Black-Rot Esca). See page 3. As such, Applicant’s argument is not persuasive.
Applicant further argues
In contrast, the claimed invention requires the single attribute classifier to be subsequently used in combination with the multi-attribute classifier.
Shinvu teaches using Black-Rot Esca classifier and the Leaf-Bligh-Healthy-BlackRot/Esca classifiers in combination. See Shinvu page 3. Applicant’s argument is not persuasive.
Applicant further argues
Accordingly, Applicant submits that the feature "causing training and generating of a single attribute classifier for classifying the particular attribute, wherein the single attribute classifier is subsequently used in combination with the multi-attribute classifier for classifying the particular attribute of the plurality of attributes" as recited in claim 1 is not anticipated by Shinvu.
This argument is rendered moot by the new grounds of rejection above.
Applicant further argues
3. Shinvu Does Not Disclose Evaluating Performance Against a Defined
Standard
Independent claim 1 further recites "determining that the performance of the multi-attributeclassifier for at least a particular attribute of the plurality of attributes falls below a defined standard", which is not taught or suggested by Shinvu. The Office Action maps the "defined standard" to the Shinvu user's informal observation that "I have one class which is misclassified." Office Action at p. 3. The Office Action further maps the comparison of performance to the standard to the lighter shading in a normalized confusion matrix. Office Action at p. 4.
These mappings do not constitute anticipation of the claimed "defined standard." The claims require a "defined" standard against which the computer determines the performance of the multi-attribute classifier for at least a particular attribute of the plurality of attributes to fall below. Shinvu's informal observation is not a defined performance standard; it is merely an after-the-fact subjective observation by a user that one class performed worse than others.
Examiner finds Shinvu suggests conventional machine learning performance standards related to whether an attribute is misclassified. For example, a misclassification of sample would have suggested to one of ordinary skill in the machine learning arts at the time of filing a “defined standard” such as accuracy and/or precision.
Applicant further argues
Accordingly, Applicant submits that the feature "determining that the performance of the multi-attribute classifier for at least a particular attribute of the plurality of attributes falls below a defined standard " as recited in claim 1 is not anticipated by Shinvu. In view of the foregoing, Shinvu does not teach or suggest all the claim limitations as set forth in independent claim 1 as amended.Accordingly, claim 1 is patentable over Shinvu, and withdrawal of the § 102 rejection of claim 1 is respectfully requested.
This argument is not persuasive. Shinvu teaches and/or suggests this element at least on page 1 (“I have one class which is misclassified and I cannot understand what to do next. I have tried to reduce the value of dropout layer, but results got worst. [sic]”). Examiner finds “misclassified” teaches falling below a standard. Examiner finds the particular attribute is the one class (i.e. black rot); See page 2 heat map .
Applicant’s remaining arguments are rendered moot by the new grounds of rejection above.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALBERT M PHILLIPS, III whose telephone number is (571)270-3256. The examiner can normally be reached 10a-6:30pm EST M-F.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Ann J Lo can be reached at (571) 272-9767. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/ALBERT M PHILLIPS, III/ Primary Examiner, Art Unit 2159
Prosecution Insights