CTFR 18/175,750 CTFR 96329 DETAILED ACTION Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA. Response to Arguments 07-37 AIA Applicant's arguments filed 12/29/2025 have been fully considered but they are not persuasive. Regarding 35 U.S.C. 101 applicant argues in page 8-11 “ An invention is patent-eligible if it claims a "new and useful process, machine, manufacture, or composition of matter." 35 U.S.C. § 101. However, the Supreme Court has interpreted§ 101 to include implicit exceptions: "Laws of nature, natural phenomena, and abstract ideas" are not patentable. E.g., Alice Corp. v. CLS Bank Int'l, 573 U.S. 208,216 (2014)… The Problem and the Claimed Solution In the present case, the claims are directed to a technical solution to problems arising in the field of machine learning, and more specifically to efficient Hidden Markov Model (HMM) architecture that provides more accurate output inferences using fewer computational resources. For example, as the Specification explains … Specification ,-i, i [0002]-[0003] (emphasis added). The techniques disclosed in the present application provide a solution to these problems, namely with an improved HMM model architecture that adaptively modifies the output based on coefficient hyperparameters. The claims embody the solution described in the Specification (e.g., include the components or steps that provide the solution) by reciting, for example: (1) "accessing a sequence of observations"; (2) "accessing a hidden Markov model (HMM) comprising a set of transition probabilities, a set of emission probabilities, a transition coefficient hyperparameter, and an emission coefficient hyperparameter"; and (3) "generating a first output inference from the HMM based on the sequence of observations. ” The applicant argues that the claim application provides a technical improvement on the art and the specific limitations reflect the improvement. Yet regarding the independent claim 1, 11 , 21, and 30 fail to provide any information on how the hidden Markov model works. As the independent claims recite “ accessing a hidden Markov model (HMM) comprising a set of transition probabilities, a set of emission probabilities, a transition coefficient hyperparameter, and an emission coefficient hyperparameter; ” and offers no information on how HMM operates based on “ a set of transition probabilities, a set of emission probabilities, a transition coefficient hyperparameter, and an emission coefficient hyperparameter” . The claim further only recites “ generating a first output inference from the HMM based on the sequence of observations. ” The independent claims only provide the general structure by which the HMM comprises and further only provides what it outputs. Further MPEP 2106.05(a) “ After the examiner has consulted the specification and determined that the disclosed invention improves technology, the claim must be evaluated to ensure the claim itself reflects the disclosed improvement in technology. ” The independent claims fail to provide an improvement on the technology and only provide an abstract idea and what the structure of the HMM as explain in the previous office action dated 12/29/2025. Applicant further argues in page 11-15 “ The Examiner concedes that the claims relate to one of the statutory categories of patentable subject matter (e.g., processes, machines). Office Action, p. 2. Thus, the analysis proceeds to Step 2A of the Alice/Mayo test… Here, like the reasoning rejected by the ARP in Desjardins, the Examiner argues that various features of the present claims "us[e] a generic computer to access the HMM." Office Action, p. 6. Applicant respectfully disagrees and submits that the Examiner is evaluating the claims at a high level of generality without considering whether the technical features confer a technological improvement to a technical problem. Like the claimed technology in Desjardins, the claimed solution of the present disclosure improves machine learning by reducing computational resources and increasing model accuracy (e.g., by exploiting coefficient hyperparameters to adaptively modify the output of the HMM). As such, the present claims provide an improvement to machine learning that integrates any alleged abstract idea into a practical application. Accordingly, Applicant submits that the claims are directed to a specific improvement in machine learning technology and the field of implementing machine learning models on physical devices, and are thus eligible subject matter under Step 2A, Prong 2 of the Alice/Mayo test… Similar to BASCOM, the present claims recite non-conventional and non-generic methods and systems, in this case for an efficient HMM model architecture. Applicant's claimed solution improves upon conventional techniques, for example, by adaptively modifying the output of an HMM using coefficient hyperparameters, thereby increasing inference accuracy and more efficiently leveraging the computational capabilities of the device. E.g., Specification ,-i,i [0046], [0053], [0068]. The improvements are accomplished by particular features in the claims, such as (1) "accessing a sequence of observations"; (2) "accessing a hidden Markov model (HMM) comprising a set of transition probabilities, a set of emission probabilities, a transition coefficient hyperparameter, and an emission coefficient hyperparameter"; and (3) "generating a first output inference from the HMM based on the sequence of observations. Accordingly, Applicant submits that the claims are eligible under Step 2B of the Alice/Mayo test and are therefore directed to eligible subject matter under 35 U.S.C. § 101. Thus, for at least these additional reasons, Applicant respectfully requests that the rejection of Claims 1-30 under 35 U.S.C. § 101 be withdrawn. " The applicant argues Step 2A, prong 2 and Step 2B by citing court cases and appeals review panels decision. Yet the applicant relies on the specification to provide the necessary detail to overcome the 101 rejection. As explain the claims have to provide the necessary structure to ensure the it reflects the disclosed improvement in the technology. The independent claims fail to provide how the HMM uses emission and transition hyperparameters. The claims also do not provide any information regarding the HMM works based on the transition coefficient hyperparameter and the emission coefficient hyperparameter. As the claims are examined based on the claim limitations and not the specification thus the argument presented by the applicant are not convincing and moot. Regarding applicants arguments for U.S.C 102 rejection, applicant argues in page 15-17 “ Applicant submits that Wang does not describe, explicitly or implicitly, each and every element of the claims. For example, Wang fails to anticipate or suggest "accessing a hidden Markov model (HMM) comprising a set of transition probabilities, a set of emission probabilities, a transition coetficient hyperparameter, and an emission coetficient hyperparameter" as recited in independent Claim 1 (emphasis added) and similar features recited in independent Claims 11, 21, and 30… As an initial matter, "hyperparameter" is a technical term of art such that one of ordinary skill would appreciate that it is distinct from a mere "parameter." Hyperparameters are generally understood to be settings for any configurable part of a machine learning model's learning process (i.e., settings that control how the model learns)…. Furthermore, Applicant submits that Wang's wtr and wem are not coefficient hyperparameters. A coefficient is well-understood to be a multiplicative factor in a mathematical expression… For the foregoing reasons, fails to describe at least "a hidden Markov model (HMM) comprising ... a transition coefficient hyperparameter[] and an emission coefficient” Regarding claim 1 and analogous claims 11 and 21, hyperparameter." Accordingly, Applicant submits that Claims 1, 11, 21, and 30, as well as claims dependent thereon, are allowable and respectfully requests withdrawal of this rejection. ” Applicant argues that prior art cited does not teach transition coefficient hyperparameter and an emission coefficient hyperoperator as claim in in independent claims 1, 11, 21, and 30. Yet the claims do not provided any function by which the hyperparameters are used or set and only limit the claim to be part of accessing the HMM. Further the specification of the claimed invention calls for hyperparameters or parameters in multiple sections such as in paragraph [0023] “ Additionally, in Equation 1, y (gamma), a (alpha), b (beta), and ζ (zeta) are new parameters or hyperparameters (discussed in more detail below), which may be static/defined (e.g., manually by a user), learned during training (e.g., based on training data) and fixed during inference, and/or dynamically determined during both training and inference (e.g., based on observed data and/or determined states) ” and paragraph 0089 line 5-7 “ In some aspects, as discussed above, the particular parameter(s) and/or hyperparameters to be used may vary depending on the particular implementation and/or the particular time step or index of the selected observation .” The specification specifies them as being both parameters or hyperparameters thus they are interpreted as parameter as cited above or other sections such paragraph 0090. Further Wang the reference in questions is controlled by emission and transitions as they directly affect the outcome of the PHMM and are used during learning (Wang page 3 3.1. Parameter Vector and Feature Function line 6-7, transition parameters w t r and emission parameters w e m . (Page 4 Algorithm 1 shows the main steps of the learning procedure.) ). Under reasonable interpretation of the claims Wang teaches the limitations of the independent claims. Applicant further argues in page 18-19 that the remaining claims should be allowable as they depend on the independent claims. However the arguments are not persuasive and moot . Claim Rejections - 35 USC § 101 07-04-01 AIA 07-04 35 U.S.C. 101 reads as follows: Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title. Claims 1-27, and 30-32 are rejected under 35 U.S.C. 101 because the claimed invention is directed to abstract idea without significantly more. The claim(s) recite(s) significantly more. The subject matter eligibility test for products and process is describe below for claim 1 in view of dependent claims. Regarding claim 1: Step 1: Is the claim to a process machine manufacture or composition of matter? Yes – Claim 1 recites a method, which a method falls under the statutory categories. Step 2A Prong 1: Does the claim recite an abstract idea, law of nature, or natural phenomenon? Yes – The claim recites the following: “ generating a first output inference from the HMM based on the sequence of observations. ” - The limitations of claim 1 recites a mental process of generating a first output inference based on the observations (see MPEP 2106.04(a)(2)III). Step 2 Prong 2: Does the claim recite additional elements that integrate the judicial exception into a particular application? No – The claim includes the additional element(s): “ A processor-implemented method, comprising: accessing a sequence of observations; ” The additional elements fall under Insignificant Extra-Solution Activity as mere data gathering by accessing a sequence of observations. See MPEP 2106.5(g). “ accessing a hidden Markov model (HMM) comprising a set of transition probabilities, a set of emission probabilities, a transition coefficient hyperparameter, and an emission coefficient hyperparameter; ” The additional elements fall under “apply it” as using a generic computer to access the HMM. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)). Step 2B: Does the claim recite additional elements that amount to significantly more than the judicial exception? No - The claim does not include additional elements that are sufficient to amount to a significantly more than the judicial exemption. As an order whole, the claim is directed to a mental process of using a hidden Markov model and neural networks to generate inferences based on observations. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of accessing fall under using generic computer to apply an exemption and mere data gathering. The method does not improve on the function of a computer, transforms an article into another article, nor is it applied by a particular machine, making the claim not patent eligible. Regarding claim 2: Step 2A Prong 2, Step 2B: The additional element(s): “ The processor-implemented method of claim 1, wherein the HMM further comprises a linear hyperparameter. ” The additional elements fall under Insignificant Extra-Solution Activity. See MPEP 2106.5(g). No additional elements. The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application. Regarding claim 3: Step 2A Prong 2, Step 2B: The additional element(s): “ The processor-implemented method of claim 1, wherein the transition coefficient hyperparameter and the emission coefficient hyperparameter were learned based on training data while training the HMM. ” The additional elements fall under Insignificant Extra-Solution Activity. See MPEP 2106.5(g). No additional elements. The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application Regarding claim 4: Step 2A Prong 2, Step 2B: The additional element(s): “ The processor-implemented method of claim 1, further comprising refining the transition coefficient hyperparameter and the emission coefficient hyperparameter using continual learning based on the first output inference. ” The additional elements fall under “apply it” as using a generic computer to use continual learning to refine the transition and emission coefficient hyperparameters. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)). Regarding claim 5: Step 2A Prong 2, Step 2B: The additional element(s): “ The processor-implemented method of claim 1, further comprising generating the transition coefficient hyperparameter and the emission coefficient hyperparameter based on one or more observations of the sequence of observations. ” The additional elements fall under “apply it” as using a generic computer to generate the transition and emission coefficient hyperparameters based on the observations. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)). Regarding claim 6: Step 2A Prong 2, Step 2B: The additional element(s): “ The processor-implemented method of claim 5, wherein generating the transition coefficient hyperparameter and the emission coefficient hyperparameter comprises processing the one or more observations of the sequence of observations using a neural network. ” The additional elements fall under “apply it” as using a generic computer to use a neural network to process the observations. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)). Regarding claim 7: Step 2A Prong 2, Step 2B: The additional element(s): “ The processor-implemented method of claim 1, wherein the HMM further comprises at least one of: a cross-correlation hyperparameter, a quadratic hyperparameter, or a nonlinear hyperparameter. ” The additional elements fall under Insignificant Extra-Solution Activity. See MPEP 2106.5(g). The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application. Regarding claim 8: Step 2A Prong 2, Step 2B: The additional element(s): “ The processor-implemented method of claim 1, wherein: at least one output inference from the HMM is generated using one or more transition probabilities and one or more emission probabilities, ” The additional elements fall under “apply it” as using a generic computer to output an inference using the transition and emission probabilities. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)). and at least one output inference from the HMM is generated using a neural network classifier that does not use the set of transition probabilities and the set of emission probabilities. The additional elements fall under “apply it” as using a generic computer to output an inference using a neural network classifier. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)). Regarding claim 9: Step 2A Prong 2, Step 2B: The additional element(s): “ The processor-implemented method of claim 1, further comprising generating an output inference, wherein generating the output inference includes processing the first output inference of the HMM using a neural network.. ” The additional elements fall under “apply it” as using a generic computer to generate an output inference using a neural network. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)). Regarding claim 10: Step 2A Prong 2, Step 2B: The additional element(s): “ The processor-implemented method of claim 1, further comprising: generating a second output inference, wherein generating the second output inference includes processing the sequence of observations using a recurrent neural network (RNN); ” The additional elements fall under “apply it” as using a generic computer to generate a second output inference using a RNN. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)). “ generating an overall output inference, wherein generating the overall output inference includes processing the first and second output inferences using a multilayer perceptron (MLP). ” The additional elements fall under “apply it” as using a generic computer to generate an overall output inference using a MLP. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)). Claims 11-20 recite a system and are analogous to the method of claims 1-10. Therefore, the rejections of claim 1-10 above applies to claims 11-20. Claims 21-27 recite a CRM and are analogous to the method of claims 1-10. Therefore, the rejections of claim 1-10 above applies to claims 21-27. Claims 30 recite a system and are analogous to the method of claims 1. Therefore, the rejections of claim 1 above applies to claims 30. Regarding claim 31: Step 2A Prong 1: Does the claim recite an abstract idea, law of nature, or natural phenomenon? Yes – The claim recites the following: “ wherein the transition coefficient hyperparameter is configured to modify one or more transition probability elements in the set of transition probabilities ” - The limitation recites a mathematical process of modifying the one or more transition probability using t he transition coefficient hyperparameter (see MPEP 2106.04(a)(2)I). Step 2A Prong 2, Step 2B: The additional element(s): No additional elements. The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application Regarding claim 32: Step 2A Prong 1: Does the claim recite an abstract idea, law of nature, or natural phenomenon? Yes – The claim recites the following: “ The processing system of Claim 11, wherein the emission coefficient hyperparameter is configured to modify one or more emission probability elements in the set of emission probabilities ” - The limitation recites a mathematical process of modifying the one or more emission probability using the emission coefficient hyperparameter (see MPEP 2106.04(a)(2)I). Step 2A Prong 2, Step 2B: The additional element(s): No additional elements. The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application Claim Rejections - 35 USC § 102 07-07-aia AIA 07-07 The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action: A person shall be entitled to a patent unless – 07-08-aia AIA (a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention. 07-12-aia AIA (a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention. 07-15 AIA Claim (s) 1-3, 5, 7, 11-13, 15, 17, 21-23, 25, 26, 30 are rejected under 35 U.S.C. 102( a)(1 ) as being anticipated by WANG Z., et al., "A Pair Hidden Markov Support Vector Machine for Alignment of Human Actions", Proceedings of the 2016 IEEE International Conference on Multimedia and Expo (ICM'16), 11 July 2016, 6 Pages, XP032953214, (“Wang”) . Regarding claim 1 and analogous claims 11, 21, 30, Wang teaches A processor-implemented method, comprising: accessing a sequence of observations (Wang Page 2 2.1. Pair Hidden Markov Model, Wang page 2 2.1 Pair Hidden Markov Model para 1, PHMM is a probabilistic model for pairwise sequence alignments. Given two sequences, s = { s 1 , … , s i , … , s L s and t = { t 1 , … , t j , … , t L t } , their alignment can be intuitively defined as a sequence of index pairs from the two sequences. However, to simplify both notations and operations, the alignment is re-defined as a sequence of only three types of symbols: M (“match”), S (“insert a gap on sequence s”) and T (“insert a gap on sequence t”). The symbols have the following meaning: assuming i and j to be the current indices over sequences s and t, respectively, 1) symbol M pairs frames si and tj and then increments both indices; 2) symbol S pairs no frames and only increments index j; and, likewise, 3) symbol T pairs no frames and only increments index i. 2.2. Structural SVM Para 1 line 1-8,, Structural SVM is a powerful classifier that extends the notion of maximum-margin classification to the case of structured prediction. This case includes the classification of structures such as sequences and graphs and problems such as alignment and ranking. In the case of alignment, the problem is to learn a scoring function, F(s; t; y), between input sequences s and t and output alignment y based on training samples of input-output pairs [ a sequence of observations ].) ; accessing a hidden Markov model (HMM) comprising a set of transition probabilities, a set of emission probabilities, a transition coefficient hyperparameter, and an emission coefficient hyperparameter (Wang page 3 3. THE PROPOSED INTEGRATION: PHMM-SSVM para 1, The integration of the PHMM in the SSVM framework (PHMM-SSWM hereafter) can be obtained by simply setting the PHMM’s joint probability as: PNG media_image1.png 41 439 media_image1.png Greyscale [ accessing a hidden Markov model (HMM) ] 3.1 3.1. Parameter Vector and Feature Function para line 1-8, In the structural SVM framework, the score for a sample (s; t; y) is obtained from the product of a parameter vector, w, and a feature function, Ψ , that provides a re-mapping of the given measurements and labels. As a common assumption, the score is assumed to be a decomposable function (a sum) over the individual labels of the assignment, y k , k = 1 … | y | . The parameter vector contains two sections: transition parameters w t r and emission parameters w e m [ comprising a set of transition probabilities, a set of emission probabilities, a transition coefficient hyperparameter, and an emission coefficient hyperparameter ].) ; and generating a first output inference from the HMM based on the sequence of observations (Wang page 2 2.1. Pair Hidden Markov Model para 3 line 1-4, In probability notation, a PHMM is a model for the joint probability, p(s, t, y), of the two sequences and their alignment. Such a model can be used to infer an optimal alignment, _y, for the two sequences as y - = a r g m a x y p(s, t, y) [ and generating a first output inference from the HMM ]. Page 3 3.2. Loss Functions para 1, The common way to measure the inaccuracy of a predicted alignment is by use of a Hamming distance between the ground-truth alignment, y, and the prediction, y - . This function is often referred to as Q-loss function in the alignment literature and noted as ∆ Q ( y , y - ) [13]. The Q-loss is decomposable over the individual operations in the alignments as ∆ Q ( y , y - ) = 1 - ∑ k = 1 | y | δ ( y k , y - k ) At its turn, δ ( y k , y - k ) returns 1=N (N: number of frame matches in the ground truth) when a ground-truth match is correctly predicted and 0 otherwise. In practice, we compute the loss by explicitly unfolding all the frame indices over sequences s and t in both the ground truth and the predicted alignment [ based on the sequence of observations ]) . Regarding claim 2 and analogous claims 12 and 22, Wang teaches The processor-implemented method of claim 1. Wang teaches wherein the HMM further comprises a linear hyperparameter (Page 3 3.3 Most-Violated Constrains para 1 line 6-10, The solution proposed by [8, 12] is a relaxed problem (6) that only considers the sub-set of the “most-violated constraints”, i.e. the constraints that set the value of penalty for each sample, i = 1 …N. The solution is proven to be ε-close to that of the fully-constrained problem, where ε is a small constant (set to 0.01 in our experiments) [ a linear hyperparameter ] that can be made arbitrary smaller at the cost of only a polynomial increase in the number of iterations of the solver. Page 4 Algorithm 1 line 3-6, PNG media_image3.png 205 573 media_image3.png Greyscale ( Examiner Note: At line 5 ε is added to PNG media_image2.png 29 24 media_image2.png Greyscale )). Regarding claim 3 and analogous claims 13 and 23, Wang teaches The processor-implemented method of claim 1. Wang teaches wherein the transition coefficient hyperparameter and the emission coefficient hyperparameter were learned based on training data while training the HMM (Wang Page 3 3.1 3.1. Parameter Vector and Feature Function para line 1-8, In the structural SVM framework, the score for a sample (s; t; y) is obtained from the product of a parameter vector, w, and a feature function, Ψ , that provides a re-mapping of the given measurements and labels. As a common assumption, the score is assumed to be a decomposable function (a sum) over the individual labels of the assignment, y k , k = 1 … | y | . The parameter vector contains two sections: transition parameters w t r and emission parameters w e m . The transition parameters are a (partial) 3 X 3 matrix indexed by labels yk-1 and yk (transitions between symbols S → T and T → S are not allowed). As emission feature, we simply consider the absolute difference of measurements si and tj in matching states; therefore, the emission parameters are a vector with the same dimensionality as the individual measurements, i.e. w e m , s i , t j ∈ R D . Logically, w e m should be assigned negative values during training so that more dissimilar measurements receive lower scores; however, we do not impose a negativity constraint on these parameters [ wherein the transition coefficient hyperparameter and the emission coefficient hyperparameter ]. Page 4 4. Experiments, The following experiments evaluate the proposed PHMMSSVM model in the temporal alignment of action videos against DTW [1] and a state-of-the-art algorithm, CTW [5]. In the first experiment, we compare the performance in aligning the “jump” action from different subjects of the Weizmann dataset [9]. In the second experiment, we compare the “clean-and-jerk” action performed by 11 subjects from the challenging Olympic Sports dataset [10]. For the SSVM training, we have set parameters C to 10 and _ to 0:01, with no noticeable sensitivity. Results are reported in terms of both Q-loss and Q4-loss (see section 3.2) [ were learned based on training data while training the HMM. ]) . Regarding claim 5 and analogous claims 15 and 25, Wang teaches the processor-implemented method of claim 1. Wang teaches further comprising generating the transition coefficient hyperparameter and the emission coefficient hyperparameter based on one or more observations of the sequence of observations ((2.1. Pair Hidden Markov Model para 1 line 1-4, line PHMM is a probabilistic model for pairwise sequence alignments. Given two sequences, s = { s 1 , … , s i , s L s } and t = { t 1 , … , t j , … , t L t }, their alignment can be intuitively defined as a sequence of index pairs from the two sequences [ based on one or more observations of the sequence of observations ]. 3.1 3.1. Parameter Vector and Feature Function para line 1-8, In the structural SVM framework, the score for a sample (s; t; y) is obtained from the product of a parameter vector, w, and a feature function, Ψ , that provides a re-mapping of the given measurements and labels. As a common assumption, the score is assumed to be a decomposable function (a sum) over the individual labels of the assignment, y k , k = 1 … | y | . The parameter vector contains two sections: transition parameters w t r and emission parameters w e m . Page 4 3.3 Most-Violated Constraints para 2, As (9) shows, the alignment y * i setting the value of penalty PNG media_image4.png 30 26 media_image4.png Greyscale can be found by a modified version of the inference, known as “loss-augmented” inference since it adds up the loss function to the score. Given that the loss function is decomposable frame-by-frame, the efficient Viterbi algorithm can also be used for this maximisation. Algorithm 1 shows the main steps of the learning procedure. ( Examiner Note: The model generates the parameters as it finds the aliment between observations ) . Regarding claim 7 and analogous claims 17 and 26, Wang teaches The processor-implemented method of claim 1. Wang teaches wherein the HMM further comprises at least one of: a cross-correlation hyperparameter, a quadratic hyperparameter, or a nonlinear hyperparameter (Wang Algorithm 1 line 9, PNG media_image5.png 37 546 media_image5.png Greyscale [ a quadratic hyperparameter ] ( i.e. w is a hyperparameter in the algorithm and it is squared making it a quadratic hyperparameter )) . Claim Rejections - 35 USC § 103 07-20-aia AIA The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action: A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made. 07-21-aia AIA Claim (s) 4, 14, 24, 31, and 32 are rejected under 35 U.S.C. 103 as being unpatentable over Wang in view of E. Dorj, C. Chen and M. Pecht, "A Bayesian Hidden Markov Model-based approach for anomaly detection in electronic systems," 2013 IEEE Aerospace Conference, Big Sky, MT, USA, 2013, pp. 1-10 (“Dorj”) . Regarding claim 4 and analogous claims 14 and 24, Wang teaches the processor-implemented method of claim 1. Wang does not explicitly teach further comprising refining the transition coefficient hyperparameter and the emission coefficient hyperparameter using continual learning based on the first output inference. However Dorj teaches further comprising refining the transition coefficient hyperparameter and the emission coefficient hyperparameter using continual learning based on the first output inference (Dorj Page 2 2. BAYESIAN HIDDEN MARKOV MODEL para 3 line 1-6, To define our Bayesian HMM, we specify Dirichlet distributions for the parameters θ={A,B,π}, with set of hyperparameters w = { w A , w B , w ( π ) } .. In Bayesian statistics, a Dirichlet distribution is considered as the conjugate prior of the categorical distribution. Page 4 2. BAYESIAN HIDDEN MARKOV MODEL Expected counts in the observation sequence, The posterior hyperparameters need to be re-estimated in each of the iterations of HMM training [further comprising refining], since the Baum-Welch algorithm is an iterative learning algorithm. In order to update the posterior hyperparameters, variables and need to be identified first. The variable is the probability of being in state Si at time t and in state Sj at time t+1, given the observation sequence O_1 O_2…O_T and defined as follows [13]: Once the counts have been estimated as above, distributions of the posterior parameters A, B, and an be defined as shown in Eqs. (10)-(12) . Page 5 (i.e the hyperparameters are refined using the first output inference) PNG media_image6.png 757 632 media_image6.png Greyscale [ using continual learning based on the first output inference ]) . Wang and Dorj are considered to be analogous to the claim invention because they are in the same field of machine learning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Wang to incorporate the teachings of Dorj to disclose refining the hyperparameters. Doing so to determine whether the convergence criteria have been meet (Dorj page 5 Bayesian HMM learning algorithm, The expected counts are used to estimate the posterior hyperparameters in the ith iteration using the conjugate prior method (Eqs. (10) - (12)). The posterior hyperparameters and the log-likelihood of the data are used to estimate the negative free energy to determine whether the convergence criteria have been met. If the convergence criteria have not been met, then the next iteration is started, or else the training is stopped. The output of the training process is the posterior parameters of the Bayesian HMM.). Regarding claim 31, Wang teaches the processor-implemented method of claim 1 and analogous claims 11, 21, and 30. Wang and Dorj are combine in the same rational as set forth above with respect to claim 4 and analogous claims 14 and 24. Dorj further teaches wherein the transition coefficient hyperparameter is configured to modify one or more transition probability elements in the set of transition probabilities (Dorj page 2 2. Bayesian Hidden Markov Model para 1 line 12 - 26 PNG media_image7.png 521 590 media_image7.png Greyscale ) Dorj Page 2 2. BAYESIAN HIDDEN MARKOV MODEL para 3 line 1-6, To define our Bayesian HMM, we specify Dirichlet distributions for the parameters θ = { A , B , π } , with set of hyperparameters w = { w A , w B , w ( π ) } [ transition coefficient hyperparameter ]. In Bayesian statistics, a Dirichlet distribution is considered as the conjugate prior of the categorical distribution. Dorj Page 2-3 2. BAYESIAN HIDDEN MARKOV MODEL, Equation 7 and 8 PNG media_image8.png 263 596 media_image8.png Greyscale [ is configured to modify one or more transition probability elements in the set of transition probabilities ]) . Regarding claim 32, Wang teaches the processing system of claim 11. Wang and Dorj are combine in the same rational as set forth above with respect to claim 4 and analogous claims 14 and 24. Dorj further teaches wherein the emission coefficient hyperparameter is configured to modify one or more emission probability elements in the set of emission probabilities (Dorj page 2 2. Bayesian Hidden Markov Model para 1 line 12 - 26 PNG media_image7.png 521 590 media_image7.png Greyscale ) Dorj Page 2 2. BAYESIAN HIDDEN MARKOV MODEL para 3 line 1-6, To define our Bayesian HMM, we specify Dirichlet distributions for the parameters θ = { A , B , π } , with set of hyperparameters w = { w A , w B , w ( π ) } [ emission coefficient hyperparameter ]. In Bayesian statistics, a Dirichlet distribution is considered as the conjugate prior of the categorical distribution. Dorj Page 2-3 2. BAYESIAN HIDDEN MARKOV MODEL, Equation 7 and 8 PNG media_image8.png 263 596 media_image8.png Greyscale PNG media_image9.png 49 515 media_image9.png Greyscale [ configured to modify one or more emission probability elements in the set of emission probabilities ]) . 07-21-aia AIA Claim (s) 6 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Wang in view of Pasa, Luca, Alberto Testolin, and Alessandro Sperduti. "Neural Networks for Sequential Data: a Pre‐training Approach based on Hidden Markov Models." Neurocomputing 169 (2015): 323-333 (“Pasa”) . Regarding claim 6 and analogous claim 16, Wang teaches the processor-implemented method of claim 5. Wang does not explicitly teach wherein generating the transition coefficient hyperparameter and the emission coefficient hyperparameter comprises processing the one or more observations of the sequence of observations using a neural network. Pasa wherein generating the transition coefficient hyperparameter and the emission coefficient hyperparameter comprises processing the one or more observations of the sequence of observations using a neural network (Pasa page 326 2.2 Recurrent Neural Networks, PNG media_image10.png 299 516 media_image10.png Greyscale page 327 3. Out pre-training approach line 15-36, The first element of the sequence is then sampled according to the emission distribution associated with that hidden state, and the same process is iteratively repeated by selecting the next target state according to the learned state transition probabilities. A straight forward, naïve implementation of the random sampling process can be obtained by first computing the cumulative probability distribution from the target distribution and then selecting the element corresponding to a random number drawn from the interval[0,1].Notably, our pretraining procedure does not need any form of bootstrapping from the original sequences, because the smooth dataset is generated by sampling the HMM in a completely unconstrained fashion. The simplified, HMM-generated dataset is then used to pretrain the more powerful nonlinear model, with the aim of transferring the knowledge acquired by the HMM to the recurrent neural network. There current network pre-training phase uses the same algorithm that is used for the normal training phase. After completing the pretraining phase on the smooth dataset, the neural network is then fine-tuned using the original music sequences, in order to allow the nonlinear model to extract more complex structure from the data distribution. The pseudocode for the proposed HMM-based pretraining method is given in Algorithm1, and a flowchart of the procedure is illustrated in Fig.3 [ processing the one or more observations of the sequence of observations using a neural network. ]) . Wang and Pasa are considered to be analogous to the claim invention because they are in the same field of machine learning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Wang to incorporate the teachings of Pasa to disclose using RNN to process the observations. Doing so to implement a determinist model using a RNN (Pasa page 324 1. Introduction left column, Because of that, they aim at reconstructing such model, with the goal of successfully extrapolating the learned function to the full sequence domain. Model-based approaches are typically computationally demanding, however, if a good approximation of the target model is learned, very good performances on the whole sequence domain can be obtained. Graphical models [18], and in particular Hidden Markov Models(HMMs), are often used as learning models. HMMs assume that each sequence item has been generated by hidden variables that are not directly observable. The way each sequence it em observed at time t is generated is described by a (parametric) probability distribution which depends on the state at time t of the HMM, i.e. the values taken by the hidden variables at time t; moreover, another(parametric) probability distribution drives the way the values assigned to hidden variables change through time. Learning aim sat tuning the se probability distributions in order to make the observed sequences more likely to be generated when sampling from the model. A deterministic alternative to HMMs is given by Recurrent Neural Networks(RNNs),which can be under-stood as non-linear dynamical systems where learning is performed by using gradient-based approaches [13]. From an abstract computational point of view, given a graphical model for sequences, it is possible to specify a RNN which constitutes a specific deterministic implementation of that graphical model [19]. Due to their non-linearity, RNNs are potentially very expressive and powerful. ) . 07-21-aia AIA Claim (s) 8, 18, and 27 are rejected under 35 U.S.C. 103 as being unpatentable over Wang in view of Graves, Alex, and Navdeep Jaitly. "Towards end-to-end speech recognition with recurrent neural networks." International conference on machine learning. PMLR, 2014, (“Graves”) . Regarding claim 8 and analogous claims 18 and 27, Wang teaches the processor-implemented method of claim 1. Wang teaches at least one output inference from the HMM is generated using one or more transition probabilities and one or more emission probabilities (Wang page 2, 2.1. Pair Hidden Markov Model para 3-4, In probability notation, a PHMM is a model for the joint probability, p(s; t; y), of the two sequences and their alignment. Such a model can be used to infer an optimal alignment _y, for the two sequences as y ̅=argmax_y p(s,t,y). Like for a conventional HMM, the joint probability of a PHMM factorises into a set of transition and emission probabilities. The transition probabilities are commonly defined as: (1) δ for transitions from M to either S or T; (2) " for staying in S or T; (3) 1 ε for transitions from either S or T to M. Note that the model bars direct transitions from S to T and the vice versa assuming that a pair of matched frames will always follow a run of gaps. Figure 1 shows the state diagram of the PHMM, while Table 1 shows the complete transition probabilities table. To complete the model, we also need to define the emission probabilities. To this aim, we note the probability of emitting aligned pair (a; b) as pa;b and the probability of emitting measurement a against a gap as qa. In the common case of numerical measurements, both p and q will be multi-variate likelihoods such as Gaussian distributions or mixture models [ generated using one or more transition probabilities and one or more emission probabilities ] 3.3. Most-Violated Constrains para 2, As (9) shows, the alignment y^(*i) setting the value of penalty ε^i can be found by a modified version of the inference, known as “loss-augmented” inference since it adds up the loss function to the score [at least one output inference]. Given that the loss function is decomposable frame-by-frame, the efficient Viterbi algorithm can also be used for this maximisation. Algorithm 1 shows the main steps of the learning procedure.), Wang does not explicitly teach and at least one output inference from the HMM is generated using a neural network classifier that does not use the set of transition probabilities and the set of emission probabilities. However Graves teaches and at least one output inference from the HMM is generated using a neural network classifier that does not use the set of transition probabilities and the set of emission probabilities (Graves page 2 I. Introduction para 4-5, Finally, the acoustic scores produced by the HMM are combined with a language model trained on a text corpus. In general the language model contains a great deal of prior information, and has a huge impact on performance. Modelling language separately from sound is perhaps the most justifiable departure from end-to-end learning, since it is easier to learn linguistic dependencies from text than speech, and arguable that literate humans do the same thing. Nonetheless, with the advent of speech corpora containing tens of thousands of hours of labelled data, it may be possible to learn the language model directly from the transcripts. The goal of this paper is a system where as much of the speech pipeline as possible is replaced by a single recurrent neural network (RNN) architecture. Although it is possible to directly transcribe raw speech waveforms with RNNs (Graves, 2012, Chapter 9) or features learned with restricted Boltzmann machines (Jaitly & Hinton, 2011), the computational cost is high and performance tends to be worse than conventional preprocessing. We have therefore chosen spectrograms as a minimal preprocessing scheme. Page 3 3. Connectionist Temporal Classification para 1. Neural networks (whether feedforward or recurrent) are typically trained as frame-level classifiers in speech recognition. This requires a separate training target for every frame, which in turn requires the alignment between the audio and transcription sequences to be determined by the HMM. However the alignment is only reliable once the classifier is trained, leading to a circular dependency between segmentation and recognition (known as Sayre’s paradox in the closely-related field of handwriting recognition). Furthermore, the alignments are irrelevant to most speech recognition tasks, where only the word-level transcriptions matter. Connectionist Temporal Classification (CTC) (Graves, 2012, Chapter 7) is an objective function that allows an RNN to be trained for sequence transcription tasks without requiring any prior alignment between the input and target sequences. Page 4 para 2 line 1-4, The output layer contains a single unit for each of the transcription labels (characters, phonemes, musical notes etc.), plus an extra unit referred to as the ‘blank’ which corresponds to a null emission Given a length T input sequence x, the output vectors yt are normalised with the softmax function, then interpreted as the probability of emitting the label (or blank) with index k at time t: where yk t is element k of yt. A CTC alignment a is a length T sequence of blank and label indices. The probability Pr(ajx) of a is the product of the emission probabilities at every time-step: [ and at least one output inference from the HMM is generated using a neural network classifier ]. 5. Decoding Para 1, Decoding a CTC network (that is, finding the most probable output transcription y for a given input sequence x) can be done to a first approximation by picking the single most probable output at every timestep and returning the corresponding transcription: More accurate decoding can be performed with a beam search algorithm, which also makes it possible to integrate a language model. The algorithm is similar to decoding methods used for HMM-based systems, but differs slightly due to the changed interpretation of the network outputs. In a hybrid system the network outputs are interpreted as posterior probabilities of state occupancy, which are then combined with transition probabilities provided by a language model and an HMM. With CTC the network outputs themselves represent transition probabilities (in HMM terms, the label activations are the probability of making transitions into different states, and the blank activation is the probability of remaining in the current state) [ is generated using a neural network classifier that does not use the set of transition probabilities and the set of emission probabilities ] . ( Examiner Note: The system uses the CTC to allow the RNN to be trained and does not use the HMM transition and emission probabilities )) . Wang and Graves are considered to be analogous to the claim invention because they are in the same field of machine learning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Wang to incorporate the teachings of Graves to disclose generating an output inference using a Neural network. Doing so to use an RNN in a speech pipeline and achieve reasonable accuracy (Graves page 1 1. Introduction para 2 line 14--26, The objective function used to train the networks is therefore substantially different from the true performance measure (sequence-level transcription accuracy). This is precisely the sort of inconsistency that end-to-end learning seeks to avoid. In practice it is a source of frustration to researchers, who find that a large gain in frame accuracy can translate to a negligible improvement, or even deterioration in transcription accuracy. An additional problem is that the frame-level training targets must be inferred from the alignments determined by the HMM. This leads to an awkward iterative procedure, where network retraining is alternated with HMM realignments to generate more accurate targets. page 2 1. Introduction para 5 and 7 The goal of this paper is a system where as much of the speech pipeline as possible is replaced by a single recurrent neural network (RNN) architecture. Although it is possible to directly transcribe raw speech waveforms with RNNs (Graves, 2012, Chapter 9) or features learned with restricted Boltzmann machines (Jaitly & Hinton, 2011), the computational cost is high and performance tends to be worse than conventional preprocessing. We have therefore chosen spectrograms as a minimal preprocessing scheme. Experiments on the Wall Street Journal speech corpus demonstrate that the system is able to recognize words to reasonable accuracy, even in the absence of a language model or dictionary, and that when combined with a language model it performs comparably to a state-of-the-art pipeline.) . 07-21-aia AIA Claim (s) 9 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Wang in view of Edmondo Trentin & Roldano Cattoni (1999) Learning Perception for Indoor Robot Navigation with a Hybrid Hidden Markov Model/Recurrent Neural Networks Approach, Connection Science, 11:3-4, 243-265, (“Trentin”) . Regarding claim 9 and analogous claims 19, Wang teaches the processor-implemented method of claim 1. Wang does not explicitly teach further comprising generating an output inference, wherein generating the output inference includes processing the first output inference of the HMM using a neural network. Trentin teaches further comprising generating an output inference, wherein generating the output inference includes processing the first output inference of the HMM using a neural network (Trentin page 254 3.3. Integrating HMMs with RNNs para 4 line 8-22 and para 7, In the present research, an approximate version of equation (3) is used, by taking: that is to say, for each state of the HMM a network is trained to estimate the state posterior given the partial sensory vector sequence . This is accomplished using a supervised gradient-descent training algorithm (as described in the previous section) on a RNN for each one of the states. The training set Tj for the j th network is obtained by labeling the F th pattern of each training sequence xL1 with the target probability values of 1.0 or 0.0 according to the fact that the robot was, or was not, in state qj at time F during acquisition of sensory signal xz. All the available training sequences X are used to train all the Q RNNs, with different distributions of the corresponding labels. This instantiates the concept of discriminative training, since models are trained on `positive’ as well as on `negative’ examples. As pointed out in Bourlard and Morgan (1994), this is no longer strictly an HMM, but the general framework is still valid and the standard decoding algorithms (e.g. Viterbi) can still be used [ wherein generating the output inference includes processing the first output inference of the HMM using a neural network ]. Page 256, The input layer consisted of 16 units (one for each sonar): the RNN was enlarged with the addition of 16 context units, as described in Section 3.2. An individual ANN was trained for each state of the HMM, and the output from the ith net was assumed to be an estimate of the corresponding state-posterior probability PNG media_image11.png 29 88 media_image11.png Greyscale given the current sonar observations (MLP case), or the state-posterior PNG media_image12.png 27 92 media_image12.png Greyscale given the sequence of sonar readings observed so far (RNN case) [ further comprising generating an output inference ]) . Wang and Trentin are considered to be analogous to the claim invention because they are in the same field of machine learning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Wang to incorporate the teachings of Trentin to use a RNN to process a first output inference. Doing so to take the advantages of RNN in keep an internal trace or memory of the past in combination with HMM (Trentin page 252 para 1, Like HMMs, RNNs are particularly suited for sequence processing, due to their ability to keep an internal trace, or memory, of the past. This memory is combined with the current input to provide a context-dependent output. Section 3.3 describes how the long-term sequence processing capabilities of HMMs can be combined with RNNs to handle the problem of non-stationarity of observations within individual states of the hidden Markov chain.) . 07-21-aia AIA Claim (s) 10 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Wang in view of Trentin and further in view of Chu et al. (US2020/0074303A1) (“Chu”) . Regarding claim 10 and analogous claim 20, Wang teaches the processor-implemented method of claim 1. Wang and Trintin are combine in the same rational as set forth above with respect to claim 9 and analogous claim 19. Wang does not explicitly teach generating a second output inference, wherein generating the second output inference includes processing the sequence of observations using a recurrent neural network (RNN) and generating an overall output inference, wherein generating the overall output inference includes processing the first and second output inferences using a multilayer perceptron (MLP). Trentin teaches generating a second output inference, wherein generating the second output inference includes processing the sequence of observations using a recurrent neural network (RNN) (Trentin page 13 3.3. Integrating HMMs with RNNs para 4 line 8-22 and para 7, In the present research, an approximate version of equation (3) is used, by taking: that is to say, for each state of the HMM a network is trained to estimate the state posterior given the partial sensory vector sequence . This is accomplished using a supervised gradient-descent training algorithm (as described in the previous section) on a RNN for each one of the states. The training set Tj for the j th network is obtained by labeling the F th pattern of each training sequence xL1 with the target probability values of 1.0 or 0.0 according to the fact that the robot was, or was not, in state qj at time F during acquisition of sensory signal xz [generating a second output inference] . All the available training sequences X are used to train all the Q RNNs [wherein generating the second output inference includes processing the sequence of observations using a recurrent neural network (RNN)], with different distributions of the corresponding labels. This instantiates the concept of discriminative training, since models are trained on `positive’ as well as on `negative’ examples. As pointed out in Bourlard and Morgan (1994), this is no longer strictly an HMM, but the general framework is still valid and the standard decoding algorithms (e.g. Viterbi) can still be used.) ; However chu teaches teach and generating an overall output inference, wherein generating the overall output inference includes processing the first and second output inferences using a multilayer perceptron (MLP) (Chu para 0078, One or multiple CNNs could input to the recurrent neural network 1300 ( or be multiple RNN s, such as one per CNN), and the recurrent neural network 1300 outputs classification predictions based on how the CNN image classifications evolve over time. The CNN may have a modified output layer, such that the CNN does not predict the pass/fail, but sends a set of classification features as an input to the recurrent neural network 1300. The features may, for example, be the results of the second fully connected layer. If multiple RNN s are used ( e.g., one per CNN) [ wherein generating the overall output inference includes processing the first and second output inferences ], the analytic system may utilize a multilayer perceptron (a general purpose ANN) to combine the RNN predictions into a final prediction of failure mode [ and generating an overall output inference ]). Wang and Chu are considered to be analogous to the claim invention because they are in the same field of machine learning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Wang to incorporate the teachings of Chu to use a multilayer perceptron to combine the output of an RNN. Doing so to generate a single final prediction (Chu para 0078 line 18-26, The CNN may have a modified output layer, such that the CNN does not predict the pass/ fail, but sends a set of classification features as an input to the recurrent neural network 1300. The features may, for example, be the results of the second fully connected layer. If multiple RNN s are used ( e.g., one per CNN), the analytic system may utilize a multilayer perceptron (a general purpose ANN) to combine the RNN predictions into a final prediction of failure mode.). Pertinent Prior Art 07-96 AIA The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Lee et al. (US20110004936A1) – teaches a botnet detection system by using hybrid hidden Markov model as shown in Figure 1 using transition and emission parameter estimators. Yu, J., Liang, S., Tang, D. et al. A weighted hidden Markov model approach for continuous-state tool wear monitoring and tool life prediction. Int J Adv Manuf Technol 91 , 201–211 (2017) (“Yu”) – teaches a weighted hidden Markov model . Conclusion 07-40 AIA Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL . See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a). A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action. Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALFREDO CAMPOS whose telephone number is (571)272-4504. The examiner can normally be reached 7:00 - 4:00 pm M - F. 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, Michael J. Huntley can be reached at (303) 297-4307. 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. /ALFREDO CAMPOS/Examiner, Art Unit 2129 /MICHAEL J HUNTLEY/Supervisory Patent Examiner, Art Unit 2129 Application/Control Number: 18/175,750 Page 2 Art Unit: 2129 Application/Control Number: 18/175,750 Page 3 Art Unit: 2129 Application/Control Number: 18/175,750 Page 4 Art Unit: 2129 Application/Control Number: 18/175,750 Page 5 Art Unit: 2129 Application/Control Number: 18/175,750 Page 6 Art Unit: 2129 Application/Control Number: 18/175,750 Page 7 Art Unit: 2129 Application/Control Number: 18/175,750 Page 8 Art Unit: 2129 Application/Control Number: 18/175,750 Page 9 Art Unit: 2129 Application/Control Number: 18/175,750 Page 10 Art Unit: 2129 Application/Control Number: 18/175,750 Page 11 Art Unit: 2129 Application/Control Number: 18/175,750 Page 12 Art Unit: 2129 Application/Control Number: 18/175,750 Page 13 Art Unit: 2129 Application/Control Number: 18/175,750 Page 14 Art Unit: 2129 Application/Control Number: 18/175,750 Page 15 Art Unit: 2129 Application/Control Number: 18/175,750 Page 16 Art Unit: 2129 Application/Control Number: 18/175,750 Page 17 Art Unit: 2129 Application/Control Number: 18/175,750 Page 18 Art Unit: 2129 Application/Control Number: 18/175,750 Page 19 Art Unit: 2129 Application/Control Number: 18/175,750 Page 20 Art Unit: 2129 Application/Control Number: 18/175,750 Page 21 Art Unit: 2129 Application/Control Number: 18/175,750 Page 22 Art Unit: 2129 Application/Control Number: 18/175,750 Page 23 Art Unit: 2129 Application/Control Number: 18/175,750 Page 24 Art Unit: 2129 Application/Control Number: 18/175,750 Page 25 Art Unit: 2129 Application/Control Number: 18/175,750 Page 26 Art Unit: 2129 Application/Control Number: 18/175,750 Page 27 Art Unit: 2129 Application/Control Number: 18/175,750 Page 28 Art Unit: 2129 Application/Control Number: 18/175,750 Page 29 Art Unit: 2129 Application/Control Number: 18/175,750 Page 30 Art Unit: 2129 Application/Control Number: 18/175,750 Page 31 Art Unit: 2129 Application/Control Number: 18/175,750 Page 32 Art Unit: 2129 Application/Control Number: 18/175,750 Page 33 Art Unit: 2129 Application/Control Number: 18/175,750 Page 34 Art Unit: 2129 Application/Control Number: 18/175,750 Page 35 Art Unit: 2129
Prosecution Insights