DETAILED ACTION
This communication is in response to the amendment filed 10/29/25 in which claims 1, 4, 13, and 20 were amended, and claims 3, 5, 15 were canceled. Claims 1, 4, 6-13, and 16-20 are pending.
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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 10/29/25 has been entered.
Response to Arguments
Applicant, in pertinent part, argues:
In rejecting claim 1 (and similarly rejecting claim 13), page 8 of the Office action alleges that Henning discloses "[i]dentifying a quantile of defect values corresponding to one of the defect values; and assigning the input vector to a cluster label of the plurality of first cluster labels based on the quantile of defect values (but see Henning pg. 4849 Introduction ('Quantile-based clustering represents the clusters by optimally chosen quantiles.'), pg. 4852."
However, as best understood, the cited portions of Henning only appear to disclose, at best, the clustering of points (apparently equated with the recited "defect values") into quantiles, but are silent as to assigning any alleged input vector, separate from the clustered points, to a particular label, much less assigning any alleged input vector to a particular label based on any alleged quantiles.
Moreover, even in Chu, the entire wafer inspection data/original data 30 (apparently equated with the recited "input vectors" and "defect values" collectively) is clustered, and then assigned grades thereto. However, the cited portions of Chu do not appear to disclose any alleged assigning the original data 30, separate from the original data 30 that was clustered, to any particular label based on the clustering. (See e.g., Chu, paras. [0058]-[0062]).
Applicant’s arguments have been considered. However, Applicant’s arguments are moot in light of the new reference to Blowers and remapping of the argued limitation. Specifically, Abdulaal in combination with Blowers teaches cluster-labeled vector training data that may be used to train a supervised machine learning model. Henning further teaches quantile-based clustering, a specific type of clustering technique, which can be used to cluster the vector training data taught by Abdulaal and Blowers.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or non-obviousness.
Claims 1, 6, 9-13, 16, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Abdulaal (US 2020/0226490 A1; published Jul. 16, 2020) in view of Covell (US 8,712,930 B1; patented Apr. 29, 2014), Powers, S., et al., "Customized training with an application to mass spectrometric imaging of cancer tissue. arXiv [stat. AP] (2016) [online] (“Powers”), Blowers (US 2012/0271782 A1; published Oct. 25, 2012), and Hennig, Christian, Cinzia Viroli, and Laura Anderlucci. "Quantile-based clustering." (2019): 4849-4883 (“Henning”).
Regarding claim 1, Abdulaal discloses [a] method of training a system for making predictions relating to products manufactured via a manufacturing process, the method comprising: (see ¶ 2 (anomaly detection))
receiving, by a processor of the system, a plurality of input vectors and a plurality of defect values corresponding to the plurality of input vectors from a data collection circuit, (FIG. 1 (real-time machine metrics as input) ¶ 26 (“It is to be appreciated that while the technologies disclosed herein are primarily described in the context of identifying anomalies in multi-dimensional machine metrics that are indicative of incidents impacting infrastructure components, the technologies described herein can be utilized to identify anomalies (“defect values”) in other types of data in other configurations.”), ¶ 27 (“The ML classifier training 102 is performed using unlabeled training data that includes multi-dimensional machine metrics generated by infrastructure components.”))
identifying, by the processor, a plurality of first cluster labels corresponding to the plurality of input vectors based on the defect values; (¶ 36 (“Once the unlabeled training data 202 has been pre-processed, the pre-processed training data 210 can be clustered and boosted using a weak unsupervised learner 214. This process, which is described in greater detail below, clusters the pre-processed training data 210 to generate cluster-labeled training data 222.”), ¶ 37 (“The cluster-labeled training data 222 includes labels indicating the probability (i.e. the size of a cluster relative to the size of the full data set) that a particular cluster of training data is anomalous. Detected anomalies can first be used as incident indicators, when no other indicators are available (e.g. when historical labels are unavailable). This changes over time and updates periodically as users send their feedback through the UI described below.”))
training, by the processor, a cluster classifier based on the input vectors and the corresponding first cluster labels; (¶ 43 (“Once the training data 202 has been clustered in the manner described above, post-processing 224 can be performed on the cluster-labeled training data 222. In one configuration, for example, a supervised learner 236 can perform supervised machine learning on the cluster-labeled training 222 data to train a ML classifier 104. For example, in one configuration the supervised learner 236 fits the cluster-labeled training data to a classification tree. Other supervised ML techniques can be utilized in other configurations such as, but not limited to, a support vector machine (“SVM”).”)).
Although Abdulaal teaches once the ML classifier has been trained it can be deployed in the production environment for use in classification of real-time machine metrics (“input vectors”), see ¶ 28, Abdulaal does not expressly disclose reassigning, by the processor, the input vectors to a plurality of second cluster labels based on outputs of the cluster classifier, the outputs having a one-to-one correspondence with the input vectors (but see Covell 6:25—52 (“The cluster labeling module 240 assigns a label to each cluster generated by the clustering module 230. The cluster labeling module 240 performs an initial assignment of labels to the clusters. The initial assignment of labels to the clusters can be random. Each exemplar of a cluster is considered to be mapped to the label corresponding to the cluster. The machine learning module 220 uses the mapping from exemplars to the labels to generate training data sets for the prediction model 175. The machine learning module 220 initially trains the prediction model 175 using the initial assignment of the clusters to the labels which could be random. A random assignment from clusters to labels may not be learnable by the generated prediction model 175. As a result, the prediction model may not accurately predict the assigned labels when applied to exemplars from the cluster. However, the cluster labeling module 240 reassigns the labels to the cluster based on criteria that improves predictability of a prediction model 175 generated from the new mapping. In an embodiment, the reassignment of the labels is based on the prediction model 175 generated from the previous assignment. For example, the prediction model 175 generated by the previous assignment is used to predict labels for exemplars from each cluster. The predicted labels obtained from each cluster are used to determine the reassignment of the labels. The machine learning module 220 trains the prediction model 175 based on the new assignment from clusters to labels. This process is repeated iteratively in order to improve the prediction model 175.”)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Abdulaal to incorporate the teachings of Covell to reassign the labels of the cluster-labeled data based on a prediction model generated from the previous assignment, at least because the previous assignment may not have been correctly predicted the labels.
Abdulaal further discloses retraining, by the processor, the cluster classifier based on the input vectors and the second cluster labels; and (¶ 30 (“In one configuration, additional training data is obtained through human confirmation 114 or rejection of classifications made by the ML classifier 104. In these configurations, an interface can be provided through which a user can confirm or reject classifications of real-time machine metrics 108 made by the ML classifier 104. For example, and without limitation, data identifying real-time machine metrics 108 classified as an anomaly cluster of high incident likelihood can be presented in a UI. A user can then provide an indication by way of the UI indicating whether the real-time machine metrics 108 indicate or do not indicate an incident. This indication can be utilized to perform additional supervised training of the ML classifier 104 such as, for example, updating 116 incident probability inferences generated using Bayesian learning.”), ¶ 43 (“Once the training data 202 has been clustered in the manner described above, post-processing 224 can be performed on the cluster-labeled training data 222. In one configuration, for example, a supervised learner 236 can perform supervised machine learning on the cluster-labeled training 222 data to train a ML classifier 104. For example, in one configuration the supervised learner 236 fits the cluster-labeled training data to a classification tree. Other supervised ML techniques can be utilized in other configurations such as, but not limited to, a support vector machine (“SVM”).”).
Abdulaal teaches performing supervised machine learning on the cluster-labeled training data to train ML classifier 104 and performing classification of pre-processed machine metrics 210A (including metrics obtained by further splitting of clusters) to determine if machine metrics correspond to a cluster labeled as being anomalous. See Abdulaal ¶¶ 42-43, 65. Yet, Abdulaal does not expressly disclose training, by the processor, a plurality of machine learning models corresponding to the second cluster labels, wherein each of the plurality of machine learning models is configured to predict a defect value based on one or more input vectors corresponding to one of the second cluster labels. However, Powers teaches customized training in which a test set is partitioned into subsets and a separate machine learning model is trained to make predictions for each subset. See Abstract. In particular, each subset of the test set uses only its own “customized” subset of the training set to fit each model. Powers 5.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Abdulaal to incorporate the teachings of Powers to train a separate machine learning model for each cluster in order to make predictions if machine metrics corresponding to a cluster are anomalous, at least because doing so would lead to a model that is locally linear but rich globally. Powers 5.
Abdulaal does not expressly teach a system for making predictions relating to products manufactured via a manufacturing process and that the plurality of input vectors comprising trace data, and the plurality of defect visibility values of products of the manufacturing process corresponding to the trace data (but see Blowers ¶ 15 (“Both within the paper industry, as well as in other manufacturing environments, various research efforts (9) (10) have explored using neural networks to model the process dynamics. Some research has demonstrated the ability of time-delay neural networks to capture the dynamics of the process. Others (11) (8) have explored the possibility of knowledge based neural network models.”), ¶ 51 (“In the present invention, a computer means monitor and industrial-like process or any other system or process which is amenable to being monitored. Sensor measurements are made at various points in the process being monitored and communicated via common computer data channels or media to the computer means. Within the computer a software program executes the appropriate computer-implementable steps so as to assemble the sensor measurements or object parameters into a vector, or feature vector, so the entire vector represents an observation from a multivariate population [input vectors comprising trace data]. When applied to a real world multi-sensor environment, the operation and internal complexity of the process becomes represented in terms of the collection of vectors which describe the different observations, or states, at different points in time. When applied to an object, the features which add the most value to discriminate one set of objects from another become that objects feature vector. A feature vector is an n-dimensional vector of numerical features that come from the various data collection points. The information contained in such feature vector may include (1) well defined (structured) parameters describing an object to be characterized; (2) sensor measurements assembled into a vector, so the entire vector represents an observation from a multivariate population at different points in time.”)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Abdulaal to incorporate the teachings of Blowers to assemble the multivariate sensor measurements into an input vector or a feature vector, at least because doing so would represent an observation from a multivariate population.
Abdulaal ¶ 26 teaches that the training data is in the form of multi-dimensional metrics and Abdulaal ¶ 36 teaches generating cluster-labeled training data from the unlabeled training data. As explained above, Blowers further teaches representing multivariate trace data from an industrial manufacturing process as a collection of vectors for the purpose of training a classifier. Thus, Abdulaal as modified by Blowers teaches cluster-labeled vectors generated from multivariate sensor measurements. However, the references do not specifically disclose quantile-based clustering as the clustering technique used to generate the cluster-labeled training data and thus does not specifically disclose wherein identifying the plurality of first cluster labels comprises: identifying a quantile of defect values corresponding to one of the defect values; and assigning the input vector to a cluster label of the plurality of first cluster labels based on the quantile of defect values (but see Henning pg. 4849 Introduction (“Quantile-based clustering represents the clusters by optimally chosen quantiles.”), pg. 4852 Quantile-based clustering (“We now introduce a new clustering strategy based on the idea of assigning points to the closest quantile. Measuring “closeness” by the squared Euclidean distance is associated with the mean, in the sense that means optimize (1). Quantiles can also be characterized by minimizing a sum of discrepancies, although these discrepancies are not symmetric; they depend on which side of the quantile a point is. Using these discrepancies in “K-means style” leads to a simple clustering method that allows for within-cluster skewness.” [points are assigned to quantiles identified by minimizing a sum of discrepancies])).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Abdulaal to incorporate the teachings of Henning to use quantile-based clustering instead of k-means clustering at least because doing so is “faster and simpler than many modern clustering methods, at the same time being more flexible than K-means.” Henning pg. 4849-50.
Claims 13 and 20 are respective method and apparatus claims corresponding to claim 1 and are similarly rejected.
Regarding claim 6, Abdulaal, in view of Covell, Powers, Blowers, and Henning, discloses the invention of claim 1 as discussed above. Abdulaal further discloses wherein the reassigning the input vectors to the plurality of second cluster labels comprises:
inputting the input vectors to the cluster classifier; (FIG. 1 (real-time machine metrics input to the machine learning classifier))
receiving the plurality of second cluster labels from the cluster classifier as outputs in response to the inputting of the input vectors; and (FIG. 1 (classification of machine metrics to identify anomalies 110))
assigning the input vectors to corresponding ones of the plurality of second cluster labels (FIG. 1 (classification of machine metrics to identify anomalies 110)).
Claim 19 is a method claim corresponding to claim 6 and is similarly rejected.
Regarding claim 9, Abdulaal, in view of Covell, Powers, Blowers, and Henning, discloses the invention of claim 1 as discussed above. Abdulaal further discloses wherein the training the cluster classifier comprises:
inputting, by the processor, the input vectors and the corresponding first cluster labels as training data to the cluster classifier; and (FIG. 2 (cluster-labeled training data is fed into supervised learner))
training, by the processor, the cluster classifier to identify the first cluster labels given the input vectors using a supervised machine learning algorithm (FIG. 2 (supervised learner is trained to generated machine learning classifier)).
Regarding claim 10, Abdulaal, in view of Covell, Powers, Blowers, and Henning, discloses the invention of claim 1 as discussed above. Abdulaal further discloses wherein the retraining the cluster classifier comprises:
inputting, by the processor, the input vectors and the corresponding second cluster labels as training data to the cluster classifier; and (FIG. 2 (cluster-labeled training data is input to a supervised learner))
training, by the processor, the cluster classifier to identify the second cluster labels given the input vectors using a supervised machine learning algorithm (¶ 43 (“Once the training data 202 has been clustered in the manner described above, post-processing 224 can be performed on the cluster-labeled training data 222. In one configuration, for example, a supervised learner 236 can perform supervised machine learning on the cluster-labeled training 222 data to train a ML classifier 104.”)).
Regarding claim 11, Abdulaal, in view of Covell, Powers, Blowers, and Henning, discloses the invention of claim 1 as discussed above. Abdulaal further discloses wherein the training the plurality of machine learning models comprises:
training one of the plurality of machine learning models based on ones of the input vectors within a same cluster label of the second cluster labels and corresponding ones of the defect values (FIG. 2A (Bayesian Learning 230), ¶ 46 (“In some configurations, labels 228 are available for instances of the training data 202 and 210. For, example, labels can be collected in the manner described above with regard to FIG. 1 using human confirmation 114. The labels 228 can indicate whether a particular cluster is representative of an incident. In these configurations, Bayesian learning 230 can be performed on the cluster-labeled training data 222 using the training data labels 228 to assign incident probability inferences to the clustered-labeled training data 222.”)).
Claim 16 is a method claim corresponding to claim 11 and is similarly rejected.
Regarding claim 12, Abdulaal, in view of Covell, Powers, and Henning, discloses the invention of claim 1 as discussed above. Abdulaal further discloses wherein a cluster label of the plurality of first cluster labels is different from a corresponding cluster label of the plurality of second cluster labels (¶ 41 (“During boosting 212, cluster assessment 216 is performed on the clusters of training data in order to determine if each cluster is a candidate for splitting into multiple clusters. A completeness score is then computed for clusters to identify candidates for splitting. The completeness score indicates whether a cluster is complete and should not be split or is not complete and should be split. As will be described in great detail below, the completeness score for each cluster can be computed by determining whether the distance between instances of training data in selected cluster are approximately similar in length to a median non-zero distance between instances of training data in the cluster.”) (different clusters have different labels)).
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Abdulaal, Covell, Powers, Blowers, and Henning as applied to claim 1 above, and further in view of Chu (US 2020/0380655 A1; published Dec. 3, 2020).
Regarding claim 4, Abdulaal, in view of Covell, Powers, Henning, and Chu, discloses the invention of claim 3 as discussed above. Abdulaal does not expressly disclose wherein the trace data comprise multivariate sensor data from a plurality of sensors used in the manufacturing process. However, Chu teaches collecting source data from a wafer manufacturing process, the source providing facilities including various devices for continuously generating data such as each facility in a production line, inspection equipment for each process, inspection equipment for a finished product, a sensor, an IoT device, or the like. ¶ 60. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Abdulaal to incorporate the teachings of Chu to obtain the multi-dimensional machine metrics as source data from a manufacturing process, at least because doing so would enable utilizing the technology “to identify anomalies in other types of data in other configurations. Other configurations will be apparent to those of ordinary skill in the art.” Abdulaal ¶ 26.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Bu (US 2022/0237433 A1) DATA DRIVEN RECOGNITION OF ANOMALIES AND CONTINUATION OF SENSOR DATA.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SHAHID K KHAN whose telephone number is (571)270-0419. The examiner can normally be reached M-F, 9-5 est.
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, Andrew Jung can be reached on (571)270-3779. 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.
/SHAHID K KHAN/Primary Examiner, Art Unit 2146