DETAILED ACTION
This office action is in response to the amendments filed 03/05/2026.
Claims 1, 8, and 15 have been amended. Claims 7, 14, and 20 have been canceled. Claims 1-6, 8-13, and 15-19 are currently 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 .
Response to Arguments
Prior Art Rejections:
Applicant's arguments regarding the prior art rejections (pg. 14-16) have been fully considered but they are not persuasive.
Applicant argues that the amended independent claim limitation which recites “...wherein extracting the feature importance also comprises performing permutation importance to quantify the feature importance for the features of the first dataset input to the machine learning model, wherein performance of the permutation importance records an increase in loss function of a prediction accuracy of the machine learning model as inputted features of the first dataset are shuffled one column at a time and where a shuffled column is returned to an original order before shuffling of a next column.” is not taught or suggested by any of the cited references. Applicant specifically argues that while La Cava discloses permuting each feature to calculate a permutation importance, La Cava does not disclose the column-wise shuffling recited in the claim. However, examiner respectfully notes that the feature permutation described by La Cava is equivalent to the claimed column-wise shuffling. Specifically, one of ordinary skill in the art will recognize that a column of a dataset represents a feature, and thus the shuffling of feature/predictor
x
j
described by La Cava (pg. 576, paragraph 2) is equivalent to the claimed shuffling of a column. Additionally, in La Cava, the permutation importance of each feature is computed based on a comparison between the normal data set and the permuted dataset in which only that feature is shuffled. No two features are shuffled at the same time, and thus a shuffled column is returned to its original order before shuffling the next column.
The prior art rejections have been updated to include the amended limitations and to clarify the reasoning given for the limitations that were not amended.
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 1-4, 8-11, and 15-17 are rejected under 35 U.S.C. 103 as being unpatentable over
Peipei et al., China Patent CN 106934035 B (hereinafter ‘Peipei’) in view of
Wenchel et al., Untied States Patent US 11922280 B2 (hereinafter ‘Wenchel’),
La Cava et al., “Interpretation of machine learning predictions for patient outcomes in electronic health records” (hereinafter ‘La Cava’),
Dalli et al., United States Patent Application Publication US 20210133630 A1 (hereinafter ‘Dalli’), and
Jia et al., U.S. Patent Application Publication US 20210097329 A1 (hereinafter ‘Jia’).
Regarding Claim 1: A computer-implemented method for versioning a machine learning model, the computer-implemented method comprising:
Peipei teaches ingesting, by the versioning service, the first dataset used to train the machine learning model; ([0012] Peipei teaches the first dataset: “Divide a set of multi-label data streams…into N data blocks…Dt represents the t-th data block in the multi-label data stream.” Data block Dt corresponds to the first dataset.)
Peipei teaches performing, by the versioning service, feature exploration of the first dataset and extracting from the first dataset, feature importance (f1) of the machine learning model; ([0025] Peipei teaches feature exploration of the first dataset: “Count the feature distribution of the t-th data block Dt.” [0025-0035] Peipei teaches the computations to obtain “the correlation between the all-dimensional feature vectors…and the class label…of the t-th data block Dt.” This correlation is a measure of feature importance.)
Peipei teaches ranking, by the versioning service [using the LIME explainable AI framework], top features of the first dataset used to train the machine learning model by the feature importance, up to a configured threshold number of features ([0037-0045] Peipei teaches the computations to “sort the correlation between the all-dimensional feature vectors…and the class label…in descending order” to obtain the “sorted feature vector set.” Sorting in descending order is equivalent to ranking, and as above, the correlation is a measure of feature importance. [0046] Peipei teaches the threshold number of features: “Select the first K feature vectors and their corresponding importance from the sorted feature vector set of the t-th data block Dt.” K corresponds to the threshold.)
Peipei teaches pre-processing, by the versioning service, to extract features (f2) of a second dataset; ([0012] Peipei teaches the second dataset: “Dt+1 represents the t+1-th data block in the multi-label data stream.” Data block Dt+1 corresponds to the second dataset. [0025] Peipei teaches preprocessing the second dataset: “Count the feature distribution of…the t+1-th data block Dt+1.”)
Peipei does not appear to explicitly disclose:
training, by a versioning service, the machine learning model using a first dataset;
wherein the extracting is performed using a local interpretable model-agnostic explanations (LIME) explainable artificial intelligence (AI) framework to interpret predictions of the machine learning model resulting from inputted features of the first dataset,
ranking, by the versioning service using the LIME explainable AI framework,
comparing, by the versioning service, changes in features between f1 and f2 for up to the configured threshold number of features; and
upon comparing, by the versioning service, the changes in the features between f1 and f2, and the changes between f1 and f2 are non-overlapping features:
highlighting set (f1−f2) in f1 which have an addition or deletion of categories within a feature; and
if the set (f1−f2) is ranked within the top features up to the configured threshold number of features for f1, re-training, by the versioning service, the machine learning model using a merged set of top features comprising features of importance from f1 and f2 up to the configured threshold number of features.
However, Wenchel teaches training, by a versioning service, the machine learning model using a first dataset; (Col. 6, lines 55-56: “The calculation or generation of the metrics 104 can be based on training data…” Col. 18, lines 20-22: “In some embodiments, the method also includes training, the at least one model via the processor.”)
Wenchel teaches wherein the extracting is performed using a local interpretable model-agnostic explanations (LIME) explainable artificial intelligence (AI) framework to interpret predictions of the machine learning model resulting from inputted features of the first dataset, and (Col. 15-16, lines 65-3: “Explanation service…calculates explainability values, for example using one or more techniques such as LIME and SHAP to determine feature values.” Col. 6, lines 16-17: “explanations can include indications of data features and their associated feature importance.” LIME is used to extract the importance of inputted features by calculating explainability values (i.e. interpreting predictions).)
Wenchel teaches ranking, by the versioning service using the LIME explainable AI framework (Col. 9, lines 11-14: “Data drift+Explainability (e.g., feature importance)—facilitates sorting, in order of decreasing importance, by combination of global/local feature importance & data drift metric(s).” As shown above, Wenchel’s feature importance metrics are calculated by the LIME explainable AI framework.)
Wenchel teaches comparing, by the versioning service, changes in features between f1 and f2 for up to the configured threshold number of features; and upon comparing, by the versioning service, the changes in the features between f1 and f2, and the changes between f1 and f2 are non-overlapping features: (Wenchel teaches a machine learning monitoring system which detects changes in statistical metrics between 2 windows of data. Col. 5 lines 44-46: the system is capable of “tracking the change iteratively, in various moments, parameters, or other measures over time.” Tracking changes between two windows of data corresponds to comparing changes in features between the two datasets.)
Wenchel teaches highlighting set (f1−f2) in f1 which have an addition or deletion of categories within a feature; and if the set (f1−f2) is ranked within the top features up to the configured threshold number of features for f1, (Col. 5 lines 46-51 Wenchel teaches that “measures over time” as described above includes “measures of spread, such as range.” Identifying features with a change in range or spread is equivalent to identifying features with an addition or deletion of categories.)
Wenchel teaches re-training, by the versioning service, the machine learning model (Wenchel’s machine learning monitoring system generates alerts in response to changes in metrics in the data. Col. 6 line 63 – col. 7 line 3: Wenchel teaches “one or more of the alerts…can trigger an automatic action that includes a remediation. Examples of remediation actions can include… retraining of the ML model.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Peipei and Wenchel. Peipei teaches detecting data drift between windows of data based on changes in the distributions of important features. Wenchel teaches monitoring the health of a machine learning model and its data, and generating alerts including retraining recommendations in response to calculated metrics. One of ordinary skill would have motivation to combine Peipei and Wenchel because “[t]he sample data used to train a ML model may change over time, or be sampled from a true distribution in a statistically biased way that might impact the performance (e.g., overall accuracy, inference speed, accuracy with regard to a specific subgroup in the data) of the ML model” (Wenchel, col. 2 lines 41-46). To mitigate impacts to performance, retraining is recommended in response to detection of these changes in metrics as is known in the art: “Standard heuristics for adjusting ML-based systems to changes in data distributions or use cases over time include … retraining the ML models” (Wenchel, col. 2 lines 46-49).
Peipei and Wenchel do not appear to explicitly disclose wherein extracting the feature importance also comprises performing permutation importance to quantify the feature importance for the features of the first dataset input to the machine learning model, wherein performance of the permutation importance records an increase in loss function of a prediction accuracy of the machine learning model as inputted features of the first dataset are shuffled one column at a time and where a shuffled column is returned to an original order before shuffling of a next column;
However, La Cava teaches wherein extracting the feature importance also comprises performing permutation importance to quantify the feature importance for the features of the first dataset input to the machine learning model, wherein performance of the permutation importance records an increase in loss function of a prediction accuracy of the machine learning model as inputted features of the first dataset are shuffled one column at a time and where a shuffled column is returned to an original order before shuffling of a next column; (Pg. 576, para. 2: “We use a model-agnostic permutation importance score first proposed by Breiman et al.16 to estimate the importance of the features in the trained models. Permutation importance is defined as the mean decrease in accuracy of the trained model when each feature is permuted. We calculate the permutation importance of predictor
x
j
∈
x
by the following steps: 1. Create a permuted test set
{
y
i
,
x
'
i
}
i
=
1
N
t
in which
x
j
∈
x
is randomly shuffled.
N
t
is the number of test samples. 2. Generate predictions on the normal test set,
y
^
(
x
)
, and permuted predictions,
y
^
(
x
'
)
. 3. The permutation importance (PI) is the mean decrease in accuracy due to the perturbed feature…” Permutation importance is used to extract feature importance by recording the decrease in accuracy (i.e. increase in loss function of prediction accuracy) when features (i.e. columns) are permuted (i.e. shuffled). The permutation importance of each feature is computed based on a comparison between the normal data set and the permuted dataset in which only that feature is shuffled (i.e. a shuffled column is returned to its original order before shuffling the next column).)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Peipei, Wenchel, and La Cava. Peipei teaches detecting data drift between windows of data based on changes in the distributions of important features. Wenchel teaches monitoring the health of a machine learning model and its data, and generating alerts including retraining recommendations in response to calculated metrics including feature importance. La Cava teaches using permutation importance to measure feature importance. One of ordinary skill would have motivation to combine Peipei, Wenchel, and La Cava because “permutation importance, a model-agnostic method, produces the most clinically relevant interpretations, as long as the underlying model produces good predictions. Among models with high predictive performance on test sets, permutation importance scores were highly correlated, and interpretable” (La Cava, pg. 573, para. 1).
Peipei, Wenchel, and La Cava do not appear to explicitly disclose wherein the configured threshold number of features is a percentage (n%) of total features extracted;
However, Dalli teaches wherein the configured threshold number of features is a percentage (n%) of total features extracted; ([0057]: “The summarization technique may provide a method for simplifying explanations… A simplified explanation may also include a threshold such that only the top n features [according to feature importance] are considered, where n is either a static number or percentage value.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Peipei, Wenchel, La Cava, and Dalli. Peipei teaches detecting data drift between windows of data based on changes in the distributions of important features. Wenchel teaches monitoring the health of a machine learning model and its data, and generating alerts including retraining recommendations in response to calculated metrics including feature importance. La Cava teaches using permutation importance to measure feature importance. Dalli teaches a method for explainable AI, including simplifying explanations by only considering a top percentage of important features. One of ordinary skill would have motivation to combine Peipei, Wenchel, La Cava, and Dalli because Dalli’s percentage threshold simplifies feature importance explanations by reducing the number of features “so as to provide the user with a moderate, as opposed to excessive, amount of information to process” (Dalli, Para. 0070) while remaining dynamic and conditional on the number of features.
Peipei, Wenchel, La Cava, and Dalli do not appear to explicitly disclose re-training, by the versioning service, the machine learning model using a merged set of top features comprising features of importance from f1 and f2 up to the configured threshold number of features.
However, Jia teaches re-training, by the versioning service, the machine learning model using a merged set of top features comprising features of importance from f1 and f2 up to the configured threshold number of features. (Jia teaches retraining a machine learning model using a merged set of important features created by removing unimportant features and adding new features. [0032]: “If an importance metric of a machine learning feature is below a threshold value, the machine learning feature is to be removed from the model (e.g., model retrained to remove feature).” [0035]: “If management of the features has caused a machine learning feature to be added, transformed, or modified, the new version of the machine learning model is generated/retrained with training data that includes the new/transformed/modified feature.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Peipei, Wenchel, La Cava, Dalli, and Jia. Peipei teaches detecting data drift between windows of data based on changes in the distributions of important features measured by cosine similarity. Wenchel teaches monitoring the health of a machine learning model and its data, and generating alerts including retraining recommendations in response to calculated metrics including feature importance. La Cava teaches using permutation importance to measure feature importance. Dalli teaches a method for explainable AI, including simplifying explanations by only considering a top percentage of important features. Jia teaches managing (e.g. adding, removing) features of machine learning model training data based on feature importance metrics. One of ordinary skill would have motivation to combine Peipei, Wenchel, La Cava, Dalli, and Jia because removing unimportant features “allows a more computationally efficient model to be generated (e.g., less processing and storage required to deploy the model) as well as reduces the amount of data required to be collected and stored” (Jia, para. 0025). By adding new important features, “the new version of the machine learning model may perform more efficiently and/or be more accurate” (Jia, para. 0035).
Regarding Claim 2, Peipei, Wenchel, La Cava, Dalli, and Jia teach The computer-implemented method of claim 1, as shown above.
Wenchel also teaches further comprising: upon comparing, by the versioning service, the changes in the features between f1 and f2, the changes between f1 and f2 are non-overlapping features and the set (f1−f2) is not ranked within the top features up to the configured threshold number of features for f1, storing, by the versioning service, the top features up to the configured threshold number of features in a feature store. (Col. 6 lines 14-17 Wenchel teaches storing features: “explanations can be stored…for example in common records of a database or other storage medium. Such explanations can include indications of data features and their associated feature importance.” The database or other storage medium corresponds to the feature store.)
Regarding Claim 3, Peipei, Wenchel, La Cava, Dalli, and Jia teach The computer-implemented method of claim 1, as shown above.
Wenchel also teaches further comprising: upon comparing, by the versioning service, the changes in features between f1 and f2, and finding no change in the features between f1 and f2 within the top features up to the configured threshold number of features, outputting by the versioning service, a recommendation that re-training of the machine learning model is not required. (Col. 6 line 63 – col. 7 line 3: “Optionally, one or more of the alerts 105 can specify... one or more remediation actions that are to be taken… Examples of remediation actions can include… retraining of the ML model.” Since alerts are always generated in response to the calculated metrics, but specifying a remediation action in the alert is optional, it follows that in the case when the comparison of metrics indicates that retraining is not required, the alert should not include a recommendation to retrain the model.)
Regarding Claim 4, Peipei, Wenchel, La Cava, Dalli, and Jia teach The computer-implemented method of claim 1, as shown above.
Peipei also teaches wherein upon comparing, by the versioning service, changes in features between f1 and f2, the second dataset includes a new feature set with attributes not found in f1, the computer-implemented method further comprises creating, by the versioning service, a correlation matrix between f1 and the new feature set of f2 having the attributes that are not found in f1. ([0046] Peipei teaches to “use the cosine similarity function to calculate the similarity… between the feature distribution set of the t-th data block Dt and the feature distribution set of the t+1-th data block Dt+1.” This is equivalent to calculating their correlation—according to specification paragraph [0020] of the instant application, “the versioning service may evaluate the correlations between the new features [and the current features]…using cosine similarity.”)
Claims 8-11 are product claims, containing substantially the same elements as method claims 1-4, respectively. Peipei, Wenchel, La Cava, Dalli, and Jia teach the elements of claims 1-4, as shown above.
Wenchel also teaches A computer program product for versioning a machine learning model comprising: one or more computer readable storage media having computer-readable program instructions stored on the one or more computer readable storage media, said program instructions executes a computer-implemented method comprising: (Col. 20, lines 31-36: “Some embodiments described herein relate to devices with a non-transitory computer-readable medium (also can be referred to as a non-transitory processor-readable medium or memory) having instructions or computer code thereon for performing various computer-implemented operations.”)
Claims 15-17 are system claims, containing substantially the same elements as method claims 1-2 and 4, respectively. Peipei, Wenchel, La Cava, Dalli, and Jia teach the elements of claims 1-2 and 4, as shown above.
Wenchel also teaches A computer system for versioning a machine learning model comprising: a processor; and a computer-readable storage media coupled to the processor, wherein the computer- readable storage media contains program instructions executing a computer-implemented method comprising: (Col. 20, lines 29-36: “Each of the devices described herein can include one or more processors as described above. Some embodiments described herein relate to devices with a non-transitory computer-readable medium (also can be referred to as a non-transitory processor-readable medium or memory) having instructions or computer code thereon for performing various computer-implemented operations.”)
Claims 5-6, 12-13, and 18-19 are rejected under 35 U.S.C. 103 as being unpatentable over Peipei in view of Wenchel, La Cava, Dalli, and Jia, and further in view of
Huang et al., United States Patent US 10824661 B1 (hereinafter ‘Huang’).
Regarding Claim 5, Peipei, Wenchel, La Cava, Dalli, and Jia teach The computer-implemented method of claim 4, as shown above.
Peipei also teaches further comprising: computing, by the versioning service, cosine similarity [and vector distance] between the features of f1 and the new feature set of f2 having the attributes that are not found in f1; ([0046] Peipei teaches to “use the cosine similarity function to calculate the similarity … between the feature distribution set of the t-th data block Dt and the feature distribution set of the t+1-th data block Dt+1” as in claim 4.)
Peipei teaches determining, by the versioning service, an amount of overlap [in the vector distance] between the top features of f1 up to the configured threshold number of features and the new feature set of f2; and upon the amount of overlap [in vector distance] is insignificant or null, ([0047] Peipei teaches comparing the “feature distribution difference” of the “first K feature vectors” to a threshold to determine if “concept drift occurs.” The first K features corresponds to the configured threshold number of features (n), and the threshold described by Peipei corresponds to determining if the overlap in vector distance is insignificant or null.)
Wenchel teaches outputting, by the versioning service, a recommendation to re-train the machine learning model (Wenchel's machine learning monitoring system generates alerts in response to changes in metrics in the data. Col. 6 line 63 - col. 7 line 3: "one or more of the alerts 105 can specify… one or more remediation actions that are to be taken… Examples of remediation actions can include… retraining of the ML model.")
Peipei, Wenchel, La Cava, Dalli, and Jia do not appear to explicitly disclose computing vector distance
However, Huang teaches computing vector distance (Huang teaches a method for establishing a mapping between a first and second set of textual topics. In the method, for a pair of topics, “a similarity value is determined… based on a distance between the terms in the first topic and the terms in the second topic” (col. 13, lines 23-26). Huang elaborates, “each of the topics is represented by a vector” and “The similarity of the vectors may be determined based on the cosine distance or the Euclidian distance” (col. 13, lines 31-43). Euclidian distance is equivalent to vector distance.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Peipei, Wenchel, La Cava, Dalli, Jia, and Huang. Peipei teaches detecting data drift between windows of data based on changes in the distributions of important features measured by cosine similarity. Wenchel teaches monitoring the health of a machine learning model and its data, and generating alerts including retraining recommendations in response to calculated metrics including feature importance. La Cava teaches using permutation importance to measure feature importance. Dalli teaches a method for explainable AI, including simplifying explanations by only considering a top percentage of important features. Jia teaches managing (e.g. adding, removing) features of machine learning model training data based on feature importance metrics. Huang teaches methods for determining distance between vectors, including cosine distance, Euclidian distance, and semantic similarity. One of ordinary skill would have motivation to combine Peipei, Wenchel, La Cava, Dalli, Jia, and Huang because vector (Euclidian) distance and cosine distance are common alternatives for measuring the distance between vectors (Huang, col. 13, lines 42-44).
Regarding Claim 6, Peipei, Wenchel, La Cava, Dalli, and Jia teach The computer-implemented method of claim 4, as shown above.
Wenchel teaches outputting, by the versioning service, a recommendation to re-train the machine learning model (Wenchel's machine learning monitoring system generates alerts in response to changes in metrics in the data. Col. 6 line 63 - col. 7 line 3: "one or more of the alerts 105 can specify… one or more remediation actions that are to be taken… Examples of remediation actions can include… retraining of the ML model.").
Peipei, Wenchel, La Cava, Dalli, and Jia do not appear to explicitly disclose further comprising: computing, by the versioning service, semantic distance between the features of f1 and the new feature set of f2, wherein overlap in the semantic distance between features of importance within f1 and the new feature set of f2 indicates a time-revised concept in f2 over an original feature in f1; and upon identifying the time-revised concept in f2 over the original feature in f1
However, Huang teaches computing semantic distance (As discussed in claim 5, Huang’s method for mapping between topics teaches determining a similarity value between two sets of terms, which can be calculated by cosine distance or Euclidian distance. Huang elaborates, “Other methods that provide a similarity of the first and the second topics may be used…deep learning methods may be used to obtain semantic similarity” (col. 13, line 63 – col 14 line 2).)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Peipei, Wenchel, La Cava, Dalli, Jia, and Huang. Peipei teaches detecting data drift between windows of data based on changes in the distributions of important features measured by cosine similarity. Wenchel teaches monitoring the health of a machine learning model and its data, and generating alerts including retraining recommendations in response to calculated metrics including feature importance. La Cava teaches using permutation importance to measure feature importance. Dalli teaches a method for explainable AI, including simplifying explanations by only considering a top percentage of important features. Jia teaches managing (e.g. adding, removing) features of machine learning model training data based on feature importance metrics. Huang teaches methods for determining distance between vectors, including cosine distance, Euclidian distance, and semantic similarity. One of ordinary skill would have motivation to combine Peipei, Wenchel, La Cava, Dalli, Jia, and Huang because semantic similarity is a known viable alternative to cosine distance and Euclidian distance when measuring similarity between vectors (Huang, col. 14, lines 1-2).
Claims 12-13 are product claims, containing substantially the same elements as method claims 5-6, respectively. Peipei, Wenchel, La Cava, Dalli, Jia, and Huang teach the elements of claims 5-6, as shown above.
Claims 18-19 are system claims, containing substantially the same elements as method claims 5-6, respectively. Peipei, Wenchel, La Cava, Dalli, Jia, and Huang teach the elements of claims 5-6, as shown above.
Conclusion
Claims 1-6, 8-13, and 15-19 are rejected.
THIS ACTION IS MADE FINAL. 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 BENJAMIN M ROHD whose telephone number is (571)272-6445. The examiner can normally be reached Mon-Thurs 8:00-6:00 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, Viker Lamardo can be reached on (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.
/B.M.R./Examiner, Art Unit 2147
/VIKER A LAMARDO/Supervisory Patent Examiner, Art Unit 2147
Prosecution Insights