DETAILED ACTION
This Office Action is in response to the communications filed on March 11, 2026 for Application No. 17/554,048, in which claims 1-3, 5-7, and 9-11 are presented for examination. The amendments filed on March 11, 2026 have been entered, where claims 1, 3, 5, 7, 9, and 11 are amended.
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 Objections
Claims 1-3, 5-7, and 9-11 are objected to because of the following informalities:
“a decoy against the training data estimation attack” (Claim 1, ln. 13-14; Claim 5, ln. 12-13; Claim 9, ln. 14-15), as currently formulated, lacks sufficient antecedent basis for “the training data estimation attack” (objection applies equally to dependent claims 2-3, 6-7, and 10-11).
“the second machine learning model exhibiting the enhanced robustness” (Claim 1, ln. 19; Claim 5, ln. 18; Claim 9, ln. 20), as currently formulated, lacks sufficient antecedent basis for “the enhanced robustness” (objection applies equally to dependent claims 2-3, 6-7, and 10-11).
Appropriate correction is required.
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.
Claims 1-3, 5-7, and 9-11 are rejected under 35 U.S.C. 103 as being unpatentable over Dupre et al. (hereinafter Dupre) (“Improving Dataset Volumes and Model Accuracy with Semi-Supervised Iterative Self-Learning”) in view of Sharma et al. (hereinafter Sharma) (Pat. No. US 11,755,743 B2).
PNG
media_image1.png
396
452
media_image1.png
Greyscale
Dupre, Figure 1
Regarding Claim 1, Dupre teaches a non-transitory computer-readable storage medium storing an information processing program that causes at least one computer (Pg. 4347, Sec. “Appendix III Used Hardware”, “All our experiments were conducted on a Dell Precision Tower 7910 XCTO server with two GPUs TITAN Xp, Intel Xeon E5-2623 v3, 128GB DDR4 RAM and 512GB SSD”)
to execute a process, the process comprising (Pg. 4339, Sec. “III. Methodology”, “Algorithm 1 Iterative Learning”; Pg. 4342, Sec. “IV. Results”, Subsec. “A. Experiment Environment”, “an iterative loop is set up as described in Algorithm 1 and is run for the three models across the three datasets):
generating, from an initial input comprised of . . . [confidently labeled samples and low confidence samples], a first machine learning model which has been trained with first training data (Pg. 4339, Col. 1, Fig. 1, “An overview of the Iterative Learning (IL) cycle, in which a model is trained on labelled data and used to classify new unlabelled samples. The class decisions for the new samples are then evaluated based on the confidence and then added to the labelled training dataset and the process repeated”, where comprising data of an initial input, “classif[ied]” “unlabelled samples”, are “added to the labelled training dataset” in order to generate a machine learning model that is trained with the training data, “a model is trained on labelled data”, and where, in the “Iterative Learning (IL) cycle”, the “model” and the “labelled training dataset” can be considered the first model and the first training data during the first iteration; see also Pg. 4340, Col. 1, Para. 1-3, “the unlabelled samples are classified and the process of updating the training set is run . . . The training set is then updated to include these confidently labelled samples and the process repeats with the model being retrained from scratch on the newly updated training set” and Pg. 4341, Col. 1, Para. 2, “Using a defined threshold Tc, samples can now be approved for inclusion in the labelled dataset Dl updated for use in the next training iteration”, where the initial input, “unlabelled samples” after “classifi[cation]”, includes both “confidently labelled samples” and low confidence samples, which are below “defined threshold Tc”);
setting an objective function based on an output of the first machine learning model and a target label (Pg. 4342, Col. 2, Para. 2, “For all experiments the same cross entropy loss function was used. Stochastic gradient decent is also utilised for all experiments with a starting learning rate of 0.01, with a scheduled step to 0.001 after 100 epochs. Weights for the metrics ca, cb, and cc, were set to 1, 0.5 and 0.25 respectively”, where the “cross entropy loss function” is an objective function, which is set through training using “Stochastic gradient decent”, which in turn is based on model outputs, and the training uses target “labels”, see Pg. 4337, Col. 2, Para. 3, “the model being trained is only ever exposed to what it considers fully labelled data”);
calculating a gradient of the objective function with respect to . . . [the training data] (Pg. 4342, Col. 2, Para. 2, “Stochastic gradient decent is also utilised for all experiments with a starting learning rate of 0.01, with a scheduled step to 0.001 after 100 epochs. Weights for the metrics ca, cb, and cc, were set to 1, 0.5 and 0.25 respectively”, where “all experiments” use “Stochastic gradient decent”, which involves derivative values, “gradient[s]”, of the objective function, “cross entropy loss function” discussed above, “for all experiments”, which a person of ordinary skill in the art would understand that the “gradient[s]” are generated with respect to the inputted training data, see Pg. 4340, Col. 1, Para. 1, “The training set is then updated to include these confidently labelled samples and the process repeats with the model being retrained from scratch on the newly updated training set”);
generating artificial . . . data by iteratively updating the . . . [low confidence samples] based on the calculated gradient (Pg.4340, Para. 1, “data candidates are selected to be incorporated into the training set . . . The training set is then updated to include these confidently labelled samples and the process repeats with the model being retrained from scratch on the newly updated training set”, where artificial data is generated by iteratively, “the process repeats”, updating the initial input to incorporate samples in an updating training set, “data candidates are selected to be incorporated into the training set”, with the iteratively updated low confidence samples being discarded, “The training set is then updated to include these confidently labelled samples”, see also Pg. 4342, Col. 2, Para. 2, “Stochastic gradient decent is also utilised for all experiments with a starting learning rate of 0.01, with a scheduled step to 0.001 after 100 epochs. Weights for the metrics ca, cb, and cc, were set to 1, 0.5 and 0.25 respectively”, where the gradient is used to iteratively update the low confidence samples because the retrained modal is used to iteratively calculate the updated initial input, which includes the low confidence samples, and “stochastic gradient decent” is used for “all experiments”, including the model “retain[ing]”; see also Pg. 4339, Col. 1, Fig. 1, “a model is . . . used to classify new unlabelled samples. The class decisions for the new samples are then evaluated based on the confidence and then added to the labelled training dataset and the process repeated”, where the updated “training dataset” is artificial because all of the initial input data are generated for “unlabelled samples” by the “model”),
the updating modifying the . . . [low confidence samples] to create an artificial data point . . . [that functions as training data] by increasing a classification certainty for the target label by the first machine learning model (Pg. 4341, Col. 2, Para. 4, “By leveraging these incremental updates the model can be utilised to identify only those samples that it is most confident belong to a respective class. As a result the model develops its knowledge of specific classes and is therefore better able to identify additional samples in latter iterations . . . a subset of new, unlabelled, samples get projected closer to the existing manifolds due to already learned characteristics of respective classes. These samples are then labelled and added to the training dataset and the model is re-trained. The manifolds are now updated, reflecting the information brought in by the added training samples”, where the iterative updating of low confidence samples, “leveraging these incremental updates the model can be utilised to identify . . . samples”, is used to create an artificial data point that functions as training data, “These samples are then labelled and added to the training dataset”, by increasing a classification certainty for a target label, “in latter iterations . . . a subset of new, unlabelled, samples get projected closer to the existing manifolds”, by the first machine learning model, “the model is re-trained”, which, as discussed above, determines certainty using the target “labels”, see Pg. 4337, Col. 2, Para. 3, “the model being trained is only ever exposed to what it considers fully labelled data”);
acquiring second training data by combining the first training data and the artificial . . . data (Pg.4340, Para. 1, “The training set is then updated to include these confidently labelled samples”; Pg. 4339, Fig. 1, “Update [->] Labelled Dataset”, where, during a subsequent iteration of the “Iterative Learning (IL) cycle”, the first training data, the “Labelled Dataset” used to “Train” the previous iteration of the “Model”, is combined with the artificial data, “confidently labelled samples”, to acquire second training data, “training set is then updated to include these confidently labelled samples”; and where the “confidently labelled samples” are artificial because they are generated for “unlabelled samples” by the “model”, see Pg. 4339, Col. 1, Fig. 1, “a model is . . . used to classify new unlabelled samples. The class decisions for the new samples are then evaluated based on the confidence and then added to the labelled training dataset and the process repeated”); and
training a second machine learning model by using the second training data, the second machine learning model . . . (Pg. 4340, Para. 1, “the process repeats with the model being retrained from scratch on the newly updated training set”; Pg. 4339, Fig. 1, “Train”, where this training occurs during the next iteration; see also Pg. 4340, Para. 1, “The training set is then updated to include these confidently labelled samples and the process repeats with the model being retrained from scratch on the newly updated training set”, wherein “retrained from scratch” falls within the broadest reasonable interpretation of training a second model; see also Pg. 4339, Fig. 2, “Train Model”, wherein the retraining occurs during the second iteration of the “Iterative Learning cycle”).
Dupre does not explicitly disclose . . . comprised of noise as meaningless data (where the low confidence samples in the initial input data are not specifically described as comprising noise as meaningless data; but see Pg. 4337, Col. 1-2, Para. 2-2, “the biggest target of semi-supervised learning techniques, developing ways in which unlabelled or noisy data can be utilised without the need for expensive and time consuming processes which can ‘clean’ the data . . . many learning frameworks are resistant to the presence of noise . . . a model and its loss function can be manipulated to become more robust to these issues . . . even in data where over 30 percent of binary labels were inverted, good accuracy was still obtained within their tasks”, where “semi-supervised learning” frameworks are discussed as utilizing training data comprised of noise) (subsequent recitations of noise omitted)
. . . decoy . . . that functions as a decoy against the training data estimation attack . . . exhibiting the enhanced robustness against the training data estimation attack . . . (where the artificial data that is combined with the training data does not include the low confidence samples and the updated training data is not specifically described as decoy data used to enhance robustness against training data estimation attacks) (subsequent recitations of decoy omitted).
However, Sharma teaches . . . [an initial input] comprised of noise as meaningless data . . . [which is added to a training dataset] (Pg. 11, Col. 6, Ln. 44-47, “There may be different ways to add noise to a machine learning model. Noise can be added to a model at the input level (such as by adding randomized inputs to the training data”, where “randomized inputs” are meaningless data, which is added as “[n]oise” to “the training data”; see generally Abstract, “This disclosure describes methods and systems for protecting machine learning models against privacy attacks. A machine learning model may be trained using a set of training data . . . An amount of noise may be added to the machine learning model to make a privacy guarantee value of the machine learning model”)
[to generate artificial] decoy [data] . . . that functions as a decoy against the training data estimation attack (Pg. 11, Col. 5-6, Ln. 18-47, “A model may leak membership information in response to a membership inference attack as a result of overfitting. Overfitting may occur where a machine learning algorithm builds a model that fits too closely to a limited set of data . . . In situations where overfitting has occurred, an adversary may use a model's confidence for a prediction provided in response to a sample input to infer whether the sample is or is not a member of the training data . . . Adding noise to causal models may further strengthen them against privacy attacks . . . There may be different ways to add noise to a machine learning model. Noise can be added to a model at the input level (such as by adding randomized inputs to the training data)”, where the “noise” is added to the “training data” to generate data that protects against attacks “to infer whether the sample is or is not a member of the training data”, which is within the broadest reasonable interpretation of a training data estimation attack because it estimates, “infer[s]” “membership of the training data”; and where the updated “training data” is within the broadest reasonable interpretation of artificial decoy data because it comprises artificial values, “randomized inputs”, which distract from the signal of sensitive data that would otherwise “result of overfitting”)
. . . [to produce a machine learning model] exhibiting the enhanced robustness against the training data estimation attack . . . (Pg. 11, Col. 5-6, Ln. 4-58, “A privacy leak (or breach) may occur when an adversary can use a model's output to infer the values of sensitive attributes of the training data . . . One example of a privacy attack is a membership inference attack . . . [to] infer whether a particular data sample was present in the training data . . . A model may leak membership information in response to a membership inference attack as a result of overfitting. Overfitting may occur where a machine learning algorithm builds a model that fits too closely to a limited set of data . . . In situations where overfitting has occurred, an adversary may use a model's confidence for a prediction provided in response to a sample input to infer whether the sample is or is not a member of the training data . . . Adding noise to causal models may further strengthen them against privacy attacks . . . The amount of noise added and the level or step at which noise is added may impact . . . how much less likely it is that the model will be susceptible to a privacy attack . . . The robustness of a machine learning model against privacy attacks may be referred to as a privacy guarantee of the machine learning model”).
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the generation of a trained model from an initial input comprising low confidence samples, including: calculation of an objective function gradient with respect to training data; generation of artificial data by iteratively updating the low confidence samples based on the calculated gradient, wherein low confidence samples are modified for use as artificial data points in training data, such that the model has increased classification certainty for target labels; acquiring second training data by combining the training data with the artificial data; and training a second model with the training data of Dupre with the initial input comprised of noise as meaningless data, which is added to a training dataset to generate artificial decoy data used as training data to produce a machine learning model with enhanced robustness against training data estimation attacks of Sharma in order to utilize Dupre’s confidence based addition of samples to training data (Dupre, Pg 4341, Col. 1, Para. 2, “Using a defined threshold Tc, samples can now be approved for inclusion in the labelled dataset Dl updated for use in the next training iteration”) in order to add both clean and noisy data to the training data (Sharma, Pg. 11, Col. 6, Ln. 44-47, “There may be different ways to add noise to a machine learning model. Noise can be added to a model at the input level (such as by adding randomized inputs to the training data”) for use in training of a machine learning model (Dupre, Pg. 4339, Col. 1, Fig. 1, “An overview of the Iterative Learning (IL) cycle, in which a model is trained on labelled data and used to classify new unlabelled samples. The class decisions for the new samples are then evaluated based on the confidence and then added to the labelled training dataset and the process repeated), which will allow for iterative and task-specific balancing of model accuracy and training data privacy protection (Sharma, Pg. 11, Col. 6, Ln. 40 - 53, “Adding noise to a machine learning model may reduce the accuracy of the machine learning model but reduce the likelihood that an adversary can use a privacy attack to learn information about the training data . . . The amount of noise added and the level or step at which noise is added may impact how much the accuracy of the model is reduced and how much less likely it is that the model will be susceptible to a privacy attack”; Dupre, Pg. 4340-4341, Col. 2-1, Para. 4-1, “The weighting itself is found experimentally by evaluating the accuracy of each metric . . . Importantly these values may change based on application as certain metrics may be more informative in different problems”; see also Sharma, Pg. 14, Col. 11, Ln. 29-36, “a membership attack may be possible whenever the distribution of output scores for training data is different from the test data . . . if an adversary knows a true label for a target input, then the adversary may guess the input to be a member of the training data set whenever the loss is lower”) in order to deploy accurate models that contain sufficient data privacy protections for particular use cases (Sharma, Pg. 10, Col. 4, Ln. 38-48, “In certain situations, the training data may contain private or personally identifiable information . . . For example . . . hospitals and doctors, as well as the individuals themselves, would not want others to know whether a particular individual is included in a set of training data used to build a model for diagnosing HIV”).
Regarding Claim 2, Dupre in view of Sharma teaches the non-transitory computer-readable storage medium of claim 1, wherein the training is retraining the first machine learning model by using the second training data (Dupre, Pg. 4340, Para. 1, “The training set is then updated to include these confidently labelled samples and the process repeats with the model being retrained from scratch on the newly updated training set”, wherein “retrained from scratch” falls within the broadest reasonable interpretation of retraining; Dupre, Pg. 4339, Fig. 2, “Train Model”, wherein the retraining occurs throughout the “Iterative Learning cycle”, which includes use of a second “training set” during the second iteration of the “Learning cycle”).
Regarding Claim 3, Dupre in view of Sharma teaches the non-transitory computer-readable storage medium of claim 1, wherein the generating the artificial decoy data includes using an optimization technique (Dupre, Pg. 4342, Sec. IV. Results, Subsec. “A. Experiment Environment”, Para. 3, “Stochastic gradient decent is also utilized for all experiments with a starting learning rate of 0.01, with a scheduled
step to 0.001 after 100 epochs”, where “stochastic gradient decent” is within the broadest reasonable interpretation of an optimization technique; and is used to train the model for “all experiments”, which includes the generating of initial inputs and confidence-based selection of the second training data, see Dupre, Pg. 4339, Col. 1, Fig. 1, “An overview of the Iterative Learning (IL) cycle, in which a model is trained on labelled data and used to classify new unlabelled samples. The class decisions for the new samples are then evaluated based on the confidence and then added to the labelled training dataset and the process repeated”, which in view of Sharma is used to include noise as a decoy, such that the second training data is artificial decoy data, Sharma, Pg. 11, Col. 5-6, Ln. 18-47, “A model may leak membership information in response to a membership inference attack as a result of overfitting. Overfitting may occur where a machine learning algorithm builds a model that fits too closely to a limited set of data . . . In situations where overfitting has occurred, an adversary may use a model's confidence for a prediction provided in response to a sample input to infer whether the sample is or is not a member of the training data . . . Adding noise to causal models may further strengthen them against privacy attacks . . . There may be different ways to add noise to a machine learning model. Noise can be added to a model at the input level (such as by adding randomized inputs to the training data)”, where the “noise” is added to the “training data” to generate data that protects against attacks “to infer whether the sample is or is not a member of the training data” and where the updated “training data” is within the broadest reasonable interpretation of artificial decoy data because it comprises artificial values, “randomized inputs”, which distract from the signal of sensitive data that would otherwise “result of overfitting”).
The reasons for obviousness were discussed in regard to the rejection of Claim 1 above and remain applicable here.
Regarding Claim 5, Dupre teaches an information processing method (Pg. 4339, Sec. “III. Methodology”, “Algorithm 1 Iterative Learning”; Pg. 4342, Sec. “IV. Results”, Subsec. “A. Experiment Environment”, “an iterative loop is set up as described in Algorithm 1 and is run for the three models across the three datasets)
for a computer to execute a process (Pg. 4347, Sec. “Appendix III Used Hardware”, “All our experiments were conducted on a Dell Precision Tower 7910 XCTO server”).
The remaining limitations are substantially the same as the limitations of Claim 1, therefore it is rejected under the same rationale.
Regarding Claim 6, the additional elements of the dependent claim are substantially the same as the limitations of Claim 2, therefore it is rejected under the same rationale.
Regarding Claim 7, the additional elements of the dependent claim are substantially the same as the limitations of Claim 3, therefore it is rejected under the same rationale.
Regarding Claim 9, Dupre teaches an information processing apparatus comprising: one or more memories; and one or more processors coupled to the one or more memories and the one or more processors configured to (Pg. 4347, Sec. “Appendix III Used Hardware”, “All our experiments were conducted on a Dell Precision Tower 7910 XCTO server with two GPUs TITAN Xp, Intel Xeon E5-2623 v3, 128GB DDR4 RAM and 512GB SSD”, where the “TITAN Xp” and “Intel Xeon E5-2623 v3” processors are coupled with the “128GB DDR4 RAM and 512GB SSD” memories in the “Dell Precision Tower 7910 XCTO server” information processing apparatus):
The remaining limitations are substantially the same as the limitations of Claim 1, therefore it is rejected under the same rationale.
Regarding Claim 10, the additional elements of the dependent claim are substantially the same as the limitations of Claim 2, therefore it is rejected under the same rationale.
Regarding Claim 11, the additional elements of the dependent claim are substantially the same as the limitations of Claim 3, therefore it is rejected under the same rationale.
Response to Arguments
Applicant’s arguments filed on March 11, 2026 have been fully considered. Each argument is addressed below.
I. Applicant argues the objections to the specification should be withdrawn (Applicant Arguments/Remarks, 03/11/2026, “Specification Objections”, Pg. 6).
Applicant’s amendments have overcome each and every objection to the specification, as previously recited in the 12/11/2025, 08/14/2025, and 04/22/2025 Office Actions. As a result, the objections to the specification have been withdrawn.
II. Applicant argues the rejections of claims 1-3, 5-7, and 9-11, under 35 U.S.C. § 101, should be withdrawn (Applicant Arguments/Remarks, 03/11/2026, “Claim Rejections Under 35 U.S.C. §101”, Pg. 6-7).
Applicant’s amendments have overcome each and every 35 U.S.C. §101-based rejection to the claims, as previously recited in the 12/11/2025 Office Action. As a result, the rejections to the claims, under 35 U.S.C. §101, have been withdrawn.
III. Applicant argues the rejections of claims 1-3, 5-7, and 9-11, under 35 U.S.C. § 103, should be withdrawn (Applicant Arguments/Remarks, 03/11/2026, “Claim Rejections Under 35 U.S.C. §103”, Pg. 8-9).
In response to Applicant’s amendments, the previously communicated rejections under 35 U.S.C. § 103, have been withdrawn. However, Applicants arguments are not persuasive in light of the new grounds for rejection, under 35 U.S.C. § 103, discussed in detail above. The new grounds of rejection rely on new prior art of record to teach the new combination of elements in the amended independent claims, which were not presented in any of the previously presented claims. As a result, Applicant arguments against the previously communicated rejections under 35 U.S.C. § 103 are rendered moot.
However, for clarity of the record and to expedite prosecution, arguments that remain relevant to the new grounds of rejection are discussed below.
Specifically, Applicant argues a person of ordinary skill in the art “would have no motivation to pursue a solution” where Dupre is modified to “creat[e] entirely new, non-real data from noise for the purpose of deception and security” because Dupre deals with a distinct domain of “improv[ing] model’s accuracy on real-world classification tasks (Pg. 9, Para. 1).
According to MPEP 2143(I), “example rationales that may support a conclusion of obviousness include: . . . Some teaching, suggestion, or motivation in the prior art that would have led one of ordinary skill to modify the prior art reference or to combine prior art reference teachings to arrive at the claimed invention . . . courts have made clear that the teaching, suggestion, or motivation test is flexible and an explicit suggestion to combine the prior art is not necessary. The motivation to combine may be implicit and may be found in the knowledge of one of ordinary skill in the art, or, in some cases, from the nature of the problem to be solved”.
Here, as discussed in further detail above, a person with ordinary skill in the art would have been motivated to modify Dupre’s confidence-based addition of samples to training data (Dupre, Pg 4341, Col. 1, Para. 2, “Using a defined threshold Tc, samples can now be approved for inclusion in the labelled dataset Dl updated for use in the next training iteration”) to arrive at the subject matter of the claims in order to balance model accuracy and training data privacy protection (Sharma, Pg. 11, Col. 6, Ln. 40 - 53, “Adding noise to a machine learning model may reduce the accuracy of the machine learning model but reduce the likelihood that an adversary can use a privacy attack to learn information about the training data . . . The amount of noise added and the level or step at which noise is added may impact how much the accuracy of the model is reduced and how much less likely it is that the model will be susceptible to a privacy attack”; Dupre, Pg. 4340-4341, Col. 2-1, Para. 4-1, “The weighting itself is found experimentally by evaluating the accuracy of each metric . . . Importantly these values may change based on application as certain metrics may be more informative in different problems”).
As a result, the argument is not persuasive.
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 MATTHEW BRYCE GOLAN whose telephone number is (571)272-5159. The examiner can normally be reached Monday through Friday, 8:00 AM to 5:00 PM ET.
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, Alexey Shmatov can be reached at (571) 270-3428. 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.
/MATTHEW BRYCE GOLAN/Examiner, Art Unit 2123
/ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123
Prosecution Insights