DETAILED ACTION
This final rejection is responsive to the amendment filed on February 24, 2026. Claims 1-3, 5-6, 8-10, 12-13, 15-17, and 19-25 are pending. Claims 1, 8, and 15 are independent. Claims 4, 7, 11, 14, and 18 are canceled. Claims 21-25 are added.
Claim rejections under 35 USC §101 are withdrawn in light of applicant’s arguments – see Response to Arguments below.
Claim rejections under 35 USC §103 are withdrawn, however, a new grounds of rejection under 35 USC §103 has been made in light of applicant’s amendment.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 103
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 1-2, 5-6, 8-9, 12-13, 15-16, and 19-25 are rejected under 35 U.S.C. 103 as being unpatentable over Ganguly (US20220198297), hereinafter Ganguly, in view of Sharpe et al. (US20240037427), hereinafter Sharpe, in view of Settipalli and Dasireddy (Reducing Unintended bias in Text Classification using Multitask Learning), hereinafter Settipalli.
Ganguly and Settipalli were cited in applicant’s IDS filed on February 5, 2026.
Regarding claim 1, Ganguly teaches the method:
clustering a data set into first clusters to obtain first cluster assignments for records of the data set; (Ganguly, paragraph 0035: Furthermore, system 400 can include an input subspace analyzer 460. This input subspace analyzer 460 can cluster the data or abstract representations of the data from intermediate layers of the machine-learning model 150.” – clustering the input data is analogous to the first cluster assignment.)
training a multiheaded inference model using the records and the first cluster assignments; (Ganguly, paragraph 0033: “Turning now to FIG. 3, a schematic diagram is provided of a neural network architecture 300 for jointly learning to perform well in a primary task and poorly in a pseudo-bias task. Data 310 is received from the machine-learning model 150. The data 310 is used to create a shared layer of abstract representations 320, which is not specific to primary task output or bias output. The shared layer of abstract representations 320 is then split into subsections comprising a primary task specific latent representation 330 and one or more debiasing specific latent representations 340.” – training a shared layer abstraction representation then being split into subsections comprising the primary task specific latent representation and one or more debiasing specific latent representations is analogous to training a multiheaded inference model.)
performing modified split training of the multiheaded inference model to obtain an updated multiheaded inference model; (Ganguly, paragraph 0029: “The multi-objective learning component 140 can train the machine-learning model 150 by creating a set of pseudo-task variables. These pseudo-task variables can be used to set a “bias task” in the same way in which the primary task variables are used to train the machine-learning model 150 on the primary task.” – The training being a multi-objective learning to train the model is analogous to the modified split training.)
…a predictive power level of the multiheaded inference model with respect to the first cluster assignments and a predictive power level for bias features, (Ganguly, paragraph 0031: “This can be achieved by employing the following multi-objective loss function:
L
=
P
y
x
;
θ
p
,
θ
s
-
∑
i
-
1
n
P
y
x
;
θ
i
B
,
θ
s
,
which learns to perform above a first threshold on a primary task and perform below a second threshold (note the − sign in the second half of the function) on a pseudo-bias task. In various embodiments, the performance is effective based on a determination that the likelihood is maximized and/or the performance is poor based on an inverse determination that the likelihood is maximized.” – The loss function being optimized based on the performance of the primary task and pseudo-bias task is analogous to the predictive power level of the model with respect to the cluster assignment and bias features.)
clustering the updated data set into second clusters to obtain second cluster assignments for records of the updated data set; and providing computer implemented services using the second cluster assignments, (Ganguly, paragraph 0051: “At 1104, the system clusters (e.g., via input subspace analyzer 460 operatively coupled to a processor 110) the data or abstract representations of the data. At 1105, the system maps (e.g., via reverse mapper component 470 operatively coupled to a processor 110) a subset of the predictions to the clustered data or clustered abstract representations of the data. At 1106, the system reinforces (e.g., via reverse mapper component 470 operatively coupled to a processor 110) the mapping with extrinsic data that either suggests which subset of the clustered data or clustered abstract representations of the data should be considered for setting pseudo-task variables or serves as a set of bias examples to facilitate associating prediction classes with at least one of the clusters of the clustered data or clustered abstract representations of the data.” And paragraph 0052: “The multi-objective learning component 140 then outputs bias-mitigated predictions 1290, by training the machine-learning model 150 to perform well on the primary task, and poorly on the bias task (or inversely, well on the debiasing or fairness task).” The subsets of the cluster data being reverse mapped is analogous to the second cluster assignments. After the reverse mapping, this is used as an updated data set which is then further used to output bias-mitigated predictions, e.g., the computer implemented services.)
wherein the modified split training is performed using a training data set that associates features of the data set with the first cluster assignments and the bias features, (Ganguly, Fig. 3 and paragraph 0033: “Data 310 is received from the machine-learning model 150. The data 310 is used to create a shared layer of abstract representations 320, which is not specific to primary task output or bias output. The shared layer of abstract representations 320 is then split into subsections comprising a primary task specific latent representation 330 and one or more debiasing specific latent representations 340.” – The subsections comprising a primary task and one or more debiasing specific representations is analogous to using a training dataset that associates features of the data set with the first cluster assignment and the bias features.)
wherein obtaining the updated data set comprises: for each feature of the data set: (Ganguly, Fig. 5 shows that the data set is updated based off the predictions.)
identifying a first level of contribution of the feature to a predictive power level of first inferences for the first cluster assignments; identifying a second level of contribution of the feature to a predictive power level of second inferences for bias features; (Ganguly, paragraph 0031: Once pseudo-task variables have been set, then the multi-objective learning component 140 can train the machine-learning model 150. This can be achieved by employing the following multi-objective loss function:
L
=
P
y
x
;
θ
p
,
θ
s
-
∑
i
-
1
n
P
y
x
;
θ
i
B
,
θ
s
,
which learns to perform above a first threshold on a primary task and perform below a second threshold (note the − sign in the second half of the function) on a pseudo-bias task. – The first term in the loss function is the predictive power level of first inferences for the first cluster and the second term is the predictive power level of second inferences for bias features.)
Ganguly does not explicitly teach:
obtaining an updated data set using the updated multiheaded inference model by removing at least one feature from the data set based on a predictive power level of the multiheaded inference model with respect to the first cluster assignments and a predictive power level for bias features, the updated data set comprises less data than the data set;
identifying a partial latent bias score by adding a first scalar value associated with the first level of contribution and a second scalar value associated with the second level of contribution; and
adding the partial latent bias score to a listing of bias scores.
However, Sharpe teaches:
obtaining an updated data set using the updated multiheaded inference model by removing at least one feature from the data set based on a predictive power level of the multiheaded inference model with respect to the first cluster assignments and a predictive power level for bias features, (Sharpe, paragraph 0018: “The ML explanation system 102 may determine a first importance metric having a greatest importance among the feature set, for example, based on the sorting of the plurality of importance metrics. This may allow the ML explanation system 102 to determine an order in which to drop (e.g., remove) features from the dataset and may enable the ML explanation system 102 to evaluate the effectiveness of the importance metrics.” - The ML explanation system being analogous to the multiheaded inference model that is taught by Ganguly above. The importance metrics is analogous to predictive power levels while Ganguly already teaches that the predictive power levels are the predictive power levels based on the first cluster assignment and bias features.) the updated data set comprises less data than the data set; (Sharpe, paragraph 0004: “Evaluating, the model and importance features in this way eliminates the need for replacing data with perturbations and their associated drawbacks described above.” And paragraph 0022: “The second dataset may include each feature of the feature set that was included in the first dataset except for the subset of the first dataset that was filtered (e.g., removed).”)
Sharpe is considered analogous to the claimed invention as it is in the same field of endeavor, machine learning. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to have modified Ganguly, which already teaches training a multiheaded inference model but does not explicitly teach updating the dataset by removing at least one feature based on predictive power levels, to include the teachings of Sharpe which does teach updating the dataset by removing at least one feature based on predictive power levels in order to "provide for the comparison of the effectiveness of different model explanation techniques and allow for the identification of features in data that contribute to decisions made by a model." (Sharpe, paragraph 0004)
Ganguly and Sharpe do not explicitly teach:
identifying a partial latent bias score by adding a first scalar value associated with the first level of contribution and a second scalar value associated with the second level of contribution; and
adding the partial latent bias score to a listing of bias scores.
However, Settipalli teaches:
identifying a partial latent bias score by adding a first scalar value associated with the first level of contribution and a second scalar value associated with the second level of contribution; and (Settipalli, page 33, section 3.5.1, paragraph 1: “The loss function that we used is as follows
PNG
media_image1.png
79
582
media_image1.png
Greyscale
The loss function mentioned above is a weighted cross-entropy loss function. By default, the values of α and β are configured as α = 0.6 and β =1. But while dealing with nontoxic examples with identity information the value of β is changed to 3. By using this custom loss function which changes according to the training examples the model is able to put more focus on non-toxic examples with have the identity terms 34 Chapter 4. Experiment during the training process which helps the model reduce identity bias.”- Where the first term of the loss function is analogous to the first level of contribution and β is the first scalar value, the second term of the loss function is analogous to the second level of contribution and α is the second scalar value.)
adding the partial latent bias score to a listing of bias scores. (Settipalli, page 34, section 4.5.2, paragraph 3: “The training data is divided into batches of fixed batch size, then the loss is calculated for every training example in the batch and finally, the total loss is back propagated through the network at once.” – The training being in batches before the total loss is propagated means that the partial latent bias, e.g., after the batch is added to a listing of bias scores in order to update all at once.)
Settipalli is considered analogous to the claimed invention as it is in the same field of endeavor, machine learning. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to have modified Ganguly and Sharpe, which already teaches updating the data set using the first level of contribution of the feature to a predictive power level of first inferences and a second level of contribution to a predictive power level second inferences of bias features but does not explicitly teach identifying the partial latent bias score by adding a first scalar value associated with the first level of contribution and a second scalar value associated with the second level of contribution and adding the partial latent bias to a listing of bias scores, to include the teachings of Settipalli which does teach identifying the partial latent bias score by adding a first scalar value associated with the first level of contribution and a second scalar value associated with the second level of contribution and adding the partial latent bias to a listing of bias scores in order to "put more focus on non-toxic examples with have the identity terms during the training process which helps the model reduce identity bias." (Settipalli, section 4.5.1, paragraph 2)
Regarding claim 2, Ganguly, Sharpe, and Settipalli teach the method of claim 1, as cited above.
Ganguly further teaches:
a shared body; a feature head that generates the first inferences for the first cluster assignments; and a bias feature head that generates the second inferences for the bias features. (Ganguly, Fig. 3 and paragraph 0033: “The data 310 is used to create a shared layer of abstract representations 320, which is not specific to primary task output or bias output. The shared layer of abstract representations 320 is then split into subsections comprising a primary task specific latent representation 330 and one or more debiasing specific latent representations 340. Primary-task class probabilities 350 assign primary-task category labels. Similarly, debias-task class probabilities 360 assign debiasing task category labels. In an embodiment, a neural network architecture such as this can be implemented by the multi-objective learning component 140.” – The shared layer of abstraction representations is analogous to the shared body while the primary task specific latent representation is analogous to the feature head and the debiasing specific latent representations are analogous to the bias feature head.)
Regarding claim 5, Ganguly, Sharpe, and Settipalli teach the method of claim 1, as cited above.
Ganguly does not explicitly teach:
for each feature of the data set: ranking the feature among the features of the data set based on the first level of contribution and the second level of contribution to identify a worst ranked feature;
using the worst ranked feature as the at least one feature.
However, Sharpe further teaches:
for each feature of the data set: ranking the feature among the features of the data set based on the first level of contribution and the second level of contribution to identify a worst ranked feature; and (Sharpe, paragraph 0073: “The method of any of the preceding embodiments, wherein determining a first importance metric of the plurality of importance metrics comprises sorting the plurality of importance metrics in order from greatest importance to least importance.” – Ordering the importance metrics from greatest importance to least importance is analogous to ranking the feature based on the first level of contribution and second level of contribution.)
using the worst ranked feature as the at least one feature. (Sharpe, paragraph 0018: “This may allow the ML explanation system 102 to determine an order in which to drop (e.g., remove) features from the dataset and may enable the ML explanation system 102 to evaluate the effectiveness of the importance metrics.” – Determining the order in which to drop features is analogous to using the worse ranked feature as the at least one feature that is removed in claim 1.)
Regarding claim 6, Ganguly, Sharpe, and Settipalli teach the method of claim 5, as cited above.
Ganguly does not explicitly teach:
wherein the features are ranked based on a difference between the first level of contribution and the second level of contribution, and the worst ranked feature having a lowest difference of the differences associated with the features.
However, Sharpe further teaches:
wherein the features are ranked based on a difference between the first level of contribution and the second level of contribution, and the worst ranked feature having a lowest difference of the differences associated with the features. (Sharpe, paragraph 0025: “By comparing the first and second sets of performance metrics, the ML explanation system 102 may determine which importance metrics should be used to explain a classification or other output made by the machine learning model.” – Comparing the first and second set of performance metrics to determine the importance metric to be used is analogous to the features being ranked based on the difference between the levels of contribution. Sorting the plurality of importance metrics, as taught in claim 5, shows that the method is capable of sorting based off the differences, as it is being used here to determine which importance metrics are used. This indicates that the worst ranked feature having a lowest difference associated with the features is the at least one feature that is being removed in claim 1.)
Regarding claim 8, Claim 8 has all the same limitations of claim 1 which are taught by Ganguly, Sharpe, and Settipali – see claim 1 above.
Ganguly additionally teaches:
A non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, cause the processor to perform operations for managing an inference model, the operations comprising: (Ganguly, paragraph 0056: “The present invention can be a system, a method, an apparatus and/or a computer program product at any possible technical detail level of integration. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.”)
Regarding claim 9, Ganguly, Sharpe, and Settipalli teach the non-transitory machine-readable medium of claim 8, as cited above.
Claim 9 additionally has the same limitations of claim 2 which are taught by Ganguly, Sharpe, and Settipalli – see claim 2 above.
Regarding claim 12, Ganguly, Sharpe, and Settipalli teach the non-transitory machine-readable medium of claim 8, as cited above.
Claim 12 additionally has the same limitations of claim 5 which are taught by Ganguly, Sharpe, and Settipalli – see claim 5 above.
Regarding claim 13, Ganguly, Sharpe, and Settipalli teach the non-transitory machine-readable medium of claim 12, as cited above.
Claim 13 additionally has the same limitations of claim 6 which are taught by Ganguly, Sharpe, and Settipalli – see claim 6 above.
Regarding claim 15, Claim 15 has all the same limitations of claim 1 which are taught by Ganguly, Sharpe, and Settipalli – see claim 1 above.
Ganguly additionally teaches:
A data processing system, comprising: a processor; and a memory coupled to the processor to store instructions, which when executed by the processor, cause the processor to perform operations for managing an inference model, the operations comprising: (Ganguly, paragraph 0056: “The present invention can be a system, a method, an apparatus and/or a computer program product at any possible technical detail level of integration. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.”)
Regarding claim 16, Ganguly, Sharpe, and Settipalli teach the data processing system of claim 15, as cited above.
Claim 16 additionally has the same limitations of claim 2 which are taught by Ganguly, Sharpe, and Settipalli – see claim 2 above.
Regarding claim 19, Ganguly, Sharpe, and Settipalli teach the data processing system of claim 15, as cited above.
Claim 19 additionally has the same limitations of claim 5 which are taught by Ganguly, Sharpe, and Settipalli – see claim 5 above.
Regarding claim 20, Ganguly, Sharpe, and Settipalli teach the data processing system of claim 19, as cited above.
Claim 20 additionally has the same limitations of claim 6 which are taught by Ganguly, Sharpe, and Settipalli – see claim 6 above.
Regarding claim 21, Ganguly, Sharpe, and Settipalli teach the method of claim 2, as cited above.
Ganguly further teaches:
wherein the shared body and the feature head form a first inference generation path that predicts one or more output labels. (Ganguly, Fig. 3 and paragraph 0033: “Primary-task class probabilities 350 assign primary-task category labels.” – The shared body is the shared layer of abstract representations while the feature head is the primary task specific latent representation which feeds into the primary-task class probabilities and further outputs the primary-task category label which is analogous to the predicted one or more outputs.)
Regarding claim 22, Ganguly, Sharpe, and Settipalli teach the method of claim 2, as cited above.
Ganguly further teaches:
wherein the shared body and the bias feature head form a second inference generation path that predicts the bias features with an associated confidence level. (Ganguly, Fig. 3 and paragraph 0033: “Similarly, debias-task class probabilities 360 assign debiasing task category labels.” – The shared body is the shared layer of abstract representations while the bias feature head is the debiasing specific latent representations and debias-task class probabilities and further outputs the pseudo debiasing task category labels which is analogous to the confidence level of the bias features.)
Regarding claim 23, Ganguly, Sharpe, and Settipalli teach the non-transitory machine-readable medium of claim 9, as cited above.
Claim 23 additionally has the same limitations of claim 21 which are taught by Ganguly, Sharpe, and Settipalli – see claim 21 above.
Regarding claim 24, Ganguly, Sharpe, and Settipalli teach the non-transitory machine-readable medium of claim 9, as cited above.
Claim 24 additionally has the same limitations of claim 22 which are taught by Ganguly, Sharpe, and Settipalli – see claim 22 above.
Regarding claim 25, Ganguly, Sharpe, and Settipalli teach the data processing system of claim 16, as cited above.
Claim 25 additionally has the same limitations of claim 21 which are taught by Ganguly, Sharpe, and Settipalli – see claim 21 above.
Claims 3, 10, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Ganguly in view of Sharpe in view of Settipalli in view of Graham et al. (One model is all you need: Multi-task learning enables simultaneous image segmentation and classification), hereinafter Graham.
Regarding claim 3, Ganguly, Sharpe, and Settipalli teach the method of claim 2, as cited above.
Ganguly, Sharpe, and Settipalli do not explicitly teach:
modifying weights of the feature head to increase the predictive power level of the first inferences;
modifying weights of the bias feature head to increase the predictive power level of the second inferences; and
modifying weights of the shared body using the feature head and the bias feature head.
However, Graham teaches:
modifying weights of the feature head to increase the predictive power level of the first inferences; modifying weights of the bias feature head to increase the predictive power level of the second inferences; and modifying weights of the shared body using the feature head and the bias feature head. (Graham, section 3.2, page 5, column 1, paragraph 1: “After passing a batch through the network, the weights in the encoder 𝛷 are always updated irrespective of how the batch is sampled, whereas the weights of decoder 𝛹𝑡 are only updated when at least one example from task 𝑡 is present.” – eqn 1 defines the loss function which is modifying the weights in order to increase predictive power, e.g. minimizing the loss function between the predictions (inferences) and the ground truth would necessarily increase the predictive power level of the inferences. The weights of the encoder is analogous to the weights of the body, while the weights of the decoder being dependent on the task t shows that each of the feature head and bias feature head, which are taught by Ganguly above, are being updated separately depending on the task i.e. the weights of the decoder layers corresponding to the task of the feature head will be modified separately from the weights of the decoder layers corresponding to the bias feature head.)
Graham is considered analogous to the claimed invention as it is in the same field of endeavor, machine learning. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to have modified Ganguly, Sharpe, and Settipalli, which already teaches training a multiheaded inference model with a shared body, a feature head, and a bias feature head but does not explicitly teach modifying weights of the shared body, the feature head, and the bias feature head, to include the teachings of Graham which does teach modifying weights of the shared body, the feature head, and the bias feature head as training each of the heads separately “capture[s] the contribution of the feature representation.” (Graham, section 3.3, page 5, column 1, paragraph 6)
Regarding claim 10, Ganguly, Sharpe, and Settipalli teach the non-transitory machine-readable medium of claim 9, as cited above.
Claim 10 additionally has the same limitations of claim 3 which are taught by Ganguly, Sharpe, Settipalli, and Graham – see claim 3 above.
Regarding claim 17, Ganguly, Sharpe, and Settipalli teach the data processing system of claim 16, as cited above.
Claim 17 additionally has the same limitations of claim 3 which are taught by Ganguly, Sharpe, Settipalli, and Graham – see claim 3 above.
Response to Arguments
Applicant’s arguments, filed February 24, 2026, with respect to claim rejections under 35 USC §101 have been fully considered and are persuasive. In particular, applicant points out that “the claimed invention improves the quality of services provided using the results of the unsupervised learning process and reduces the computational resources (e.g., data allocation size, bandwidth requirements to transmit/receive the data, computational power to process the data) required to provide the services,” on pages 10-11, which integrates the judicial exception into a practical application. Therefore rejection of claims 1-20 under 35 USC §101 has been withdrawn.
Applicant’s arguments, filed February 24, 2026, with respect to the rejection(s) of claim(s) 1-20 under 35 USC §103 have been fully considered and are persuasive. In particular, examiner agrees that the prior rejection does not teach the amended independent claims. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Ganguly, Sharpe, and Settipalli for claims 1-2, 5-6, 8-9, 11-13, 15-16, and 19-25 and Ganguly, Sharpe, Settipalli, and Graham for claims 3, 10, and 17. See section Claim Rejections – 35 USC §103 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 JACQUELINE MEYER whose telephone number is (703)756-5676. The examiner can normally be reached M-F 8:00 am - 4:30 pm 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, Tamara Kyle can be reached at 571-272-4241. 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/afpply/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.
/J.C.M./Examiner, Art Unit 2144
/TAMARA T KYLE/Supervisory Patent Examiner, Art Unit 2144
Prosecution Insights