CTNF 18/688,257 CTNF 101321 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. Information Disclosure Statement The information disclosure statements (IDS) submitted on 03/11/2024 and 04/18/2024 have been considered by the examiner. Claim Objections Claim 14 is objected to because of the following informality: In claim 14, line 1, “the method of any of claim 11” should read “the method of any of claim 11.” Appropriate correction is required. 07-30-03-h AIA Claim Interpretation 07-30-03 AIA The following is a quotation of 35 U.S.C. 112(f): (f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof. The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph: An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof. 07-30-05 The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked. As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph: (A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function; (B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and (C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function. Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function. Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function. Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Claims 11, 17, and 21 recite limitations that use words like “means” (or “step”) or similar terms with functional language and do invoke 35 U.S.C. 112(f): Claim 11; recites the limitation, “processing the minibatch with the encoder model to generate a plurality of embeddings… , ” [Lines 5-6]. Claim 11; recites the limitation, “processing the plurality of embeddings for the training inputs of the minibatch with the classification model to generate a plurality of classifications… , ” [Lines 7-8]. Claim 17; recites the limitation, “processing the input data with an encoder model to generate an input embedding… , ” [Lines 4-5]. Claim 17; recites the limitation, “processing the input embedding with a classification model to generate an output classification… , ” [Lines 6-7]. Claim 17; recites the limitation, “processing the first input and the second input with an encoder model to generate a first embedding and a second embedding… , ” [Lines 10-11]. Claim 17; recites the limitation, “processing the first embedding and a second embedding with a classification model to generate a first classification and a second classification… , ” [Lines 12-13]. Claim 21; recites the limitation, “processing the first input with an encoder model to generate a first embedding… , ” [Line 8]. Claim 21; recites the limitation, “processing the first embedding with a classification model to generate a first classification… , ” [Lines 9-10]. Claim 21; recites the limitation, “processing the second input with the encoder model to generate a second embedding… , ” [Line 11]. Claim 21; recites the limitation, “processing the second embedding with the classification model to generate a second classification… , ” [Lines 12-13]. Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof. After a careful analysis, as disclosed above, and a careful review of the specification the following limitations in claims 11, 17, and 21: “encoder model” ( Fig. 3, #310 called encoder model, Paragraph [0112] – “Figure 3 depicts a training process 300 that includes a machine-learned model with an encoder model 310 and a classification model 322.” Paragraph [0129] – “The encoder model and the classifier model can be part of a larger machine-learned model. The larger machine-learned model can include a convolutional neural network or other image processing neural network, such as an attention-based neural network.” Thus, the encoder model does have a sufficient structure associated with it wherein it is part of an image processing neural network. “classification model” ( Fig. 3, #322 called classification model, Paragraph [0112] – “Figure 3 depicts a training process 300 that includes a machine-learned model with an encoder model 310 and a classification model 322.” Paragraph [0129] – “The encoder model and the classifier model can be part of a larger machine-learned model. The larger machine-learned model can include a convolutional neural network or other image processing neural network, such as an attention-based neural network.” Thus, the classification model does have a sufficient structure associated with it wherein it is part of an image processing neural network. If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. Claim Rejections - 35 USC § 103 07-06 AIA 15-10-15 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 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. 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 of this title, 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 1-7, 10, 16-19, and 21 are rejected under 35 U.S.C. 103 as being unpatentable over LI (US 20210374553 A1), hereinafter referenced as LI in view of CHEN (US 20210233541 A1), hereinafter referenced as CHEN . Regarding claim 1, LI teaches a computer-implemented method for training a classification model (Figs. 5A-5C, Paragraph [0057] – LI discloses FIGS. 5A-5C provides an example flow diagram illustrating a method of noise-robust contrastive learning. Paragraph [0065] – LI discloses a classification loss may be computed using the class predictions, e.g., generated by the classifier 255.) , the method comprising: obtaining, by a computing system (Fig. 3, #300 called computing device, Paragraph [0046] – LI discloses computing device 300 may be implemented as a stand-alone subsystem, as a board added to a computing device, and/or as a virtual machine.) comprising one or more processors (Fig. 3, #310 called processor, Paragraph [0046] – LI discloses computing device 300 includes a processor 310 coupled to memory 320.) , a training dataset (Fig. 5A, Paragraph [0058] – LI discloses method 500 starts with step 502, at which a training set of data samples may be obtained, each data sample having a noisy label. For example, the training set of data samples may be received as part of input 340 via the data interface 315 in FIG. 3. See also Paragraph [0046].) , wherein the training dataset comprises a first input (Fig. 5A, Paragraph [0060] – LI discloses at step 505, for each data sample, a weak augmented data sample (e.g., 241) and a strong augmented data sample may be generated [wherein weak augmented data sample is a first input].) and a second input (Fig. 5A, Paragraph [0060] – LI discloses at step 505, for each data sample, a weak augmented data sample (e.g., 241) and a strong augmented data sample may be generated [wherein strong augmented data sample is a second input].) ; processing, by the computing system (Fig. 3, #300 called computing device, Paragraph [0046] – LI discloses FIG. 3 is a simplified diagram of a computing device for implementing the noise-robust contrastive learning described in FIGS. 1-2, according to some embodiments. Operation of computing device 300 is controlled by processor 310.) , the first input (Fig. 2, #241 called a weak augmented data sample, Paragraph [0060]) with an encoder model (Fig. 2, #245a-c called a deep encoder CNN, Paragraph [0033] – LI discloses a deep encoder, e.g., a convolutional neural network (CNN) 245a-c (or collectively referred to as 245) that encodes an image x i or an augmented sample of image x i to a high-dimensional feature v i ;) to generate a first embedding (Fig. 2, Paragraph [0035] – LI discloses the weakly-augmented data sample 241 and the strongly-augmented data sample 242 are each input to the CNN 245a or 245b, respectively, each of which encodes the augmented input into a high-dimensional feature v i , respectively [wherein high-dimensional feature v i is a first embedding]. The high-dimensional features from the CNN 245a-b are then input to the autoencoders 260a-b, respectively, each of which performs z i =W ev v i the linear projection from high-dimensional features to low-dimensional embeddings.) ; processing, by the computing system (Fig. 3, #300 called computing device, Paragraph [0046]) , the first embedding with a classification model (Fig. 2, #255 called classifier, Paragraph [0033] – LI discloses a classifier 255 (e.g., a fully-connected layer followed by softmax) that receives v i as input and outputs class predictions;) to generate a first classification (Fig. 2, Paragraph [0042] – LI discloses the output high-dimensional feature v i , are passed to the classifier 255. Given the softmax output from the classifier 255, p(y; x i ), a classification loss may be defined as the cross-entropy loss. The cross-entropy loss L ce may be used to update the classifier 255, e.g., via the backpropagation path shown in dashed line from block 261. See also Paragraph [0033].) ; processing, by the computing system (Fig. 3, #300 called computing device, Paragraph [0046]) , the second input (Fig. 2, #242 called a strongly augmented data sample, Paragraph [0060]) with the encoder model (Fig. 2, #245a-c called a deep encoder CNN, Paragraph [0033] – LI discloses a deep encoder, e.g., a convolutional neural network (CNN) 245a-c (or collectively referred to as 245) that encodes an image x i or an augmented sample of image x i to a high-dimensional feature v i ;) to generate a second embedding (Fig. 2, Paragraph [0035] – LI discloses the weakly-augmented data sample 241 and the strongly-augmented data sample 242 are each input to the CNN 245a or 245b, respectively, each of which encodes the augmented input into a high-dimensional feature v i , respectively [wherein high-dimensional feature v i is a second embedding]. The high-dimensional features from the CNN 245a-b are then input to the autoencoders 260a-b, respectively, each of which performs z i =W ev v i the linear projection from high-dimensional features to low-dimensional embeddings.) ; processing, by the computing system (Fig. 3, #300 called computing device, Paragraph [0046]) , the second embedding with the classification model (Fig. 2, #255 called classifier, Paragraph [0033] – LI discloses a classifier 255 (e.g., a fully-connected layer followed by softmax) that receives v i as input and outputs class predictions;) to generate a second classification (Fig. 2, Paragraph [0042] – LI discloses the output high-dimensional feature v i , are passed to the classifier 255. Given the softmax output from the classifier 255, p(y; x i ), a classification loss may be defined as the cross-entropy loss. The cross-entropy loss L ce may be used to update the classifier 255, e.g., via the backpropagation path shown in dashed line from block 261. See also Paragraph [0033].) ; determining, by the computing system (Fig. 3, #300 called computing device, Paragraph [0046]) , a similarity measure between the first embedding and the second embedding based on a feature similarity (Fig. 5B, Paragraph [0062] – LI discloses at step 510, a consistency contrastive loss [wherein a consistency contrastive loss is a similarity measure] is computed by comparing the first embedding generated from the weakly augmented sample 241 and the second embedding generated from the strongly augmented sample 242, e.g., according to Eq. (1).) ; evaluating, by a computing system (Fig. 3, #300 called computing device, Paragraph [0046]) , a loss function (Fig. 5C, Paragraph [0067] – LI discloses step 520, where a combined loss is computed by adding the weighted sum of all the losses across the batch of samples, e.g., according to Eq. (8).) , wherein the loss function (Fig. 5C, Paragraph [0044] – LI discloses combined loss module 280 may then compute an overall training objective, as to minimize a weighted sum of all losses: L = L ce +ω cc L cc+ ω pc L pc_mix +ω recon L recon, Eq. (8). Paragraph [0045] – LI further discloses the weight parameters may be set as ω cc = 1, ω recon =1, and change ω pc only across datasets. See also Paragraph [0067].) comprises a loss term that evaluates a difference between the first classification and the second classification (Fig. 5B, Paragraph [0065] – LI discloses at step 516, a classification loss may be computed using the class predictions, e.g., generated by the classifier 255. The computation may be done according to Eq. (7). Paragraph [0042] – LI discloses a classification loss may be defined as the cross-entropy loss [wherein ℒ ce is the cross-entropy loss function, see Eq. (7)].) Although LI further teaches weighted by the similarity measure (Fig. 5C, Paragraph [0036] – LI discloses the consistency contrastive loss can be computed at 265 by comparing the normalized embeddings from autoencoders 260a-b [wherein ℒ cc is the consistency contrastive loss function that is a similarity measure, see Eq. (1).] Paragraph [0044] – LI further discloses a combined loss is computed by adding the weighted sum of all the losses across the batch of samples, e.g., according to Eq. (8).) ; LI fails to explicitly teach and adjusting, by the computing system, one or more parameters of the classification model based at least in part on the loss function. However, CHEN explicitly teaches and adjusting, by the computing system (Fig. 3, Paragraph [0064] – CHEN discloses method 300 is performed by a server executing machine-readable software code of the neural network architectures, though it should be appreciated that the various operations may be performed by one or more computing devices and/or processors.) , one or more parameters of the classification model based at least in part on the loss function (Fig. 3, Paragraph [0071] – CHEN discloses the server performs a loss function (e.g., LMCL, LDA) and updates hyper-parameters (or other types of weight values) of the neural network architecture.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI of having a computer-implemented method for training a classification model, the method comprising: obtaining, by a computing system comprising one or more processors, a training dataset, wherein the training dataset comprises a first input and a second input; processing, by the computing system, the first input with an encoder model to generate a first embedding; processing, by the computing system, the first embedding with a classification model to generate a first classification; processing, by the computing system, the second input with the encoder model to generate a second embedding, with the teachings of CHEN having and adjusting, by the computing system, one or more parameters of the classification model based at least in part on the loss function. Wherein LI’s computer implemented method for training a classification model wherein having and adjusting, by the computing system, one or more parameters of the classification model based at least in part on the loss function. The motivation behind this modification would have been to provide an enhanced method of training a classification model that improves accuracy and minimizes error, since both LI and CHEN relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and CHEN relates to systems and methods for managing, training, and deploying neural network architecture for audio processing; the result of training the neural network architecture is to minimize the amount of error between a predicted output and an expected output. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and CHEN (US 20210233541 A1), Paragraph [0005, 0070]. Regarding claim 2, LI in view of CHEN teach the computer-implemented method of claim 1, LI fails to explicitly teach further comprising: evaluating, by the computing system, a second loss function that evaluates a difference between the first classification and a first label, wherein the first label comprises a respective label for the first input, and wherein the first label is obtained from the training dataset; and adjusting, by the computing system, one or more parameters of the classification model based at least in part on the second loss function. However, CHEN explicitly teaches further comprising: evaluating, by the computing system (Fig. 6, Paragraph [0090] – CHEN discloses neural network 600 may be executed by any computing device comprising a processor capable of performing the operations of the neural network 600 and by any number of such computing devices.) , a second loss function that evaluates a difference between the first classification and a first label (Fig. 6, Paragraph [0092] – CHEN discloses loss layer 610 tunes the embedding extractor 606 by performing the loss function (e.g., LMCL) to determine the distance (e.g., large margin cosine loss) between the determined genuine and spoof classifications [wherein determined genuine and spoof classifications are a first classification], as indicated by supervised labels [wherein supervised labels are a first label] or previously generated clusters.) , wherein the first label comprises a respective label for the first input (Fig. 6, Paragraph [0091] – CHEN discloses the server feeds the training audio signals 602 into the input layers 601, where the training audio signals may include any number of genuine and spoofed or false audio signals. Paragraph [0092] – CHEN further discloses spoof embedding extractor 606 is trained by performing a loss layer 610 for learning and tuning spoof embedding according to labels associated with the training audio signals 602.) , and wherein the first label is obtained from the training dataset (Fig. 6, Paragraph [0092] – CHEN discloses spoof embedding extractor 606 is trained by performing a loss layer 610 for learning and tuning spoof embedding according to labels associated with the training audio signals 602.) ; and adjusting, by the computing system (Fig. 6, Paragraph [0090] - CHEN discloses neural network 600 may be executed by any computing device comprising a processor capable of performing the operations of the neural network 600 and by any number of such computing devices.) , one or more parameters of the classification model based at least in part on the second loss function (Fig. 6, Paragraph [0092] – CHEN discloses server fixes the hyper-parameters of the embedding extractor 606 and/or fully-connected layers 608 when predicted outputs (e.g., classifications, feature vectors, embeddings) converge with the expected outputs within a threshold margin of error. Paragraph [0071] – CHEN further discloses the server performs a loss function (e.g., LMCL, LDA) and updates hyper-parameters (or other types of weight values) of the neural network architecture. The server determines the error between the predicted output and the expected output by comparing the similarity or difference between the predicted output and expected output.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI in view of CHEN of having a computer-implemented method for training a classification model, the method comprising: determining, by the computing system, a similarity measure between the first embedding and the second embedding based on a feature similarity; evaluating, by a computing system, a loss function, wherein the loss function comprises a loss term that evaluates a difference between the first classification and the second classification weighted by the similarity measure; with the teachings of CHEN having further comprising: evaluating, by the computing system, a second loss function that evaluates a difference between the first classification and a first label, wherein the first label comprises a respective label for the first input, and wherein the first label is obtained from the training dataset; and adjusting, by the computing system, one or more parameters of the classification model based at least in part on the second loss function. Wherein LI’s computer implemented method for training a classification model further comprising: evaluating, by the computing system, a second loss function that evaluates a difference between the first classification and a first label, wherein the first label comprises a respective label for the first input, and wherein the first label is obtained from the training dataset; and adjusting, by the computing system, one or more parameters of the classification model based at least in part on the second loss function. The motivation behind this modification would have been to provide an enhanced method of training a classification model that improves accuracy and minimizes error, since both LI and CHEN relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and CHEN relates to systems and methods for managing, training, and deploying neural network architecture for audio processing; the result of training the neural network architecture is to minimize the amount of error between a predicted output and an expected output. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and CHEN (US 20210233541 A1), Paragraph [0005, 0070]. Regarding claim 3, LI in view of CHEN teach the computer-implemented method of claim 2, LI further teaches wherein the second loss function comprises a cross entropy loss function (Fig. 2, Paragraph [0042] – LI discloses a classification loss may be defined as the cross-entropy loss.) . Regarding claim 4, LI in view of CHEN teach the computer-implemented method of claim 2, LI further teaches wherein the loss function (Fig. 2, Paragraph [0044] – LI discloses the prototypical contrastive loss, the consistency contrastive loss, the cross-entropy loss and the reconstruction loss may be combined to jointly train the neural network.) and the second loss function (Fig. 2, Paragraph [0044] – LI discloses the prototypical contrastive loss, the consistency contrastive loss, the cross-entropy loss and the reconstruction loss may be combined to jointly train the neural network.) are weighted portions of a combined loss function (Fig. 2, Paragraph [0044] – LI discloses combined loss module 280 may then compute an overall training objective, as to minimize a weighted sum of all losses: : L = L ce +ω cc L cc+ ω pc L pc_mix +ω recon L recon, Eq. (8).) . Regarding claim 5, LI in view of CHEN teach the computer-implemented method of claim 1, LI further teaches wherein the loss function (Fig. 5C, Paragraph [0044] – LI discloses combined loss module 280 may then compute an overall training objective, as to minimize a weighted sum of all losses: L = L ce +ω cc L cc+ ω pc L pc_mix +ω recon L recon, Eq. (8). Paragraph [0045] – LI further discloses the weight parameters may be set as ω cc = 1, ω recon =1, and change ω pc only across datasets. See also Paragraph [0067].) comprises a neighbor consistency regularization loss function (Fig. 2, Paragraph [0036] – LI discloses by mapping different views (augmentations) of the same image to neighboring embeddings, the consistency contrastive loss encourages the neural network 130 to learn discriminative representation that is robust to low-level image corruption [wherein the consistency contrastive loss is a neighbor consistency regularization loss function, see also Eq. (1) wherein L cc is the consistency contrastive loss function].) . Regarding claim 6, LI in view of CHEN teach the computer-implemented method of claim 5, LI further teaches wherein the neighbor consistency regularization loss function (Fig. 2, Eq. (1) wherein L cc is the consistency contrastive loss function, Paragraph [0036]) is configured to penalize a divergence of a classification of a particular embedding from a weighted combination of neighbor classifications for one or more neighboring embeddings to the particular embedding in an embedding space (Fig. 2, Paragraph [0036] – LI discloses consistency contrastive loss maximizes the inner product between the pair of positive embeddings z ^ i and z ^ j ( i ) corresponding to the same source image, while minimizing the inner product between 2(b−1) pairs of negative embeddings corresponding to different images. By mapping different views (augmentations) of the same image to neighboring embeddings, the consistency contrastive loss encourages the neural network 130 to learn discriminative representation that is robust to low-level image corruption.) . Regarding claim 7, LI in view of CHEN teach the computer-implemented method of claim 1, LI further teaches wherein the first input (Fig. 2, Paragraph [0033] – LI discloses weakly augmented data sample 241 is input to CNN 245 a. ) comprises one or more first images (Fig. 2, Paragraph [0034] – LI discloses given a minibatch of b images, weak-augmentation and strong-augmentation may be applied to each image.) , wherein the second input (Fig. 2, Paragraph [0033] – LI discloses the strongly augmented data sample 242 is input to CNN 245b.) comprises one or more second images (Fig. 2, Paragraph [0034] – LI discloses given a minibatch of b images, weak-augmentation and strong-augmentation may be applied to each image.) , and wherein the first classification and the second classification are image classifications (Fig. 1, Paragraph [0030] – LI discloses the neural network 130 learns to predict a class label of an input image with significant accuracy that the class prototype corresponding to predicted class label is the closest among all class prototypes to the embedding of the input image, and with consistency that the same predicted class label can be generated for the same source image. See also Fig. 5B, Paragraph [0065].) . Regarding claim 10, LI in view of CHEN teach the computer-implemented method of claim 1, LI further teaches wherein the first embedding comprises a first feature representation (Fig. 2, Paragraph [0035] – LI discloses the weakly-augmented data sample 241 and the strongly-augmented data sample 242 are each input to the CNN 245a or 245b, respectively, each of which encodes the augmented input into a high-dimensional feature v i , respectively. The high-dimensional features from the CNN 245a-b are then input to the autoencoders 260a-b, respectively, each of which performs z i =W ev v i the linear projection from high-dimensional features to low-dimensional embeddings.) , wherein the second embedding comprises a second feature representation (Fig. 2, Paragraph [0035] – LI discloses the weakly-augmented data sample 241 and the strongly-augmented data sample 242 are each input to the CNN 245a or 245b, respectively, each of which encodes the augmented input into a high-dimensional feature v i , respectively. The high-dimensional features from the CNN 245a-b are then input to the autoencoders 260a-b, respectively, each of which performs z i =W ev v i the linear projection from high-dimensional features to low-dimensional embeddings.) , and wherein the feature similarity is determined based at least in part on the second feature representation comprising one or more similar features to the first feature representation (Fig. 5B, Paragraph [0062] – LI discloses at step 510, a consistency contrastive loss is computed by comparing the first embedding generated from the weakly augmented sample 241 and the second embedding generated from the strongly augmented sample 242, e.g., according to Eq. (1).) . Regarding claim 16, LI in view of CHEN teach the method of claim 1, LI fails to explicitly teach further comprising: obtaining a first input label; and wherein evaluating the loss function comprises evaluating a difference between the first classification and the first input label, and wherein the one or more parameters are adjusted based at least in part on the first input label. However, CHEN explicitly teaches further comprising: obtaining a first input label (Fig. 1, Paragraph [0047] – CHEN discloses the analytics server 102 employs supervised training to train the neural network, where the analytics database 104 includes labels associated with the training audio signals that indicate which signals contain speech portions. See also Paragraph [0092].) ; and wherein evaluating the loss function (Fig. 6, Paragraph [0092] – CHEN discloses loss layer 610 tunes the embedding extractor 606 by performing the loss function (e.g., LMCL).) comprises evaluating a difference between the first classification and the first input label (Fig. 6, Paragraph [0092] – CHEN discloses loss layer 610 tunes the embedding extractor 606 by performing the loss function (e.g., LMCL) to determine the distance (e.g., large margin cosine loss) between the determined genuine and spoof classifications, as indicated by supervised labels or previously generated clusters.) , and wherein the one or more parameters are adjusted based at least in part on the first input label (Fig. 6, Paragraph [0071] – CHEN discloses the server performs a loss function (e.g., LMCL, LDA) and updates hyper-parameters (or other types of weight values) of the neural network architecture. The server determines the error between the predicted output and the expected output by comparing the similarity or difference between the predicted output and expected output. See also Paragraph [0092].) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI in view of CHEN of having a computer-implemented method for training a classification model, the method comprising: determining, by the computing system, a similarity measure between the first embedding and the second embedding based on a feature similarity; evaluating, by a computing system, a loss function, wherein the loss function comprises a loss term that evaluates a difference between the first classification and the second classification weighted by the similarity measure, with the teachings of CHEN having further comprising: obtaining a first input label; and wherein evaluating the loss function comprises evaluating a difference between the first classification and the first input label, and wherein the one or more parameters are adjusted based at least in part on the first input label. Wherein LI’s computer implemented method for training a classification model further comprising: obtaining a first input label; and wherein evaluating the loss function comprises evaluating a difference between the first classification and the first input label, and wherein the one or more parameters are adjusted based at least in part on the first input label. The motivation behind this modification would have been to provide an enhanced method of training a classification model that improves accuracy and minimizes error, since both LI and CHEN relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and CHEN relates to systems and methods for managing, training, and deploying neural network architecture for audio processing; the result of training the neural network architecture is to minimize the amount of error between a predicted output and an expected output. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and CHEN (US 20210233541 A1), Paragraph [0005, 0070]. Regarding claim 17, LI teaches a computer-implemented method of classifying an input (Figs. 5A-5C, Paragraph [0057] – LI discloses FIGS. 5A-5C provides an example flow diagram illustrating a method of noise-robust contrastive learning. Paragraph [0065] – LI discloses a classification loss may be computed using the class predictions, e.g., generated by the classifier 255.) with a classification model (Fig. 2, #255 called classifier, Paragraph [0065]) , comprising: obtaining input data (Fig. 2, Paragraph [0031] – LI discloses diagram 200 shows that an input data sample 240, similar to one of the images 110a-n with a noisy label shown in FIG. 1, is used for noise-robust contrastive learning.) ; processing the input data with an encoder model (Fig. 2, #245a-c called a deep encoder CNN, Paragraph [0033] – LI discloses a deep encoder, e.g., a convolutional neural network (CNN) 245a-c (or collectively referred to as 245) that encodes an image x i or an augmented sample of image x i to a high-dimensional feature v i ) to generate an input embedding (Fig. 2, Paragraph [0035] – LI discloses the weakly-augmented data sample 241 and the strongly-augmented data sample 242 are each input to the CNN 245a or 245b, respectively, each of which encodes the augmented input into a high- dimensional feature v i , respectively. The high-dimensional features from the CNN 245a-b are then input to the autoencoders 260a-b, respectively, each of which performs z i =W ev v i the linear projection from high-dimensional features to low-dimensional embeddings.) , wherein the input embedding comprises an embedding in an embedding space (Fig. 2, Paragraph [0035] – LI discloses high-dimensional features from the CNN 245a-b are then input to the autoencoders 260a-b, respectively, each of which performs z i =W ev v i the linear projection from high-dimensional features to low-dimensional embeddings.) ; processing the input embedding with a classification model (Fig. 2, #255 called classifier, Paragraph [0033] – LI discloses a classifier 255 (e.g., a fully-connected layer followed by softmax) that receives v i as input and outputs class predictions;) to generate an output classification (Fig. 2, Paragraph [0042] – LI discloses the output high-dimensional feature v i , are passed to the classifier 255. Given the softmax output from the classifier 255, p(y; x i ), a classification loss may be defined as the cross-entropy loss. The cross-entropy loss L ce may be used to update the classifier 255, e.g., via the backpropagation path shown in dashed line from block 261. See also Paragraph [0033].) , wherein the classification model (Fig. 2, #255 called classifier, Paragraph [0065]) was trained by: obtaining a training dataset (Fig. 5A, Paragraph [0058] – LI discloses method 500 starts with step 502, at which a training set of data samples may be obtained, each data sample having a noisy label. For example, the training set of data samples may be received as part of input 340 via the data interface 315 in FIG. 3. See also Paragraph [0046].) , wherein the training dataset comprises a first input (Fig. 5A, Paragraph [0060] – LI discloses at step 505, for each data sample, a weak augmented data sample (e.g., 241) and a strong augmented data sample may be generated [wherein weak augmented data sample is a first input].) and a second input (Fig. 5A, Paragraph [0060] – LI discloses at step 505, for each data sample, a weak augmented data sample (e.g., 241) and a strong augmented data sample may be generated [wherein strong augmented data sample is a second input].) ; processing the first input (Fig. 2, #241 called a weak augmented data sample, Paragraph [0060]) and the second input (Fig. 2, #242 called a strongly augmented data sample, Paragraph [0060]) with an encoder model (Fig. 2, #245a-c called a deep encoder CNN, Paragraph [0033] – LI discloses a deep encoder, e.g., a convolutional neural network (CNN) 245a-c (or collectively referred to as 245) that encodes an image x i or an augmented sample of image x i to a high-dimensional feature v i ;) to generate a first embedding and a second embedding (Fig. 2, Paragraph [0035] – LI discloses the weakly-augmented data sample 241 and the strongly-augmented data sample 242 are each input to the CNN 245a or 245b, respectively, each of which encodes the augmented input into a high-dimensional feature v i , respectively [wherein high-dimensional feature v i is a first and second embedding]. The high-dimensional features from the CNN 245a-b are then input to the autoencoders 260a-b, respectively, each of which performs z i =W ev v i the linear projection from high-dimensional features to low-dimensional embeddings.) ; processing the first embedding and a second embedding with a classification model (Fig. 2, #255 called classifier, Paragraph [0033] – LI discloses a classifier 255 (e.g., a fully-connected layer followed by softmax) that receives v i as input and outputs class predictions.) to generate a first classification and a second classification (Fig. 2, Paragraph [0042] – LI discloses the output high-dimensional feature v i , are passed to the classifier 255. Given the softmax output from the classifier 255, p(y; x i ), a classification loss may be defined as the cross-entropy loss. The cross-entropy loss L ce may be used to update the classifier 255, e.g., via the backpropagation path shown in dashed line from block 261. See also Paragraph [0033].) ; determining a similarity measure between the first embedding and the second embedding based on a feature similarity (Fig. 5B, Paragraph [0062] – LI discloses at step 510, a consistency contrastive loss [wherein a consistency contrastive loss is a similarity measure] is computed by comparing the first embedding generated from the weakly augmented sample 241 and the second embedding generated from the strongly augmented sample 242, e.g., according to Eq. (1).) ; evaluating a loss function (Fig. 5C, Paragraph [0067] – LI discloses step 520, where a combined loss is computed by adding the weighted sum of all the losses across the batch of samples, e.g., according to Eq. (8).) , wherein the loss function (Fig. 5C, Paragraph [0044] – LI discloses combined loss module 280 may then compute an overall training objective, as to minimize a weighted sum of all losses: L = L ce +ω cc L cc+ ω pc L pc_mix +ω recon L recon, Eq. (8). Paragraph [0045] – LI further discloses the weight parameters may be set as ω cc = 1, ω recon =1, and change ω pc only across datasets. See also Paragraph [0067].) comprises a loss term that evaluates a difference between the first classification and the second classification (Fig. 5B, Paragraph [0065] – LI discloses at step 516, a classification loss may be computed using the class predictions, e.g., generated by the classifier 255. The computation may be done according to Eq. (7). Paragraph [0042] – LI discloses a classification loss may be defined as the cross-entropy loss [wherein ℒ ce is the cross-entropy loss function, see Eq. (7)].) weighted by the similarity measure (Fig. 5C, Paragraph [0036] – LI discloses the consistency contrastive loss can be computed at 265 by comparing the normalized embeddings from autoencoders 260a-b [wherein ℒ cc is the consistency contrastive loss function that is a similarity measure, see Eq. (1).] Paragraph [0044] – LI further discloses a combined loss is computed by adding the weighted sum of all the losses across the batch of samples, e.g., according to Eq. (8).) ; Although LI further teaches and providing the output classification for the input data (Fig. 3, Paragraph [0049] – LI discloses noise-robust contrastive learning module 330 may generate an output 350, e.g., such as a class label corresponding to the input image.) . LI fails to explicitly teach and adjusting one or more parameters of the classification model based at least in part on the loss function; However, CHEN explicitly teaches and adjusting one or more parameters of the classification model based at least in part on the loss function ((Fig. 3, Paragraph [0071] – CHEN discloses the server performs a loss function (e.g., LMCL, LDA) and updates hyper-parameters (or other types of weight values) of the neural network architecture.) ; Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI of having computer-implemented method of classifying an input with a classification model, comprising: obtaining input data; processing the input data with an encoder model to generate an input embedding, wherein the input embedding comprises an embedding in an embedding space; processing the input embedding with a classification model to generate an output classification, wherein the classification model was trained by: obtaining a training dataset, wherein the training dataset comprises a first input and a second input; processing the first input and the second input with an encoder model to generate a first embedding and a second embedding; processing the first embedding and a second embedding with a classification model to generate a first classification and a second classification, with the teachings of CHEN having and adjusting one or more parameters of the classification model based at least in part on the loss function. Wherein LI’s computer-implemented method of classifying an input with a classification model wherein having and adjusting one or more parameters of the classification model based at least in part on the loss function. The motivation behind this modification would have been to provide an enhanced method of classifying an input with a classification model that improves accuracy and minimizes error, since both LI and CHEN relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and CHEN relates to systems and methods for managing, training, and deploying neural network architecture for audio processing; the result of training the neural network architecture is to minimize the amount of error between a predicted output and an expected output. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and CHEN (US 20210233541 A1), Paragraph [0005, 0070]. Regarding claim 18, LI in view of CHEN teach the method of claim 17, LI further teaches wherein the input data comprises image data (Fig. 2, Paragraph [0031] – LI discloses diagram 200 shows that an input data sample 240, similar to one of the images 110a-n with a noisy label shown in FIG. 1, is used for noise-robust contrastive learning.) , and wherein the output classification comprises one or more object classifications based on one or more features in the image data (Fig. 1, Paragraph [0030] – LI discloses the neural network 130 learns to predict a class label of an input image with significant accuracy that the class prototype corresponding to predicted class label is the closest among all class prototypes to the embedding of the input image, and with consistency that the same predicted class label can be generated for the same source image. See also Paragraph [0042].) . Regarding claim 19, LI in view of CHEN teach the method of claim 17, LI fails to explicitly teach wherein the output classification comprises a prediction score descriptive of a level of certainty for one or more possible classifications. However, CHEN explicitly teaches wherein the output classification comprises a prediction score descriptive of a level of certainty for one or more possible classifications (Fig. 6, Paragraph [0093] – CHEN discloses classifier 612 includes one or more layers trained to determine the whether the outputs (e.g., classifications, feature vectors, embeddings) of the embedding extractor 606 and/or fully-connected layers 608 are within a given distance from a threshold value established during the training phase according to the LMCL and/or LDA algorithms. By executing the classifier 612, the server classifies an inbound audio signal as genuine or spoofed based on the neural network architecture's 600 output(s).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI in view of CHEN of having computer-implemented method of classifying an input with a classification model, comprising: obtaining input data; processing the input data with an encoder model to generate an input embedding, wherein the input embedding comprises an embedding in an embedding space; processing the input embedding with a classification model to generate an output classification, and providing the output classification for the input data, with the teachings of CHEN having wherein the output classification comprises a prediction score descriptive of a level of certainty for one or more possible classifications. Wherein LI’s computer-implemented method of classifying an input with a classification model wherein the output classification comprises a prediction score descriptive of a level of certainty for one or more possible classifications. The motivation behind this modification would have been to provide an enhanced method of classifying an input with a classification model that improves accuracy and minimizes error, since both LI and CHEN relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and CHEN relates to systems and methods for managing, training, and deploying neural network architecture for audio processing; the result of training the neural network architecture is to minimize the amount of error between a predicted output and an expected output. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and CHEN (US 20210233541 A1), Paragraph [0005, 0070]. Regarding claim 21, LI teaches a computing system (Fig. 3, #300 called computing device, Paragraph [0046] – LI discloses FIG. 3 is a simplified diagram of a computing device) , the computing system comprising: one or more processors (Fig. 3, #310 called processor, Paragraph [0046] – LI discloses computing device 300 includes a processor 310 coupled to memory 320.) ; and one or more non-transitory computer-readable media (Fig. 3, #320 called memory, Paragraph [0049] – LI discloses memory 320 may include non-transitory, tangible, machine readable media) that collectively store instructions that, when executed by the one or more processors, cause the computing system to perform operations (Fig. 3, Paragraph [0049] – LI discloses memory 320 may include non-transitory, tangible, machine readable media that includes executable code that when run by one or more processors (e.g., processor 310) may cause the one or more processors to perform the methods described in further detail herein.) , the operations comprising: obtaining a training dataset (Fig. 5A, Paragraph [0058] – LI discloses method 500 starts with step 502, at which a training set of data samples may be obtained, each data sample having a noisy label. For example, the training set of data samples may be received as part of input 340 via the data interface 315 in FIG. 3. See also Paragraph [0046].) , wherein the training dataset comprises a first input (Fig. 5A, Paragraph [0060] – LI discloses at step 505, for each data sample, a weak augmented data sample (e.g., 241) and a strong augmented data sample may be generated [wherein weak augmented data sample is a first input].) and a second input (Fig. 5A, Paragraph [0060] – LI discloses at step 505, for each data sample, a weak augmented data sample (e.g., 241) and a strong augmented data sample may be generated [wherein strong augmented data sample is a second input].) ; processing the first input (Fig. 2, #241 called a weak augmented data sample, Paragraph [0060]) with an encoder model (Fig. 2, #245a-c called a deep encoder CNN, Paragraph [0033] – LI discloses a deep encoder, e.g., a convolutional neural network (CNN) 245a-c (or collectively referred to as 245) that encodes an image x i or an augmented sample of image x i to a high-dimensional feature v i ;) to generate a first embedding (Fig. 2, Paragraph [0035] – LI discloses the weakly-augmented data sample 241 and the strongly-augmented data sample 242 are each input to the CNN 245a or 245b, respectively, each of which encodes the augmented input into a high-dimensional feature v i , respectively [wherein high-dimensional feature v i is a first embedding]. The high-dimensional features from the CNN 245a-b are then input to the autoencoders 260a-b, respectively, each of which performs z i =W ev v i the linear projection from high-dimensional features to low-dimensional embeddings.) ; processing the first embedding with a classification model (Fig. 2, #255 called classifier, Paragraph [0033] – LI discloses a classifier 255 (e.g., a fully-connected layer followed by softmax) that receives v i as input and outputs class predictions;) to generate a first classification (Fig. 2, Paragraph [0042] – LI discloses the output high-dimensional feature v i , are passed to the classifier 255. Given the softmax output from the classifier 255, p(y; x i ), a classification loss may be defined as the cross-entropy loss. The cross-entropy loss L ce may be used to update the classifier 255, e.g., via the backpropagation path shown in dashed line from block 261. See also Paragraph [0033].) ; processing the second input (Fig. 2, #242 called a strongly augmented data sample, Paragraph [0060]) with the encoder model (Fig. 2, #245a-c called a deep encoder CNN, Paragraph [0033] – LI discloses a deep encoder, e.g., a convolutional neural network (CNN) 245a-c (or collectively referred to as 245) that encodes an image x i or an augmented sample of image x i to a high-dimensional feature v i ;) to generate a second embedding (Fig. 2, Paragraph [0035] – LI discloses the weakly-augmented data sample 241 and the strongly-augmented data sample 242 are each input to the CNN 245a or 245b, respectively, each of which encodes the augmented input into a high-dimensional feature v i , respectively [wherein high-dimensional feature v i is a second embedding]. The high-dimensional features from the CNN 245a-b are then input to the autoencoders 260a-b, respectively, each of which performs z i =W ev v i the linear projection from high-dimensional features to low-dimensional embeddings.) ; processing the second embedding with the classification model the second embedding with the classification model (Fig. 2, #255 called classifier, Paragraph [0033] – LI discloses a classifier 255 (e.g., a fully-connected layer followed by softmax) that receives v i as input and outputs class predictions;) to generate a second classification (Fig. 2, Paragraph [0042] – LI discloses the output high-dimensional feature v i , are passed to the classifier 255. Given the softmax output from the classifier 255, p(y; x i ), a classification loss may be defined as the cross-entropy loss. The cross-entropy loss L ce may be used to update the classifier 255, e.g., via the backpropagation path shown in dashed line from block 261. See also Paragraph [0033].) ; determining a similarity measure between the first embedding and the second embedding based on a feature similarity (Fig. 5B, Paragraph [0062] – LI discloses at step 510, a consistency contrastive loss [wherein a consistency contrastive loss is a similarity measure] is computed by comparing the first embedding generated from the weakly augmented sample 241 and the second embedding generated from the strongly augmented sample 242, e.g., according to Eq. (1).) ; evaluating a loss function (Fig. 5C, Paragraph [0067] – LI discloses step 520, where a combined loss is computed by adding the weighted sum of all the losses across the batch of samples, e.g., according to Eq. (8).) , wherein the loss function (Fig. 5C, Paragraph [0044] – LI discloses combined loss module 280 may then compute an overall training objective, as to minimize a weighted sum of all losses: L = L ce +ω cc L cc+ ω pc L pc_mix +ω recon L recon, Eq. (8). Paragraph [0045] – LI further discloses the weight parameters may be set as ω cc = 1, ω recon =1, and change ω pc only across datasets. See also Paragraph [0067].) comprises a loss term that evaluates a difference between the first classification and the second classification (Fig. 5B, Paragraph [0065] – LI discloses at step 516, a classification loss may be computed using the class predictions, e.g., generated by the classifier 255. The computation may be done according to Eq. (7). Paragraph [0042] – LI discloses a classification loss may be defined as the cross-entropy loss [wherein ℒ ce is the cross-entropy loss function, see Eq. (7)].) Although LI explicitly teaches weighted by the similarity measure (Fig. 5C, Paragraph [0036] – LI discloses the consistency contrastive loss can be computed at 265 by comparing the normalized embeddings from autoencoders 260a-b [wherein ℒ cc is the consistency contrastive loss function that is a similarity measure, see Eq. (1).] Paragraph [0044] – LI further discloses a combined loss is computed by adding the weighted sum of all the losses across the batch of samples, e.g., according to Eq. (8).) ; LI fails to explicitly teach and adjusting one or more parameters of the classification model based at least in part on the loss function. However, CHEN explicitly teaches and adjusting one or more parameters of the classification model based at least in part on the loss function (Fig. 3, Paragraph [0071] – CHEN discloses the server performs a loss function (e.g., LMCL, LDA) and updates hyper-parameters (or other types of weight values) of the neural network architecture.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI of having a computing system, the computing system comprising: one or more processors; and one or more non-transitory computer-readable media that collectively store instructions that, when executed by the one or more processors, cause the computing system to perform operations, the operations comprising: obtaining a training dataset, wherein the training dataset comprises a first input and a second input; processing the first input with an encoder model to generate a first embedding; processing the first embedding with a classification model to generate a first classification, with the teachings of CHEN of having and adjusting one or more parameters of the classification model based at least in part on the loss function. Wherein LI’s computing system wherein having and adjusting one or more parameters of the classification model based at least in part on the loss function. The motivation behind this modification would have been to provide an enhanced system for training a classification model that improves accuracy and minimizes error, since both LI and CHEN relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and CHEN relates to systems and methods for managing, training, and deploying neural network architecture for audio processing; the result of training the neural network architecture is to minimize the amount of error between a predicted output and an expected output. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and CHEN (US 20210233541 A1), Paragraph [0005, 0070] . 07-21-aia AIA Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over LI (US 20210374553 A1), hereinafter referenced as LI in view of CHEN (US 20210233541 A1), hereinafter referenced as CHEN, further in view of ISO-SIPILA (US 20220188520 A1), hereinafter referenced as ISO-SIPILA . Regarding claim 8, LI in view of CHEN teach the computer-implemented method of claim 1, LI in view of CHEN fail to explicitly teach wherein the classification model and the encoder model are jointly trained. However, ISO-SIPILA explicitly teaches wherein the classification model (Fig. 3C, Paragraph [0138] – ISO-SIPILA discloses one or more classification model(s) 326a to 326n.) and the encoder model (Fig. 3C, Paragraph [0138] – ISO-SIPILA discloses one or more embedding/encoding models 324a to 324b.) are jointly trained (Fig. 3C, Paragraph [0138] – ISO-SIPILA discloses model parameters for the NER models 324, 325, and 326 may be jointly trained based on labelled training dataset(s) associated with entities and/or entity types as described with reference to FIGS. 1a to 3b.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI in view of CHEN of having a computer-implemented method for training a classification model, the method comprising: processing, by the computing system, the first input with an encoder model to generate a first embedding; processing, by the computing system, the first embedding with a classification model to generate a first classification; processing, by the computing system, the second input with the encoder model to generate a second embedding; processing, by the computing system, the second embedding with the classification model to generate a second classification; with the teachings of ISO-SIPILA having wherein the classification model and the encoder model are jointly trained. Wherein LI’s computer implemented method for training a classification model wherein the classification model and the encoder model are jointly trained. The motivation behind this modification would have been to provide an enhanced method of training a classification model that improves accuracy, since both LI and ISO-SIPILA relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and ISO-SIPILA relates to a system and method for performing name entity recognition with deep learning on large scale datasets, providing a more robust NER system that can ensure both up-to-date, and accurate dictionaries for more efficient entity recognition. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and ISO-SIPILA (US 20220188520 A1), Paragraph [0089] . 07-21-aia AIA Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over LI (US 20210374553 A1), hereinafter referenced as LI in view of CHEN (US 20210233541 A1), hereinafter referenced as CHEN, further in view of LAI (US 20200244621 A1), hereinafter referenced as LAI . Regarding claim 9, LI in view of CHEN teach the computer-implemented method of claim 1, LI in view of CHEN fail to explicitly teach wherein the encoder model is a newly initialized model. However, LAI explicitly teaches wherein the encoder model (Fig. 5, #150 called encoder, Paragraph [0095]) is a newly initialized model (Fig. 5, Paragraph [0095] – LAI discloses the parameters in the learning algorithm 140, including the encoder 150 and decoder 160, may be randomly initialized at the start of the training process.). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI in view of CHEN of having a computer-implemented method for training a classification model, the method comprising: processing, by the computing system, the first input with an encoder model to generate a first embedding; processing, by the computing system, the first embedding with a classification model to generate a first classification; processing, by the computing system, the second input with the encoder model to generate a second embedding; processing, by the computing system, the second embedding with the classification model to generate a second classification; with the teachings of LAI having wherein the encoder model is a newly initialized model. Wherein LI’s computer implemented method for training a classification model wherein the encoder model is a newly initialized model. The motivation behind this modification would have been to provide an enhanced method of training a machine learning model that improves both accuracy and the quality of training data, since both LI and LAI relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and LAI relates to the field of creating training data, using domain name searches, selected suggested domain names and/or registered domain names, to train a learning algorithm to suggest domain names to users; learning algorithm 140 may also filter out pairs where two domains are not relevant to each other to improve the quality of the training data. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and LAI (US 20200244621 A1), Paragraph [0092] . 07-21-aia AIA Claim s 11-13 are rejected under 35 U.S.C. 103 as being unpatentable over LI (US 20210374553 A1), hereinafter referenced as LI in view of CHEN (US 20210233541 A1), hereinafter referenced as CHEN, further in view of OTT (US 20190180171 A1), hereinafter referenced as OTT . Regarding claim 11, LI in view of CHEN teach the computer-implemented method of claim 1, LI further teaches wherein the second input comprises a minibatch comprising a plurality of training inputs (Fig. 2, Paragraph [0034] – LI discloses given a minibatch of b images, weak-augmentation and strong-augmentation may be applied to each image.) ; and wherein generating the second embedding comprises: processing the minibatch with the encoder model to generate a plurality of embeddings for the training inputs of the minibatch (Fig. 2, Paragraph [0035] – LI discloses the weakly-augmented data sample 241 and the strongly-augmented data sample 242 are each input to the CNN 245a or 245b, respectively, each of which encodes the augmented input into a high-dimensional feature v i , respectively. The high-dimensional features from the CNN 245a-b are then input to the autoencoders 260a-b, respectively, each of which performs z i =W ev v i the linear projection from high-dimensional features to low-dimensional embeddings.) ; Although LI explicitly teaches processing the plurality of embeddings for the training inputs of the minibatch with the classification model (Fig. 2, #255 called classifier, Paragraph [0033]) to generate a plurality of classifications for the training inputs of the minibatch (Fig. 2, #255 called classifier, Paragraph [0033] – LI discloses a classifier 255 (e.g., a fully-connected layer followed by softmax) that receives v i as input and outputs class predictions.) ; LI in view of CHEN fails to explicitly teach determining one or more particular ones of the plurality of embeddings for the training inputs of the minibatch associated with the first embedding; and determining the second embedding based on the one or more particular ones of the plurality of embeddings. However, OTT explicitly teaches determining one or more particular ones of the plurality of embeddings for the training inputs of the minibatch associated with the first embedding (Fig. 6, Paragraph [0047] – OTT discloses the social-networking system 1160 may access a plurality of second embeddings representing a plurality of predicted second place-entities, respectively, each second place-entity corresponding to a second geographic location, wherein each second embedding is a point in the d-dimensional embedding space. At step 640, the social-networking system 1160 may calculate, for each of the second place-entities, a similarity metric between the embedding representing the first place-entity [wherein the embedding representing the first place-entity is a first embedding] and the embedding representing the second place-entity, wherein the similarity metric corresponds to a probability that the first user will be located at the second geographic location corresponding to the second place entity within a particular timeframe of the first user being located at the first geographic location at the first time. At step 650, the social-networking system 1160 may rank each of the second place-entities based on their calculated similarity metrics.) ; and determining the second embedding based on the one or more particular ones of the plurality of embeddings (Fig. 6, Paragraph [0047] – OTT discloses the social-networking system 1160 may rank each of the second place-entities based on their calculated similarity metrics. At step 660, the social-networking system 1160 may send, to the client system 1130, information associated with one or more second geographic locations corresponding to one or more second place-entities having a ranking greater than a threshold ranking.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI in view of CHEN of having a computer-implemented method for training a classification model, the method comprising: processing, by the computing system, the first input with an encoder model to generate a first embedding; processing, by the computing system, the first embedding with a classification model to generate a first classification; processing, by the computing system, the second input with the encoder model to generate a second embedding; processing, by the computing system, the second embedding with the classification model to generate a second classification; with the teachings of OTT having determining one or more particular ones of the plurality of embeddings for the training inputs of the minibatch associated with the first embedding; and determining the second embedding based on the one or more particular ones of the plurality of embeddings. Wherein LI’s computer implemented method for training a classification model wherein having determining one or more particular ones of the plurality of embeddings for the training inputs of the minibatch associated with the first embedding; and determining the second embedding based on the one or more particular ones of the plurality of embeddings. The motivation behind this modification would have been to provide an enhanced method of training a classification model that improves accuracy and computing processes related to receiving and executing queries, since both LI and OTT relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and OTT discloses a model trained by machine learning, which may take an input embedding representing a first place-entity corresponding to a first geographic location and output a predicted second place-entity corresponding to a second geographic location that the user will subsequently visit; automatically predicting a second geographic location that a user will visit subsequent to the user's presence at a first geographic location and sending the user information associated with the second geographic location may improve computing processes related to receiving and executing queries by reducing the need for the user to send queries related to the second geographic location, which may reduce the computing resources needed for such processes. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and OTT (US 20190180171 A1), Paragraph [0005, 0034]. Regarding claim 12, LI and CHEN in view of OTT teach the method of claim 11, LI and CHEN fail to explicitly teach wherein determining the one or more particular embeddings for the training inputs of the minibatch associated with the first embedding comprises determining a cosine similarity between the first embedding and each of the plurality of embeddings for the training inputs of the minibatch. However, OTT explicitly teaches wherein determining the one or more particular embeddings for the training inputs of the minibatch associated with the first embedding (Fig. 6, Paragraph [0047] – OTT discloses the social-networking system 1160 may rank each of the second place-entities based on their calculated similarity metrics. At step 660, the social-networking system 1160 may send, to the client system 1130, information associated with one or more second geographic locations corresponding to one or more second place-entities having a ranking greater than a threshold ranking.) comprises determining a cosine similarity between the first embedding and each of the plurality of embeddings for the training inputs of the minibatch (Fig. 6, Paragraph [0045] – OTT discloses an embedding a ⃑ may represent first place-entity A, and second embeddings b ⃑ and c ⃑ may represent second place-entities B and C, respectively. A cosine similarity between a ⃑ and b ⃑ may be 0.24 and a cosine similarity between a ⃑ and c ⃑ may be 0.86. This may indicate that there is a higher probability that the first user will be located at a geographic location corresponding to C than the probability that the user will be located at a geographic locations corresponding to B within the particular timeframe.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI in view of CHEN of having a computer-implemented method for training a classification model, the method comprising: processing, by the computing system, the first input with an encoder model to generate a first embedding; processing, by the computing system, the first embedding with a classification model to generate a first classification; processing, by the computing system, the second input with the encoder model to generate a second embedding; processing, by the computing system, the second embedding with the classification model to generate a second classification; with the teachings of OTT having wherein determining the one or more particular embeddings for the training inputs of the minibatch associated with the first embedding comprises determining a cosine similarity between the first embedding and each of the plurality of embeddings for the training inputs of the minibatch. Wherein LI’s computer implemented method for training a classification model wherein determining the one or more particular embeddings for the training inputs of the minibatch associated with the first embedding comprises determining a cosine similarity between the first embedding and each of the plurality of embeddings for the training inputs of the minibatch. The motivation behind this modification would have been to provide an enhanced method of training a classification model that improves accuracy and computing processes related to receiving and executing queries, since both LI and OTT relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and OTT discloses a model trained by machine learning, which may take an input embedding representing a first place-entity corresponding to a first geographic location and output a predicted second place-entity corresponding to a second geographic location that the user will subsequently visit; automatically predicting a second geographic location that a user will visit subsequent to the user's presence at a first geographic location and sending the user information associated with the second geographic location may improve computing processes related to receiving and executing queries by reducing the need for the user to send queries related to the second geographic location, which may reduce the computing resources needed for such processes. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and OTT (US 20190180171 A1), Paragraph [0005, 0034]. Regarding claim 13, LI and CHEN in view of OTT teach the method of claim 11, LI fails to explicitly teach wherein the minibatch comprises randomly selected training inputs from a training input database. However, CHEN explicitly teaches wherein the minibatch (Fig. 1, Paragraph [0047] – CHEN discloses analytics database 104 and/or the call center database 112 may contain any number of corpora of training audio signals [wherein training audio signals are the minibatch] that are accessible to the analytics server 102 via one or more networks.) comprises randomly selected training inputs from a training input database (Fig. 1, Paragraph [0047] – CHEN discloses an administrator may configure the analytics server 102 to select the speech segments to have durations that are random, random within configured limits, or predetermined at the admin device 103. Paragraph [0055] – CHEN further discloses the server may select training audio signals and/or randomly generate simulated audio segments.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI and CHEN in view of OTT of having a computer-implemented method for training a classification model, the method comprising: determining, by the computing system, a similarity measure between the first embedding and the second embedding based on a feature similarity; evaluating, by a computing system, a loss function, wherein the loss function comprises a loss term that evaluates a difference between the first classification and the second classification weighted by the similarity measure; with the teachings of CHEN having further comprising: wherein the minibatch comprises randomly selected training inputs from a training input database. Wherein LI’s computer implemented method for training a classification model wherein the minibatch comprises randomly selected training inputs from a training input database. The motivation behind this modification would have been to provide an enhanced method of training a classification model that improves accuracy and minimizes error, since both LI and CHEN relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and CHEN relates to systems and methods for managing, training, and deploying neural network architecture for audio processing; the result of training the neural network architecture is to minimize the amount of error between a predicted output and an expected output. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and CHEN (US 20210233541 A1), Paragraph [0005, 0070] . 07-21-aia AIA Claim 14 is rejected under 35 U.S.C. 103 as being unpatentable over LI (US 20210374553 A1), hereinafter referenced as LI in view of CHEN (US 20210233541 A1), hereinafter referenced as CHEN, further in view of OTT (US 20190180171 A1), hereinafter referenced as OTT, further in view of TARIQ (US 20190392268 A1), hereinafter referenced as TARIQ . Regarding claim 14, LI and CHEN in view of OTT teach the method of any of claim 11, LI and CHEN in view of OTT fail to explicitly teach wherein the minibatch comprises a balanced training data set, wherein the minibatch comprises an equal amount of training inputs for each of a plurality of predetermined classifications. However, TARIQ explicitly teaches wherein the minibatch comprises a balanced training data set (Fig. 6A, Paragraph [0087] – TARIQ discloses the first batch of images and/or the second batch of images may be data balanced to ensure that the ML model is being trained to accurately detect objects of different types.) , wherein the minibatch comprises an equal amount of training inputs for each of a plurality of predetermined classifications (Fig. 6A, Paragraph [0087] – TARIQ discloses in some instances, the first batch of images and/or the second batch of images may include a first predefined number of images that are associated with a first object classification (e.g., each of the first predefined number of images include at least one pedestrian) and a second predefined number of images that are associated with a second object classification (e.g., each of the second predefined number of images include at least one biker), though any number of classes and relative weights are contemplated.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI and CHEN in view of OTT of having a computer-implemented method for training a classification model, the method comprising: determining, by the computing system, a similarity measure between the first embedding and the second embedding based on a feature similarity; evaluating, by a computing system, a loss function, wherein the loss function comprises a loss term that evaluates a difference between the first classification and the second classification weighted by the similarity measure; with the teachings of TARIQ having wherein the minibatch comprises a balanced training data set, wherein the minibatch comprises an equal amount of training inputs for each of a plurality of predetermined classifications. Wherein LI’s computer implemented method for training a classification model wherein the minibatch comprises a balanced training data set, wherein the minibatch comprises an equal amount of training inputs for each of a plurality of predetermined classifications. The motivation behind this modification would have been to provide an enhanced method of training a classification model that improves computer vision and increases accuracy, since both LI and TARIQ relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and TARIQ relates to techniques for training a machine learning (ML) model; techniques discussed herein improve computer vision by increasing the accuracy of object detection and decreasing the compute time for obtaining accurate object identifications so that objects may be detected in real time for use in applications such as autonomous vehicle control. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and TARIQ (US 20190392268 A1), Paragraph [0021] . 07-21-aia AIA Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over LI (US 20210374553 A1), hereinafter referenced as LI in view of CHEN (US 20210233541 A1), hereinafter referenced as CHEN, further in view of HUH (US 20200034693 A1), hereinafter referenced as HUH . Regarding claim 15, LI in view of CHEN teach the method of claim 1, LI in view of CHEN fail to explicitly teach wherein the loss function comprises a bootstrapping loss function. However, HUH explicitly teaches wherein the loss function comprises a bootstrapping loss function (Fig. 3, Paragraph [0055] – HUH discloses the label-noise CNN model 310 may be, but is not limited to, a model employing a bootstrapping technique using a bootstrapping loss function as a loss function.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date the claimed invention was made to combine the teachings of LI in view of CHEN of having a computer-implemented method for training a classification model, the method comprising: determining, by the computing system, a similarity measure between the first embedding and the second embedding based on a feature similarity; evaluating, by a computing system, a loss function, wherein the loss function comprises a loss term that evaluates a difference between the first classification and the second classification weighted by the similarity measure; with the teachings of HUH having wherein the loss function comprises a bootstrapping loss function. Wherein LI’s computer implemented method for training a classification model wherein the loss function comprises a bootstrapping loss function. The motivation behind this modification would have been to provide an enhanced method of training a classification model that improves accuracy and automatically classifies bad data, since both LI and HUH relate to managing and training neural network architectures, wherein LI relates to training and use of machine learning systems and more specifically noise-robust contrastive learning with data samples having noisy labels; the neural network learns to predict a class label of an input image with significant accuracy ; the proposed method also improves the computational efficiency compared to many existing methods, and HUH relates to a method for detecting defects in a semiconductor device even for non-labeled image data or improperly labeled image data by using a convolutional neural network (CNN); defect detection system 1 has advantages in that it can automatically create and correct labels for the non-labeled image data and the noise-labelled image data so that defects during the process of fabricating a semiconductor device can be detected and classified without manually adding or correcting labels. Please see LI (US 20210374553 A1), Paragraph [0030, 0091], and HUH (US 20200034693 A1), Paragraph [0006, 0038]. Conclusion Listed below are the prior arts made of record and not relied upon but are considered pertinent to applicant’s disclosure. BOULT et al. (US 20200394557 A1) - The invention includes systems and methods, including computer programs encoded on computer storage media, for classifying inputs as belonging to a known or unknown class as well as for updating the system to improve is performance. In one system, there is a desired feature representation for unknown inputs, e.g., a zero vector, and the system includes transforming input data to produce a feature representation, using that to compute dissimilarity with the desired feature representation for unknown inputs and combining dissimilarity with other transformations of the feature representation to determine if the input is from a specific known class or if it is unknown. In one embodiment, the system transforms the magnitude of the feature representation into a confidence score. In an update method to improve performance, the system transforms inputs into feature representations which go through a scoring means and then use a robust loss function, which has different loss terms for known and unknown inputs which are then used to update the system weights to improve performance...… Fig. 1, Abstract. SHMIGELSKY et al. (US 20230337636 A1) - An animal management system has one or more imaging devices, and a computing device coupled to the one or more image devices for receiving one or more images captured by the one or more imaging devices, processing at least one image using an artificial intelligence (AI) pipeline for: (i) detecting and locating in the image one or more animals, (ii) for each detected animal: (a) generating at least one section of the detected animal, (b) determining a plurality of key points in each section, (c) generating an embedding for each section based on the plurality of key points in the section, and (d) combining the embeddings for generating an identification of the detected animal with a confidence score. Key points and bounding boxes may also have associated confidence scores.....… Fig. 1, Abstract. URTASUN et al. (US 20200160117 A1) - Systems, methods, tangible non-transitory computer-readable media, and devices associated with object localization and generation of compressed feature representations are provided. For example, a computing system can access training data including a target feature representation and a source feature representation. An attention feature representation can be generated based on the target feature representation and a machine-learned attention model. An attended target feature representation can be generated based on masking the target feature representation with the attention feature representation. A matching score for the source feature representation and the target feature representation can be determined. A loss associated with the matching score and a ground-truth matching score for the source feature representation and the target feature representation can be determined. Furthermore, parameters of the machine-learned attention model can be adjusted based on the loss....… Fig. 1, Abstract. Any inquiry concerning this communication or earlier communications from the examiner should be directed to BEZAWIT N SHIMELES whose telephone number is (571)272-7663. The examiner can normally be reached M-F 7:30am-5pm. 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, Chineyere Wills-Burns can be reached at (571) 272-9752. 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. /BEZAWIT NOLAWI SHIMELES/Examiner, Art Unit 2673 /CHINEYERE WILLS-BURNS/Supervisory Patent Examiner, Art Unit 2673 Application/Control Number: 18/688,257 Page 2 Art Unit: 2673 Application/Control Number: 18/688,257 Page 3 Art Unit: 2673 Application/Control Number: 18/688,257 Page 4 Art Unit: 2673 Application/Control Number: 18/688,257 Page 5 Art Unit: 2673 Application/Control Number: 18/688,257 Page 6 Art Unit: 2673 Application/Control Number: 18/688,257 Page 7 Art Unit: 2673 Application/Control Number: 18/688,257 Page 8 Art Unit: 2673 Application/Control Number: 18/688,257 Page 9 Art Unit: 2673 Application/Control Number: 18/688,257 Page 10 Art Unit: 2673 Application/Control Number: 18/688,257 Page 11 Art Unit: 2673 Application/Control Number: 18/688,257 Page 12 Art Unit: 2673 Application/Control Number: 18/688,257 Page 13 Art Unit: 2673 Application/Control Number: 18/688,257 Page 14 Art Unit: 2673 Application/Control Number: 18/688,257 Page 15 Art Unit: 2673 Application/Control Number: 18/688,257 Page 16 Art Unit: 2673 Application/Control Number: 18/688,257 Page 17 Art Unit: 2673 Application/Control Number: 18/688,257 Page 18 Art Unit: 2673 Application/Control Number: 18/688,257 Page 19 Art Unit: 2673 Application/Control Number: 18/688,257 Page 20 Art Unit: 2673 Application/Control Number: 18/688,257 Page 21 Art Unit: 2673 Application/Control Number: 18/688,257 Page 22 Art Unit: 2673 Application/Control Number: 18/688,257 Page 23 Art Unit: 2673 Application/Control Number: 18/688,257 Page 24 Art Unit: 2673 Application/Control Number: 18/688,257 Page 25 Art Unit: 2673 Application/Control Number: 18/688,257 Page 26 Art Unit: 2673 Application/Control Number: 18/688,257 Page 27 Art Unit: 2673 Application/Control Number: 18/688,257 Page 28 Art Unit: 2673 Application/Control Number: 18/688,257 Page 29 Art Unit: 2673 Application/Control Number: 18/688,257 Page 30 Art Unit: 2673 Application/Control Number: 18/688,257 Page 31 Art Unit: 2673 Application/Control Number: 18/688,257 Page 32 Art Unit: 2673 Application/Control Number: 18/688,257 Page 33 Art Unit: 2673 Application/Control Number: 18/688,257 Page 34 Art Unit: 2673 Application/Control Number: 18/688,257 Page 35 Art Unit: 2673 Application/Control Number: 18/688,257 Page 36 Art Unit: 2673 Application/Control Number: 18/688,257 Page 37 Art Unit: 2673 Application/Control Number: 18/688,257 Page 38 Art Unit: 2673
Prosecution Insights