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 .
Response to Amendment
The previous claim objections are withdrawn due to Applicant’s amendments.
Response to Arguments
Applicant’s arguments filed 03/02/2026 regarding the rejection under 35 USC 103 with respect to claims 1-3, 5-11, 13-17 and 19-22 have been fully considered but are not persuasive.
Beginning on page 15, Applicant asserts that the combination of Privault, Duesterwald, Li, Davis, Sanderson and Lee do not teach “a unique metadata value n of the N unique metadata values comprises one of: an identifier of an annotator of a plurality of annotators associated with the multiple annotated samples, a timestamp of a plurality of timestamps associated with the multiple annotated samples, or an identifier of a software tool of a plurality of software tools associated with the multiple annotated samples.” However, Privault teaches “a unique metadata value n of the N unique metadata values comprises one of: an identifier of an annotator of a plurality of annotators associated with the multiple annotated samples,” in paragraph [0083]: “The classification labels assigned to the documents by the classifiers M1, M2, M3, . . . Mi are compared with those provided by the original reviewers R1, R2, R3, . . . Ri, respectively.” R1, R2, R3 are the identifiers of the annotators for the documents.
Applicant’s arguments regarding the rejection under 35 USC 103 with respect to claims 1-3, 5-11, 13-17 and 19-22 concerning the limitation “retaining a second set of annotated samples that is not associated with the unique metadata value n” have been fully considered but are moot. New reference Elisha has been incorporated below to teach the newly presented limitations.
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.
Claims 1, 2, 3, 6, 7, 8, 9, 10, 11, 14, 15, 16, 17, 20 and 22 are rejected under 35 U.S.C. 103 as being unpatentable over Privault et al. (US 20100312725 A1); hereinafter Privault in view of Duesterwald et al. (US 20190042953 A1); hereinafter Duesterwald in view of Li et al. (Learning to Learn from Noisy Labeled Data); hereinafter Li in view of Elisha et al. (US 20210256420 A1); hereinafter Elisha
Claim 1 is rejected over Privault, Duesterwald, Li and Elisha.
Regarding claim 1, Privault teaches a computer-implemented method, comprising:
obtaining a training set comprising multiple annotated samples that are associated with an annotation metadata category comprising N unique metadata values, wherein
N is at least 2,
a unique metadata value n of the N unique metadata values comprises one of:
an identifier of an annotator of a plurality of annotators associated with the multiple annotated samples, (Privault [0083]: “The classification labels assigned to the documents by the classifiers M1, M2, M3, . . . Mi are compared with those provided by the original reviewers R1, R2, R3, . . . Ri, respectively.”)
each annotated sample of the multiple annotated samples comprises one or more annotations,
the unique metadata value n is associated with a first set of annotated samples of the multiple annotated samples, (Privault [0080]: “At S110, a classifier model M1, M2, M3, . . . Mi 62 is learned for each reviewer/subset reviewed. Steps S108 and S110 are described in further detail below with reference to FIG. 3. Each classifier model M1, M2, M3, . . . Mi built for a respective user from the set of documents labeled by this user can be used for result tracking (Stage C). Each such model can provide feed-back information by identifying outlier (i.e., atypical) documents and possibly mislabeled documents.”; Note: Also see Figure 2 to see set S1, S2, S3… are metadata values to distinguish each set of documents (annotated samples) reviewed and R1, R2, R3… are another set of metadata values unique from S1, S2, S3 used to distinguish the reviewer.)
each of one or more unique metadata values of the N unique metadata values is associated with a respective annotated sample of a second set of annotated samples of the multiple annotated samples,
the one or more unique metadata values are different from the unique metadata value n, and
the first set of annotated samples is different from the second set of annotated samples; and (Privault [0082]: “subset S1, which was reviewed by reviewer R1, may be classified with the model M2 trained on the labels of S2, which were provided by reviewer R2. Similarly, subset S2, which was reviewed by R2 is classified with the model M3, which was trained on the labels of S3 by R3. Finally, subset S3, reviewed by R3, is classified with the model M1, trained on the labels of S1 provided by R1.”)
training, based on the obtained training set, N machine learning (ML) models for a classification task, wherein
for n=1, .... N, ML model n is associated with the unique metadata value n of the N unique metadata values, (Privault [0080]: “At S110, a classifier model M1, M2, M3, . . . Mi 62 is learned for each reviewer/subset reviewed. Steps S108 and S110 are described in further detail below with reference to FIG. 3. Each classifier model M1, M2, M3, . . . Mi built for a respective user from the set of documents labeled by this user can be used for result tracking (Stage C). Each such model can provide feed-back information by identifying outlier (i.e., atypical) documents and possibly mislabeled documents.”; Note: Also see Figure 2 to see set S1, S2, S3… are metadata values to distinguish each set of documents (annotated samples) reviewed and R1, R2, R3… are another set of metadata values unique from S1, S2, S3 used to distinguish the reviewer.)
Privault does not teach the ML model n is trained to test a hypothesis that the unique metadata value n is associated with a lower quality of annotated samples, of the multiple annotated samples, than each of the one or more unique metadata values of the N unique metadata values,
However, Duesterwald teaches the ML model n is trained to test a hypothesis that the unique metadata value n is associated with a lower quality of annotated samples, of the multiple annotated samples, than each of the one or more unique metadata values of the N unique metadata values, (Duesterwald [0053]: “In some embodiments of the present invention, probe mod 325 filters out potentially poisonous samples by generating targeted probes to determine a trust score for a given annotator. In these and some other embodiments, the goal is to determine whether a new sample can be safely added to the ground truth of a learner based on an assessment of the trustworthiness of an annotator and the assessed risk of the sample. Probe mod 325 aims to establish trustworthiness of annotators by probing individual annotators against the ground truth (i.e., vertical probing) and probing individual annotators against other annotators (i.e., horizontal probing). The result of the horizontal and vertical probing for each annotator, together with his/her history, is summarized in a trust score that represents the current trustworthiness of the annotator. New samples from an annotator may be accepted only if the risk of the samples is sufficiently balanced with respect to the trust score of the annotator.”; and [0047]: “Processing continues to operation S270A, where risk assessment mod 315 (see FIG. 3) determines a risk score for the label. In some embodiments of the present invention, the risk score is determined based on a series of versions of the machine learning model that have been trained over time and stored. In our exemplary embodiment (shown in FIG. 5A), a plurality of reference models 502 a, 504 a, 506 a, and 508 a have been trained at different times to recognize pictures and drawings of cats and dogs using labeled samples 512 and 514 from a ground truth.”;)
It would have been obvious before the effective filing date to combine the annotation metadata of Privault with the filtering of labeled samples by multiple annotators to optimize performance of the machine learning model (Duesterwald, [0065]). Privault and Duesterwald are analogous art because they both concern training machine learning models based on annotated samples.
Privault does not teach the training of the ML model n comprises performing an iterative process, and
each iteration of the iterative process comprises:
executing an epoch in which parameters of the ML model n are updated;
calculating, based on the updated parameters, a training loss for each annotated sample of the multiple annotated samples;
[removing, from the first set of annotated samples, a third set of annotated samples that is associated with the unique metadata value n] and has the training loss higher than a threshold; and
However, Li teaches the training of the ML model n comprises performing an iterative process, and (Li [page 2, 2 Related Work]: “An iterative training method is proposed to identify and downweight noisy samples [30]. A few other methods have also been proposed that use noise-tolerant loss functions to achieve robust learning under label noise [3, 4, 28].“;)
each iteration of the iterative process comprises:
executing an epoch in which parameters of the ML model n are updated; (Li [page 8, 5. Conclusion]: “We train for 3 epochs for each iteration. During the first 2000 mini-batches in the initial iteration “; page 5, 4.2 Implementation; and “we generate multiple mini-batches with synthetic noisy labels, and use them to update the parameters. In the meta-test step, we apply a consistency loss between each updated model and a teacher model, and train the original parameters to minimize the total consistency loss. In addition, we propose an iterative training scheme, where the model from previous iteration is used to clean data and refine predictions”;)
calculating, based on the updated parameters, a training loss for each annotated sample of the multiple annotated samples and (Li [page 3, Figure 2]: “Illustration of the proposed meta-learning based noise-tolerant (MLNT) training. For each mini-batch of training data, a meta loss is minimized before training on the conventional classification loss. We first generate multiple mini-batches of synthetic noisy labels with random neighbor label transfer (marked by orange arrow). The random neighbor label transfer can preserve the underlying noise transition (e.g. DEER → HORSE, CAT ↔ DOG), therefore generating synthetic label noise in a similar distribution as the original data. For each synthetic mini-batch, we update the parameters with gradient descent, and enforce the updated model to give consistent predictions with a teacher model. The meta-objective is to minimize the consistency loss across all updated models w.r.t θ.“;)
[removing, from the first set of annotated samples, a third set of annotated samples that is associated with the unique metadata value n] and has the training loss higher than a threshold; and (See Equation 10 of Li where T is a threshold)
It would have been obvious before the effective filing date to combine the annotation metadata of Privault with the filtering of labeled samples using multiple models of Li to improve predictions (Li, page 4, 3.3. Iterative Training). Privault and Li are analogous art because they both concern training machine learning models in regards to sample quality.
Privault does not appear to explicitly teach removing, from the first set of annotated samples, a third set of annotated samples that is associated with the unique metadata value n and [has the training loss higher than a threshold; and]
retaining the second set of annotated samples that is not associated with the unique metadata value n
However, Elisha teaches removing, from the first set of annotated samples, a third set of annotated samples that is associated with the unique metadata value n and [has the training loss higher than a threshold; and] (Elisha [0063]: “Training evaluator 246 may process training data 204 for each category to identify and evaluate suspect categories and suspect vectorized training data items (e.g., samples) with suspect labels.”; and [0136]: “In step 918, erroneous samples may be selectively removed in order from the removal list to create a revised training set. For example, as shown in FIG. 2, training evaluator 246 may cause data fetcher 208 to selectively remove training samples (e.g., based on user input provided through portal 212). For example, data fetcher 208 may change an assigned category for targeted training samples so that they are not fetched as training data 204 when classifier model(s) 214 b is trained for the category in question.”)
retaining the second set of annotated samples that is not associated with the unique metadata value n (Elisha [0025]: “Suspect categories may be retained or revised.”)
It would have been obvious before the effective filing date to combine the annotation metadata of Privault with the revising of training sets of Elisha for improved ML model accuracy (Elisha, [0025]). Privault and Elisha are analogous art because they both concern training machine learning models based on annotated samples.
Claim 2 is rejected over Privault, Duesterwald, Li and Elisha with the incorporation of claim 1.
Regarding claim 2, Privault teaches wherein a performance of the trained N ML models is with respect to a validation set comprising validated annotated samples. (Privault [0069]: “A second quality check is a cross-validation with other user models, which may be performed on the server computer by the assessment component 78, implemented by processor 72. For example, in the cross validation, model M1 of set S1 of documents (annotated samples) issued from reviewer R1's tagging, is applied to categorize set S2 of documents issued from user R2 tagging. Documents for which the tagging of S2 is inconsistent with the tagging predicted by M1 are filtered out. In the same way documents S1 are processed by user model M2 and/or M3. In addition, documents that are frequently inconsistently labeled among different reviewers are flagged as "difficult documents" and culled out for a double review. Documents that are flagged as "outliers" by each model can also be collected and culled out for a double review. These outliers are generally atypical documents with respect to the overall collection.”;)
Claim 3 is rejected over Privault, Duesterwald, Li and Elisha with the incorporation of claim 1.
Regarding claim 3, Privault teaches wherein the classification task is the same as a classification task for which the obtained training set is ultimately intended. (Privault [0051]: “The memory 42 stores a classifier model (Mi=M1, M2, M3 . . . etc.) 62, which may be progressively developed by machine learning techniques using the reviewer's labels (e.g., classifications) of the documents and features (e.g., words and/or word frequencies) extracted from the documents as inputs. A variety of machine learning techniques have been developed for application to document classification.”;)
Claim 6 is rejected over Privault, Duesterwald, Li and Elisha with the incorporation of claim 1.
Regarding claim 6, Privault does not teach training a new ML model for the classification task, based on a filtered training set comprising the first set of annotated samples without the removed third set of annotated samples, wherein the new ML model is different from the N ML models.
However, Li teaches training a new ML model for the classification task, based on a filtered training set comprising the first set of annotated samples without the removed third set of annotated samples, wherein the new ML model is different from the N ML models. (Privault [pages 4-5, 3.3]: “First, we perform an initial training iteration following the method described in Algorithm 1, and acquire a model with the best validation accuracy (usually the teacher). We name that model as mentor and use θ∗ to denote its parameters. In the second training iteration, we repeat the steps in Algorithm 1 with two changes described as follows. First, if the classification loss Lc(X,Y,θ) is applied to the entire training set D, samples with wrong ground-truth labels can corrupt training. Therefore, we remove a sample from the classification loss if the mentor model assigns a low probability to the ground-truth class. In effect, the classification loss would now sample batches from a filtered training set D′ which contains fewer corrupted samples.”)
It would have been obvious before the effective filing date to combine the annotation metadata of Privault with the filtering of labeled samples using multiple models of Li to improve predictions (Li, page 4, 3.3. Iterative Training). Privault and Li are analogous art because they both concern training machine learning models in regards to sample quality.
Claim 7 is rejected over Privault, Duesterwald, Li and Elisha with the incorporation of claim 1.
Regarding claim 7, Privault does not teach training a new ML model for the classification task based on the obtained training set, wherein
the new ML model is different from the N ML models, and
in the training of the new ML model, weights are assigned to the multiple annotated samples according to quality grades associated with the one or more annotations in the multiple annotated samples.
However, Li teaches training a new ML model for the classification task based on the obtained training set, wherein in the training of the new ML model, weights are assigned to the multiple annotated samples according to quality grades associated with the one or more annotations in the multiple annotated samples. (Li [pages 4-5, 3.3]: “First, we perform an initial training iteration following the method described in Algorithm 1, and acquire a model with the best validation accuracy (usually the teacher). We name that model as mentor and use θ∗ to denote its parameters. In the second training iteration, we repeat the steps in Algorithm 1 with two changes described as follows. First, if the classification loss Lc(X,Y,θ) is applied to the entire training set D, samples with wrong ground-truth labels can corrupt training. Therefore, we remove a sample from the classification loss if the mentor model assigns a low probability to the ground-truth class. In effect, the classification loss would now sample batches from a filtered training set D′ which contains fewer corrupted samples.”; Note: The weights of a model associated with the annotated samples are updated when retrained with.)
It would have been obvious before the effective filing date to combine the annotation metadata of Privault with the filtering of labeled samples using multiple models of Li to improve predictions (Li, page 4, 3.3. Iterative Training). Privault and Li are analogous art because they both concern training machine learning models in regards to sample quality.
Claim 8 is rejected over Privault, Duesterwald, Li and Elisha with the incorporation of claim 1.
Regarding claim 8, Privault teaches wherein the obtaining of the training set and the training of the N ML models are executed by at least one hardware processor of a computer in which the computer-implemented method is implemented. (Privault [0053]: “The exemplary server 16 includes memory 70 which stores software instructions for performing certain parts of the exemplary method. A processor 72, in communication with memories 20 and 70, executes the instructions. Included among these instructions is a clustering application 74, for assigning documents in the collection 18 to (obtaining) a respective one of a set of clusters (e.g., one cluster for each reviewer)”; [0024]: “For each set, the method includes displaying documents in the set on a display device for review by a reviewer, receiving the reviewer's labels for the displayed documents, based on the reviewer's labels, assigning a class from a plurality of classes to each of the reviewed documents, and progressively training a classifier model stored in computer memory based on features extracted from the reviewed documents in the set and their assigned classes.”; and [0097]: “the tagging input of each user in the group is used to complete the training of the classifier Mi based on each user's tagging of the documents in the subset.”)
Claim 9 is rejected over Privault, Duesterwald, Li and Elisha.
Regarding claim 9, Privault teaches a system, comprising:
at least one hardware processor; and (Privault [0053]: “The exemplary server 16 includes memory 70 which stores software instructions for performing certain parts of the exemplary method. A processor 72, in communication with memories 20 and 70, executes the instructions.”;)
a non-transitory computer-readable storage medium having program code embodied therewith, the program code executable by said at least one hardware processor to: (Privault [0098]: “The method illustrated in FIGS. 2 and 3 may be implemented in a computer program product that may be executed on a computer (or by multiple computers). The computer program product may be a computer-readable recording medium on which a control program is recorded, such as a disk, hard drive, or the like.”;)
The remainder of claim 9 is claim 1 in the form of a system and is rejected for the same reasons as claim 1 stated above.
Dependent claim 10 is claim 2 in the form of a system and is rejected for the same reasons as claim 2 stated above. For the rejection of the limitations specifically pertaining to the system of claim 9, see the rejection of claim 9 above.
Dependent claim 11 is claim 3 in the form of a system and is rejected for the same reasons as claim 3 stated above. For the rejection of the limitations specifically pertaining to the system of claim 9, see the rejection of claim 9 above.
Dependent claim 14 is claim 6 in the form of a system and is rejected for the same reasons as claim 6 stated above. For the rejection of the limitations specifically pertaining to the system of claim 9, see the rejection of claim 9 above.
Dependent claim 15 is claim 7 in the form of a system and is rejected for the same reasons as claim 7 stated above. For the rejection of the limitations specifically pertaining to the system of claim 9, see the rejection of claim 9 above.
Claim 16 is rejected over Privault, Duesterwald, Li and Elisha.
Regarding claim 16, Privault teaches a computer program product comprising a non-transitory computer-readable storage medium having program code embodied therewith, the program code executable by at least one hardware processor to: (Privault [0098]: “The method illustrated in FIGS. 2 and 3 may be implemented in a computer program product that may be executed on a computer (or by multiple computers). The computer program product may be a computer-readable recording medium on which a control program is recorded, such as a disk, hard drive, or the like.”;)
The remainder of claim 16 is claim 1 in the form of a computer program product comprising a non-transitory computer-readable storage medium and is rejected for the same reasons as claim 1 stated above.
Dependent claim 17 is claim 3 in the form of a computer program product comprising a non-transitory computer-readable storage medium and is rejected for the same reasons as claim 3 stated above. For the rejection of the limitations specifically pertaining to the computer program product comprising a non-transitory computer-readable storage medium of claim 16, see the rejection of claim 16 above.
Dependent claim 20 is claims 6 and 7 in the form of a computer program product comprising a non-transitory computer-readable storage medium and is rejected for the same reasons as claims 6 and 7 stated above. For the rejection of the limitations specifically pertaining to the computer program product comprising a non-transitory computer-readable storage medium of claim 16, see the rejection of claim 16 above.
Claim 22 is rejected over Privault, Duesterwald, Li and Elisha with the incorporation of claim 1.
Regarding claim 22, Privault does not teach for n = 1, ..., N, executing the trained ML model n to test whether the hypothesis for the trained ML model n has been proved based on performance of the trained ML model n with respect to the executing of the trained ML model n.
However, Li teaches for n = 1, ..., N, executing the trained ML model n to test whether the hypothesis for the trained ML model n has been proved based on performance of the trained ML model n with respect to the executing of the trained ML model n. (Li [page 7]: “In addition, we propose an iterative training scheme, where the model from previous iteration is used to clean data and refine predictions. We evaluate the proposed method on two datasets. The results validate the advantageous performance of our method compared to state-of-the-art methods.”; page 8, 5 Conclusion; and “τ is the threshold to determine which samples are filtered out by the mentor model during the 2nd and 3nd training iteration. It controls the balance between the quality and quantity of the data that is used by the classification loss.”; See Equation (10) on page 5 to see that samples above a threshold are kept therefore having lower loss and samples that are filtered out have higher loss.)
It would have been obvious before the effective filing date to combine the annotation metadata of Privault with the filtering of labeled samples using multiple models of Li to improve predictions (Li, page 4, 3.3. Iterative Training). Privault and Li are analogous art because they both concern training machine learning models in regards to sample quality.
Claims 5, 13 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Privault, Duesterwald, Li and Elisha in view of Davis et al. (From Context to Content: Leveraging Context to Infer Media Metadata); hereinafter Davis in view of Sanderson et al. (Making Web Annotations Persistent over Time); hereinafter Sanderson
Claim 5 is rejected over Privault, Duesterwald, Li, Elisha, Davis and Sanderson.
Regarding claim 5, Privault does not teach wherein the plurality of timestamps is selected from a group consisting of:
different times of a day at which the one or more annotations were made;
different days of a week at which the one or more annotations were made; and
However, Davis teaches wherein the plurality of timestamps is selected from a group consisting of:
different times of a day at which the one or more annotations were made; (See Figure 1 of Davis wherein the gathering of contextual metadata includes time and date.)
different days of a week at which the one or more annotations were made; and (See Figure 1 of Davis wherein the annotation metadata has processing results that include “Day of Week: Tuesday 100%”)
It would have been obvious before the effective filing date to combine the annotation metadata of Privault with the day of the week metadata of Davis for effective annotation (Davis, page 190, 3. System Description). Privault and Davis are analogous art because they both concern analyzing annotation metadata.
Privault does not teach different calendar days of a month at which the one or more annotations were made.
However, Sanderson teaches different calendar days of a month at which the one or more annotations were made. (Sanderson [page 7, 5.2.1 Mementos for Given Annotation]: “To step through an example evaluation, on the 23rd of January the day's still empty current events page in Wikipedia was annotated with a segment and the content "No news is good news". The client posted this annotation to Blogger, from where it was collected and indexed by the Cheshire3 system. This process is depicted below at the top of Figure 7)
It would have been obvious before the effective filing date to combine the annotation metadata of Privault with the day of the month metadata of Sanderson to improve accuracy (page 7, column 1). Privault and Sanderson are analogous art because they both concern analyzing annotation metadata.
Dependent claim 13 is claim 5 in the form of a system and is rejected for the same reasons as claim 5 stated above. For the rejection of the limitations specifically pertaining to the system of claim 9, see the rejection of claim 9 above.
Dependent claim 19 is claim 5 in the form of a computer program product comprising a non-transitory computer-readable storage medium and is rejected for the same reasons as claim 5 stated above. For the rejection of the limitations specifically pertaining to the computer program product comprising a non-transitory computer-readable storage medium of claim 16, see the rejection of claim 16 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 DAVID H TRAN whose telephone number is (703)756-1525. The examiner can normally be reached M-F 9:30 am - 5:30 pm.
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, Viker Lamardo can be reached at (571) 270-5871. 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.
/DAVID H TRAN/Examiner, Art Unit 2147
/VIKER A LAMARDO/Supervisory Patent Examiner, Art Unit 2147
Prosecution Insights