CTFR 17/961,277 CTFR 98338 DETAILED ACTION Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia 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 07-37 Applicant's arguments filed 01/26/2026 have been fully considered but they are not fully persuasive. Regarding the 101 rejections , applicant’s arguments and amendments to the independent claims are persuasive and overcome the previous 101 rejections. Specifically, applicant’s amended limitations of performing a sampling procedure to generate neighborhood samples for a given input using said trained autoencoder while controlling a maximum distortion level provides a technical improvement because the specifically trained autoencoder is being used to generate neighborhood samples for post-hoc local explanation methods. See pg. 15 of “Remarks”: “Hence, out of distribution examples or limited types of perturbations have a negative impact on the learned explanations, such as by omitting the relevant features that affect the decision of the post-hoc local explanation method. That is, such out of distribution examples or limited types of perturbations may result in misleading or non- local explanations thereby affecting the decision of the post-hoc local explanation method. Unfortunately, there is not currently a means for generating in-distribution samples of time-series or image data for the neighborhood distribution to be used by post-hoc local explanation methods thereby improving the accuracy of the decision making ability of the post-hoc local explanation method. The claimed invention, such as claimed in claims 1, 8 and 15, addresses such a technical problem with a technical solution by generating in-distribution samples of time-series or image data for the neighborhood distribution to be used by post-hoc local explanation methods. See, e.g., paragraphs [0019, 0020 and 0127] of Applicant's specification.” Applicant’s amendments and corresponding arguments that the claimed invention provides a technical improvement to the field of post-hoc local explanation is persuasive. Therefore, the 101 rejections are withdrawn. Regarding the 103 rejections , applicant’s arguments about reference(s) Tu and Barrett have been fully considered but are not persuasive. Alleged no teaching of performing a sampling procedure to generate neighborhood samples for a given input using said trained autoencoder while controlling a maximum distortion level In Remarks/Arguments pg. 20, applicant contends: “Applicant respectfully asserts that Beckham, Tu, Barrett, and Berthelot, taken singly or in combination, do not teach or suggest "performing a sampling procedure to generate neighborhood samples for a given input using said trained autoencoder while controlling a maximum distortion level" as recited in amended claim 1 and similarly in amended claims 8 and 15.” The relevant claim limitations appear to be: performing a sampling procedure to generate neighborhood samples for a given input using said trained autoencoder while controlling a maximum distortion level in claim 1. Tu and Barrett teach: (Tu, pg. 5 col. 1, “Thus we can adjust c to control how packed the ’neighborhood’ should be in the latent space. If we use a c that’s too big, the constraint will always be dropped thus the whole model falls back to vanilla VAE. If we use a c that’s too small, the model will learn to encode samples to variational distributions with very small variance, thus break the continuity of the latent space and result in poor sampling quality.”). (Barrett, pg. 1 col. 2, “Our work looks to alleviate these issues by providing VAEs whose robustness levels can be controlled and certified by design. To this end, we show how certifiably robust VAEs can be learned by enforcing Lipschitz continuity in the encoder and decoder, which explicitly upper-bounds changes in their outputs with respect to changes in input.”). In other words, Tu teaches performing a sampling procedure to generate neighborhood samples for a given input using said trained autoencoder. Tu shows that a constraint is used to control the neighborhood size when sampling from a neighborhood using a variational autoencoder. Thus, Tu teaches generating neighborhood samples using an autoencoder. Barrett teaches controlling a maximum distortion level of an autoencoder. Barrett shows that a Lipschitz constant is used to control the robustness level of a variational autoencoder and controls the upper bound of changes allowed in the outputs of the autoencoder. Thus, Barrett teaches controlling a maximum distortion level because controlling the robustness level is interpreted as controlling the distortion level and creating an upper bound robustness level is interpreted as setting a maximum distortion level. Therefore, applicant’s arguments are not persuasive. Claim Rejections - 35 USC § 103 07-20-aia AIA 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. 07-21-aia AIA Claim s 1-3, 7-10, and 14-17 are rejected under 35 U.S.C. 103 as being unpatentable over Beckham, et al., Non-Patent Literature “On Adversarial Mixup Resynthesis” (“Beckham”) in view of Tu, et al., Non-Patent Literature “Neighbor Embedding Variational Autoencoder” (“Tu”) and further in view of Barrett, et al., Non-Patent Literature “Certifiably Robust Variational Autoencoders” (“Barrett”) . Regarding claim 1 , Beckham discloses: …method for generating in-distribution samples of data…to be used by post-hoc local explanation methods, (Beckham, pg. 3, “We also explore a strategy in which we randomly retain some components of the hidden representation from h1 and use the rest from h2, and in this case we would randomly sample a binary mask m ∈ {0, 1} k (where k denotes the number of feature maps)… where m is sampled from a Bernoulli(p) distribution (p can simply be sampled uniformly) [ …method for generating in-distribution samples of data ]” and Beckham, pg. 1, “Other goals include learning interpretable representations (Chen et al., 2016; Jang et al., 2016) [ …to be used by post-hoc local explanation methods, ]”). the method comprising: training an autoencoder to generate in-distribution samples of input data…to be used by a post-hoc local explanation method, (Beckham, pg. 3 and see Figure 2, “We also explore a strategy in which we randomly retain some components of the hidden representation from h1 and use the rest from h2, and in this case we would randomly sample a binary mask m ∈ {0, 1} k (where k denotes the number of feature maps)… where m is sampled from a Bernoulli(p) distribution (p can simply be sampled uniformly); Figure 2 shows that an autoencoder performs the generation of samples (i.e. the method comprising: training an autoencoder to generate in-distribution samples of input data )” and Beckham, pg. 1, “Other goals include learning interpretable representations (Chen et al., 2016; Jang et al., 2016) [ …to be used by a post-hoc local explanation method, ]”). wherein said input data comprises time-series or image data, (Beckham, pg. 1, “We know that data augmentation greatly helps when it comes to increasing generalisation performance of models. A practical intuition for why this is the case is that by generating additional samples, we are training our model on a set of examples that better covers those in the test set. In the case of images [ wherein said input data comprises time-series or image data, ]”). wherein said training comprises: mapping said input data into a latent dimension forming a first latent code and a second latent code by an encoder; (Beckham, pg. 3, “What we would like to do is to be able to encode an arbitrary pair of inputs h1 = f(x1) and h2 = f(x2) into their latent representation [ wherein said training comprises: mapping said input data into a latent dimension forming a first latent code and a second latent code by an encoder; ]”). obtaining a mixed code by convexly combining said first and second latent codes with a random coefficient by a mixing block; (Beckham, pg. 2, “Mixup (Zhang et al., 2018) is a regularisation technique which encourages deep neural networks to behave linearly between pairs of data points. These methods artificially augment the training set by producing random convex combinations between pairs of examples and their corresponding labels and training the network on these combinations [ obtaining a mixed code by convexly combining said first and second latent codes with a random coefficient by a mixing block; ].”). decoding said mixed code along with said input data masked with interpretable features to obtain conditional mixed reconstructions by a decoder; (Beckham, pg. 3, “What we would like to do is to be able to encode an arbitrary pair of inputs h1 = f(x1) and h2 = f(x2) into their latent representation [ along with said input data masked with interpretable features ], perform some combination between them through a function we denote Mix(h1, h2) (more on this soon), run the result through the decoder g(·), and then minimise some loss function which encourages the resulting decoded mix to look realistic [ decoding said mixed code…to obtain conditional mixed reconstructions by a decoder; ].”). and performing adversarial training against a discriminator by computing reconstruction losses of said conditional mixed reconstructions and computing discriminator losses and minimizing said reconstruction losses and said discriminator losses ; (Beckham, pg. 3 and see equation 3, “With this in mind, we propose adversarial mixup resynthesis (AMR), where part of the auto-encoder’s objective is to produce mixes which, when decoded, are indistinguishable from real images. The generator and the discriminator of AMR are trained by the following mixture of loss components; equation 3 shows that the adversarial training trains on minimizing the reconstruction and discriminator losses (i.e. and performing adversarial training against a discriminator by computing reconstruction losses of said conditional mixed reconstructions and computing discriminator losses and minimizing said reconstruction losses and said discriminator losses ; )”). While Beckham teaches a system for generating mixed samples using an autoencoder and discriminator, Beckham does not explicitly teach: A computer-implemented… …for a neighborhood distribution… and performing a sampling procedure to generate neighborhood samples for a given input using said trained autoencoder while controlling a maximum distortion level. Tu teaches: …for a neighborhood distribution… (Tu, pg. 4 col. 1, “Inspired by the successful dimension reduction and high dimensional data visualization technique, Stochastic Neighbor Embedding (Hinton & Roweis, 2002), we propose to enforce additional constraint on the learned representation, i.e. the learned low-dimensional latent manifold should maintain the topological features of the input manifold as much as possible. In this case, we would love the the latent codes z1, z2 ∈ Z of two input sample x1, x2 ∈ X to be close to each other in Z if x1, x2 are close to each other in X [ …for a neighborhood distribution… ].”). and performing a sampling procedure to generate neighborhood samples for a given input using said trained autoencoder… (Tu, pg. 5 col. 1, “Thus we can adjust c to control how packed the ’neighborhood’ should be in the latent space. If we use a c that’s too big, the constraint will always be dropped thus the whole model falls back to vanilla VAE. If we use a c that’s too small, the model will learn to encode samples to variational distributions with very small variance, thus break the continuity of the latent space and result in poor sampling quality [ and performing a sampling procedure to generate neighborhood samples for a given input using said trained autoencoder… ].”). Beckham and Tu are both in the same field of endeavor (i.e. autoencoders). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Beckham and Tu to teach the above limitation(s). The motivation for doing so is that considering a neighborhood distribution improves the output by reflecting the input distribution of the two inputs (cf. Tu pg. 4 col. 1, “If this can be achieved, we can expect the learned latent space to be much more regular than other ill-formed solutions, and potentially reduce bad local minimal solutions for downstream tasks.”). While Beckham in view of Tu teaches a system that generates mixed samples while considering neighborhood distributions, the combination does not explicitly teach: A computer-implemented… …while controlling a maximum distortion level. Barrett teaches A computer-implemented… (Barrett, pg. 21, “All models were trained on a 13-inch Macbook Pro from 2017 with 8GB of RAM and 2 CPUs [ A computer-implemented… ].”). Beckham, in view of Tu, and Barrett are both in the same field of endeavor (i.e. autoencoders). Barrett teaches a known technique of using a computer to perform machine learning functions. It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Beckham, in view of Tu, and Barrett to teach the above limitation(s). The motivation for doing so is that applying Barrett’s known technique of using a computer to perform machine learning to Beckham, in view of Tu,’s base system of performing machine learning would yield predictable results. Barrett also teaches …while controlling a maximum distortion level. (Barrett, pg. 1 col. 2, “Our work looks to alleviate these issues by providing VAEs whose robustness levels can be controlled and certified by design. To this end, we show how certifiably robust VAEs can be learned by enforcing Lipschitz continuity in the encoder and decoder, which explicitly upper-bounds changes in their outputs with respect to changes in input [ …while controlling a maximum distortion level. ].”). Beckham, Tu, and Barrett are all in the same field of endeavor (i.e. autoencoder). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Beckham, Tu, and Barrett to teach the above limitation(s). The motivation for doing so is that using Lipschitz constants in an encoder/decoder improves the consistency of the decoded outputs by ensuring the outputs reside within allowed ranges (cf. Barrett, abstract, “Specifically, we first derive actionable bounds on the minimal size of an input perturbation required to change a VAE’s reconstruction by more than an allowed amount, with these bounds depending on certain key parameters such as the Lipschitz constants of the encoder and decoder.”). Regarding claim 2 , Beckham in view of Tu and Barrett teaches the method as recited in claim 1 . Beckham further teaches: further comprising: randomly selecting a mixing sample from said input data; (Beckham, pg. 2, “These methods artificially augment the training set by producing random convex combinations between pairs of examples and their corresponding labels and training the network on these combinations [ randomly selecting a mixing sample from said input data; ].”). and obtaining latent codes of an instance of said mixing sample. (Beckham, pg. 3, “What we would like to do is to be able to encode an arbitrary pair of inputs h1 = f(x1) and h2 = f(x2) into their latent representation [ and obtaining latent codes of an instance of said mixing sample. ], perform some combination between them through a function we denote Mix(h1, h2)”). Regarding claim 3 , Beckham in view of Tu and Barrett teaches the method as recited in claim 2 . Barrett further teaches obtaining an estimation of a Lipschitz constant of said decoder. (Barrett, pg. 1 col. 2, “Our work looks to alleviate these issues by providing VAEs whose robustness levels can be controlled and certified by design. To this end, we show how certifiably robust VAEs can be learned by enforcing Lipschitz continuity in the encoder and decoder, which explicitly upper-bounds changes in their outputs with respect to changes in input [ obtaining an estimation of a Lipschitz constant of said decoder. ].”). It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Barrett with the teachings of Beckham and Tu for the same reasons disclosed in claim 1. Regarding claim 7 , Beckham in view of Tu and Barrett teaches the method as recited in claim 1 . Beckham further teaches wherein said interpretable features are obtained from an interpretable feature map. (Beckham, pg. 3, “We also explore a strategy in which we randomly retain some components of the hidden representation from h1 and use the rest from h2, and in this case we would randomly sample a binary mask m E {1, 0}k (where k denotes the number of feature maps) [ wherein said interpretable features are obtained from an interpretable feature map. ] and perform the following operation: equation (5) where m is sampled from a Bernoulli(p) distribution (p can simply be sampled uniformly) and multiplication is element-wise.”). Regarding claim 8 , the claim is similar to claim 1. Barrett further teaches the additional limitations A computer program product…the computer program product comprising one or more computer readable storage mediums having program code embodied therewith, the program code comprising programming instructions for: (Barrett, pg. 21, “All models were trained on a 13-inch Macbook Pro from 2017 with 8GB of RAM and 2 CPUs [ A computer program product…the computer program product comprising one or more computer readable storage mediums having program code embodied therewith, the program code comprising programming instructions for: ].”). Beckham, in view of Tu, and Barrett are both in the same field of endeavor (i.e. autoencoders). Barrett teaches a known technique of using a computer to perform machine learning functions. It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Beckham, in view of Tu, and Barrett to teach the above limitation(s). The motivation for doing so is that applying Barrett’s known technique of using a computer to perform machine learning to Beckham, in view of Tu,’s base system of performing machine learning would yield predictable results. Regarding claims 9-10 and 14 , the claims are similar to claims 2-3 and 7 and thus rejected under the same rationales. Regarding claim 15 , the claim is similar to claim 1. Barrett further teaches the additional limitations A system, comprising: a memory for storing a computer program…and a processor connected to said memory, wherein said processor is configured to execute program instructions of the computer program comprising: (Barrett, pg. 21, “All models were trained on a 13-inch Macbook Pro from 2017 with 8GB of RAM and 2 CPUs [ A system, comprising: a memory for storing a computer program…and a processor connected to said memory, wherein said processor is configured to execute program instructions of the computer program comprising: ].”). Beckham, in view of Tu, and Barrett are both in the same field of endeavor (i.e. autoencoders). Barrett teaches a known technique of using a computer to perform machine learning functions. It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Beckham, in view of Tu, and Barrett to teach the above limitation(s). The motivation for doing so is that applying Barrett’s known technique of using a computer to perform machine learning to Beckham, in view of Tu,’s base system of performing machine learning would yield predictable results. Regarding claims 16-17 , the claims are similar to claims 2-3 and thus rejected under the same rationales . 07-21-aia AIA Claim s 4-6, 11-13, and 18-20 are rejected under 35 U.S.C. 103 as being unpatentable over Beckham, et al., Non-Patent Literature “On Adversarial Mixup Resynthesis” (“Beckham”) in view of Tu, et al., Non-Patent Literature “Neighbor Embedding Variational Autoencoder” (“Tu”) and further in view of Barrett, et al., Non-Patent Literature “Certifiably Robust Variational Autoencoders” (“Barrett”) and Berthelot, et al., Non-Patent Literature “Understanding and Improving Interpolation in Autoencoders via an Adversarial Regularizer” (“Berthelot”) . Regarding claim 4 , Beckham in view of Tu and Barrett teaches the method as recited in claim 3 . Barrett also teaches and said Lipschitz constant of said decoder as seen in claim 3. While the combination teaches a system for generating mixed samples using an autoencoder with Lipschitz constants, the combination does not explicitly teach: computing an upper bound on a mixing coefficient using positions in said latent dimension represented by said obtained latent codes Berthelot teaches computing an upper bound on a mixing coefficient using positions in said latent dimension represented by said obtained latent codes (Berthelot, pg. 3, “To enforce this constraint we introduce a critic network, as is done in Generative Adversarial Networks (GANs) [12]. The critic is fed interpolations of existing datapoints [ using positions in said latent dimension represented by said obtained latent codes ] (i.e. xˆα as defined above). Its goal is to predict α from xˆα, i.e. to predict the mixing coefficient used to generate its input. In order to resolve the ambiguity between predicting α and 1 − α, we constrain α to the range [0, 0.5] [ computing an upper bound on a mixing coefficient ] when feeding xˆα to the critic predicting the mixing coefficient.”). Beckham, in view of Tu and Barrett, and Berthelot are both in the same field of endeavor (i.e. mixed sample generation). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Beckham, in view of Tu and Barrett, and Berthelot to teach the above limitation(s). The motivation for doing so is that setting a maximum mixing coefficient aids in ensuring that the mixing of two inputs is semantically relevant (cf. Berthelot, pg. 2, “Ideally, adjusting α from 0 to 1 will produce a sequence of realistic datapoints where each subsequent xˆα is progressively less semantically similar to x1 and more semantically similar to x2.”). Regarding claim 5 , Beckham in view of Tu, Barrett, and Berthelot teaches the method as recited in claim 4 . Barrett also teaches and said estimation of said Lipschitz constant of said decoder as seen in claim 3. Beckham further teaches sampling interpretable features of said mixing sample forming a perturbation mask using said trained autoencoder at a sampling time (Beckham, pg. 3 and see Figure 3, “We also explore a strategy in which we randomly retain some components of the hidden representation from h1 and use the rest from h2 [ sampling interpretable features of said mixing sample ], and in this case we would randomly sample a binary mask m ∈ {0, 1} k (where k denotes the number of feature maps); Figure 2 shows that an autoencoder performs the generation of the samples (i.e. forming a perturbation mask using said trained autoencoder at a sampling time )”). Berthelot further teaches with a distortion level controlled by said upper bound on said mixing coefficient (Berthelot, pg. 3, “To enforce this constraint we introduce a critic network, as is done in Generative Adversarial Networks (GANs) [12]. The critic is fed interpolations of existing datapoints (i.e. xˆα as defined above). Its goal is to predict α from xˆα, i.e. to predict the mixing coefficient used to generate its input. In order to resolve the ambiguity between predicting α and 1 − α, we constrain α to the range [0, 0.5]; the mixing coefficient is interpreted as a distortion level as the value of the mixing coefficient dictates how realistic the mixing between two samples will be (i.e. with a distortion level controlled by said upper bound on said mixing coefficient ) when feeding xˆα to the critic predicting the mixing coefficient.”). It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Berthelot with the teachings of Beckham, Tu, and Barrett for the same reasons disclosed in claim 4. Regarding claim 6 , Beckham in view of Tu, Barrett, and Berthelot teaches the method as recited in claim 5 . Tu teaches the neighbor consideration as seen in claim 1. Beckham further teaches obtaining a corresponding…sample by decoding said mixed code by said trained autoencoder along with an original instance of said input data masked by said perturbation mask. (Beckham, pg. 3, “What we would like to do is to be able to encode an arbitrary pair of inputs h1 = f(x1) and h2 = f(x2) into their latent representation, perform some combination between them through a function we denote Mix(h1, h2) (more on this soon), run the result through the decoder g(·), and then minimise some loss function which encourages the resulting decoded mix to look realistic. With this in mind, we propose adversarial mixup resynthesis (AMR), where part of the auto-encoder’s objective is to produce mixes which, when decoded, are indistinguishable from real images [ obtaining a corresponding…sample by decoding said mixed code by said trained autoencoder ]…We also explore a strategy in which we randomly retain some components of the hidden representation from h1 and use the rest from h2, and in this case we would randomly sample a binary mask m ∈ {0, 1} k (where k denotes the number of feature maps) [ along with an original instance of said input data masked by said perturbation mask. ]”). Regarding claims 11 and 18 , the claims are similar to claim 4 and thus rejected under the same rationale. Regarding claims 12 and 19 , the claims are similar to claim 5 and thus rejected under the same rationale. Regarding claims 13 and 20 , the claims are similar to claim 6 and thus rejected under the same rationale . Conclusion 07-96 AIA The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Venkataramanan, et al., “AlignMixup: Improving Representations By Interpolating Aligned Feature” discloses a system that generates mixed samples by using an autoencoder that aligns the feature tensors between two images during mixup synthesis . 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 NICHOLAS S WU whose telephone number is (571)270-0939. The examiner can normally be reached Monday - Friday 8:00 am - 4:00 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, Michelle Bechtold can be reached at 571-431-0762. 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. /N.S.W./Examiner, Art Unit 2148 /MICHELLE T BECHTOLD/Supervisory Patent Examiner, Art Unit 2148 Application/Control Number: 17/961,277 Page 2 Art Unit: 2148 Application/Control Number: 17/961,277 Page 3 Art Unit: 2148 Application/Control Number: 17/961,277 Page 4 Art Unit: 2148 Application/Control Number: 17/961,277 Page 5 Art Unit: 2148 Application/Control Number: 17/961,277 Page 6 Art Unit: 2148 Application/Control Number: 17/961,277 Page 7 Art Unit: 2148 Application/Control Number: 17/961,277 Page 8 Art Unit: 2148 Application/Control Number: 17/961,277 Page 9 Art Unit: 2148 Application/Control Number: 17/961,277 Page 10 Art Unit: 2148 Application/Control Number: 17/961,277 Page 11 Art Unit: 2148 Application/Control Number: 17/961,277 Page 12 Art Unit: 2148 Application/Control Number: 17/961,277 Page 13 Art Unit: 2148 Application/Control Number: 17/961,277 Page 14 Art Unit: 2148 Application/Control Number: 17/961,277 Page 15 Art Unit: 2148 Application/Control Number: 17/961,277 Page 16 Art Unit: 2148 Application/Control Number: 17/961,277 Page 17 Art Unit: 2148 Application/Control Number: 17/961,277 Page 18 Art Unit: 2148
Prosecution Insights