DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . This action is made final.
This office action is in response to the amendments filed on February 23, 2026.
Claims 1, 5-7, and 9 have been amended. Claims 4, and 8 have been cancelled.
Response to Amendment
The amendments filed February 23, 2026 has been entered. Claims 1-3, 5-7 and 9 remain pending in the application.
Priority
Acknowledgment is made of applicant's claim for foreign priority based on an application filed in India on 05/01/2022. It is noted, however, that applicant has not filed a certified copy of the IN202221000662 application as required by 37 CFR 1.55.
Applicant is advised of possible benefits under 35 U.S.C. 119(a)-(d) and (f), wherein an application for patent filed in the United States may be entitled to claim priority to an application filed in a foreign country.
The Office attempted to electronically retrieve the priority document on December 04, 2023 but failed.
Response to Arguments
Response to 112 arguments
Applicant’s arguments, see page 13, filed February 23, 2026, with respect to claims 6, 7, and 8 (it is acknowledged that claim 8 has been cancelled) have been fully considered and are persuasive. The rejection of November 14, 2025 has been withdrawn.
Response to 101 arguments
Applicant’s arguments, see pages 13-30, regarding Step 2A Prong Two, filed February 23, 2026, with respect to claims 1-9 have been fully considered and are persuasive. The rejection of November 14, 2025 has been withdrawn.
Response to 103 arguments
Applicant’s arguments with respect to claim(s) 1-9 have been considered but are moot in view of the new grounds of rejection necessitated by the amendment.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claim(s) 1-3, 5-7, and 9 is/are rejected under 35 U.S.C. 103 as being unpatentable over Yoon et al. ("Time-series Generative Adversarial Networks", referred to as Yoon), in view of Xu et al. ("Modeling Tabular Data using Conditional GAN", referred to as Xu), in view of Zhang et al.(“Time-series regeneration with convolutional recurrent generative adversarial network for remaining useful life estimation”, referred to as Zhang), in view of Chang et al. (US 20230092969 A1, referred to as Chang), in view of Odena et al. ("Conditional Image Synthesis with Auxiliary Classifier GANs", referred to as Odena), in view of Raimbault (“Second-order control of complex systems with correlated synthetic data” referred to as Raimbault), in view of Bogle et al. (“A moment matching approach for generating synthetic data”, referred to as Bogle).
Regarding claim 1, Yoon teaches, a processor implemented method for generating mixed variable type multivariate temporal synthetic data, the method comprising (Yoon, Page 3, Section 3: Describes providing multivariate and temporal real-time sequence data as input to a generative adversarial network to learn temporal relationships between variables over time. The examiner notes that although Yoon does not explicitly say a processor, it is understood to one of ordinary skill in the art that Yoon’s methods are implemented in a computer environment, which contains processors, GPU’s, RAM, hard drives and other computer components to run.):
splitting, via the one or more hardware processors, the pre- processed data into a training dataset, a validation dataset and a test dataset (Yoon, Page 6-7 Section 5: Describes evaluating held-out test sets and discusses hyperparameters/benchmarks, reflecting conventional workflow where a validation set is sued for tuning before final testing.; Xu, Page 2-3, Section 3: Describes partitioning the dataset into training and test sets and evaluating performance on the held-out test split, corresponding a split.);
providing the training dataset as an input to the embedding neural network to generate high dimensional real latent temporal embeddings (Yoon, Page 4, Section 4.1: Describes an embedding network which maps the real sequence to latent temporal embeddings.),
providing the high dimensional real latent temporal embeddings as an input to the recovery neural network to get a reconstructed input training dataset,… wherein the embedding and the recovery neural network is jointly trained using a supervised learning approach for reconstructing the training dataset(Yoon, Page 4, Section 4.1: Describes that there is a recovery network that maps from latent back toward feature space and optimizes a reconstruction loss (supervised reconstruction) of the input sequence using embedding and recovery pairs (joint objective)),
providing the high dimensional real latent temporal embeddings as an input to the supervisor neural network to generate a single-step-ahead high dimensional real latent temporal embeddings, corresponding to a short term prediction for value of an embedding wherein the supervisor neural network is trained using the supervised learning approach (Yoon, Page 5-6, Section 4.3: Describes a one-step supervised loss in latent space used to synchronize latent dynamics of real vs. generated sequences (supervisor). It implements a supervised next step latent loss, corresponding to a substantive requirement (single step latent prediction trained from the supervised objective).; Section 1 and 4.3 : Describes comparing the actual next step laten vector form the embedding function with the synthetic next step latent vector generated from the historical latent sequence.),
providing the high dimensional synthetic latent temporal embeddings to the trained supervisor neural network to predict single- step ahead synthetic temporal latent embeddings (Yoon, Page 5-6, Section 4.3: Describes implementing supervised, single-step latent objectives that takes latent sequences and predicts the next latent step, i.e., the supervisor function. This is applied on latent embeddings and synthetic sequences, which correspond to latent dynamics of both real and synthetic which are synchronized.),
… by retaining a conditional temporal dynamics of the input data in the mixed variable type multivariate temporal synthetic data (Yoon Abstract and Introduction : Describes retaining conditional temporal dynamics of input time series data and generated synthetic time series data because it teaches that a time series generative model should preserve temporal dynamics such that it generated sequences of respect relationships between variables across time, and further teaches using a stepwise supervised loss to capture a stepwise conditional distribution/temporal transitions in the training data. It retains temporal dynamics of the input data in generated multivariate temporal synthetic data by preserving relationships between variables over time and using a supervised stepwise loss to capture temporal transitions.);
Although Yoon teaches, a processor implemented method for generating mixed variable type multivariate temporal synthetic data; splitting, via the one or more hardware processors, the pre- processed data into a training dataset, a validation dataset and a test dataset; providing the training dataset as an input to the embedding neural network to generate high dimensional real latent temporal embeddings; providing the high dimensional real latent temporal embeddings as an input to the recovery neural network to get a reconstructed input training dataset,… wherein the embedding and the recovery neural network is jointly trained using a supervised learning approach for reconstructing the training dataset; providing the high dimensional real latent temporal embeddings as an input to the supervisor neural network to generate a single-step-ahead high dimensional real latent temporal embeddings, corresponding to a short term prediction for value of an embedding wherein the supervisor neural network is trained using the supervised learning approach; providing the high dimensional synthetic latent temporal embeddings to the trained supervisor neural network to predict single- step ahead synthetic temporal latent embeddings; and retaining a conditional temporal dynamics of the input data in the mixed variable type multivariate temporal synthetic data.
It does not teach providing, via one or more hardware processors, mixed variable type multivariate temporal real time data… as an input data, wherein the mixed variable type comprises continuous variables and discrete variables; pre-processing, via the one or more hardware processors, the input data by scaling to a fixed range for both the continuous variables and the discrete variables; determining, via the one or more hardware processors, a cluster label dependent random noise by transforming a Gaussian random noise with fixed predetermined cluster labels, wherein the Gaussian random noise is part of the input data; computing, via the one or more hardware processors, a conditioned knowledge vector corresponding to a pre-determined label value for each discrete variable; concatenating, via the one or more hardware processors, the cluster label dependent random noise with the conditioned knowledge vector to generate a condition aware synthetic noise; providing a condition aware synthetic noise as an input to the sequence generator neural network to get high dimensional synthetic latent temporal embeddings; and providing the predicted single-step ahead synthetic temporal latent embeddings as an input to the recovery neural network to generate the mixed variable type multivariate temporal synthetic data.
Xu teaches providing, via one or more hardware processors, mixed variable type multivariate temporal real time data… as an input data, wherein the mixed variable type comprises continuous variables and discrete variables (Yoon, Page 3, Section 3: Describes providing multivariate and temporal real-time sequence data as input to a generative adversarial network to learn temporal relationships between variables over time. ;Xu, Page 2-3 Section 3: Describes mixed variable type tabular data containing both continuous and discrete variables to a conditional GAN, addressing mixed data distributions. Xu supplies the condition aware/mixed variable aspects. Thus, Yoon in view of Xu teaches retaining temporal dynamics while generating mixed variable, condition aware temporal synthetic data.), …
pre-processing, via the one or more hardware processors, the input data by scaling to a fixed range for both the continuous variables and the discrete variables (Xu, Page 3, Section 4.2:Describes column-wise preprocessing by mode-specific normalization that scales continuous columns to fixed ranges suitable for the generator and conditional/one-hot vectors or embeddings for discrete columns. This corresponds to a fixed-range scaling/encoding. );
determining, via the one or more hardware processors, a cluster label dependent random noise by transforming a Gaussian random noise with fixed predetermined cluster labels, wherein the Gaussian random noise is part of the input data (Xu, Page 4, Section 4.3Describes combining Gaussian noise z with a fixed, predetermined label encoding (the conditional vector that selects a category in a discrete column). The built conditional for the selected label uses z N(0,1) as the noise input, with the concatenation notation, the noise becomes label conditioned (transformed by the fixed label) before entering the generator. Which describes preprocessing mixed type data by applying mode specific normalization to continuous columns and representing discrete columns using conditional vectors, one hot vectors, or embeddings. Thus, teaching transforming both continuous variables and discrete variables into fixed numerical representations suitable for GAN training, which corresponds to scaling encoding the input data to a fixed range for use by the generator);
computing, via the one or more hardware processors, a conditioned knowledge vector corresponding to a pre-determined label value for each discrete variable (Xu, Page 5, Section 4.3 Conditional vector: Describes a conditional knowledge vector built per discrete variable, for each discrete column Di, a mask/one-hot is created which selects a pre-determined label value k. These masks are then concatenated across all discrete variables to form a conditional.);
concatenating, via the one or more hardware processors, the cluster label dependent random noise with the conditioned knowledge vector to generate a condition aware synthetic noise (Xu, Page 4-5, Section 4.3: Describes how a fixed label indicator is constructed for the chosen discrete value and uses Gaussian noise. The concatenation is generated form input by concatenating z with the fixed label indicator, producing a label/condition-aware noise vector before generation. This corresponds to concatenating cluster label dependent random noise with conditioned knowledge vectors to create synthetic noise.);
providing a condition aware synthetic noise as an input to the sequence generator neural network to get high dimensional synthetic latent temporal embeddings (Xu Page 5, Section 4.3: Describes a condition-aware input built form concatenating Gaussian noise with a fixed label vector. Yoon, Page 4, Section 4.2: Describes a generator which maps nose to latent space, producing synthetic latent codes. These on combination work together to generate a condition aware synthetic noise, generated for high dimensional synthetic latent embeddings.),
providing the predicted single-step ahead synthetic temporal latent embeddings as an input to the recovery neural network to generate the mixed variable type multivariate temporal synthetic data (Yoon, Page 4, Section 4.1: Describes a recovery network which turns latent embeddings (including next-step/supervised-predicted ones) into feature space sequences. ;Xu, Page 6, Section 4.4: Describes how mixed-type variables (continuous scalars and discrete values) are produced in one synthetic sample. ; Together these correspond to latent recovered sequences with mixed types.);
It would have been obvious to one of ordinary skill In the art at the time of the claimed invention to combine Yoon’s time-series GAN framework with Xu’s conditional mixed-type data handling. Doing so would enable the system to generate realistic temporal data including continuous and discrete attributes to stabilize training and covering multiple input data types for better model generation, improving training stability and making more efficient use of model/computing capacity.
Although Yoon, in view of Xu teaches, providing, via one or more hardware processors, mixed variable type multivariate temporal real time data… as an input data, wherein the mixed variable type comprises continuous variables and discrete variables; pre-processing, via the one or more hardware processors, the input data by scaling to a fixed range for both the continuous variables and the discrete variables; determining, via the one or more hardware processors, a cluster label dependent random noise by transforming a Gaussian random noise with fixed predetermined cluster labels, wherein the Gaussian random noise is part of the input data; computing, via the one or more hardware processors, a conditioned knowledge vector corresponding to a pre-determined label value for each discrete variable; concatenating, via the one or more hardware processors, the cluster label dependent random noise with the conditioned knowledge vector to generate a condition aware synthetic noise; providing a condition aware synthetic noise as an input to the sequence generator neural network to get high dimensional synthetic latent temporal embeddings; and providing the predicted single-step ahead synthetic temporal latent embeddings as an input to the recovery neural network to generate the mixed variable type multivariate temporal synthetic data.
They do not teach, related to functioning and operation of industrial assets.. , wherein the continuous variables comprise independent and dependent variables
Zhang teaches, related to functioning and operation of industrial assets(Pages 6820-6822, Abstract and Introduction: Describes time series data related to functioning and operation of industrial assets, the framework generates and uses run to failure multivariate time series for RUL estimation of industrial equipment/components, including aero-engine system and Li-ion battery systems.).. , wherein the continuous variables comprise independent and dependent variables (Pages 6821-6822 Introduction and Section II: Describes using measured sensor/process variables as input variables and developing a regression relationship between those variables and a concerned performance indicator/RUL. The measured sensor/process variables correspond to independent variables, and the performance indicator/RUL correspond to a dependent/target variable.);
It would have been obvious to one of ordinary skill In the art at the time of the claimed invention to combine Yoon in view of, Xu with Zhang’s RUL estimation for industrial equipment. Doing so would enable the system to generate realistic multivariate industrial degradation, increasing available training data and reducing training cost and time.
Yoon in view of Xu, in view of Zhang teaches, by mapping low dimensional temporal sequences to high dimensional temporal latent variables using an autoregressive neural-network model realized with an unidirectional recurrent neural network with an extended memory (Yoon page 4, Section 4.1 Describes mapping temporal sequences to temporal latent variables because the embedding function maps temporal features to latent codes/latent vector spaces using a recurrent embedding network. The embedding/recovery functions are autoregressive and obey casual ordering, such that outputs at each step depend only on preceding information.; Zhang pages 6824-6826 Section III. C.: Describes using an LSTM type recurrent network which implements a CRGAN using LSTM layers, where the LSTM layers learn temporal correlations between past samples and future samples. ),
… by transforming the high dimensional temporal latent variables to the low dimensional temporal sequences using an autoregressive, casual-ordering driven neural-network model, realized with the unidirectional recurrent neural network with the extended memory (Yoon Page 4, Section 4.1: Describes transforming temporal latent variables to temporal sequences because the recovery function takes static and temporal latent codes Back to their feature representations, including ~xt = rX (ht). It further teaches that the embedding and recovery functions are auto regressive and obey casual ordering, such that outputs at each step can only depend on preceding information.; Zhang pages 6824-6826 Section III. C.: Teaches using LSTM layers in ACRGAN to learn temporal correlations between past samples and future samples corresponding to unidirectional recurrent neural network with extended memory),
…the autoregressive neural-network model implemented with the unidirectional recurrent neural network with the extended memory (As taught above), and wherein the critic neural network is initially trained on the input data to map the independent variables to the target variable to minimize a supervised loss in prediction of the target variable (Xu Pages 4-6 Sections 4.3, and 4.4: Teaches a CTGAN having a critic neural network; Zhang Page 6822 II.A : Describes implementing recurrent temporal modeling using LSTM layers in a CR GAN. It maps independent/input sensor variables to a target/dependent performance variable since the RUL estimation by learning a regression relationship between measured features/sensor data and degrading performance/RUL.)and to preserve temporal dynamics between the input data and the target variable (Yoon Pages 4-5, Sections 4.1 and 4.2 : Teaches an autoregressive recurrent neural network model for temporal sequence generation, including supervised loss for learning stepwise temporal transitions in a latent embedding space. Zhang II.A III.C further teaches implementing GAN based time series generation with LSTM layers to learn temporal correlations in multivariate run to failure data. It uses measures sensor/process variables as input variables to predict a target RUL/performance variable by minimizing prediction error, thereby learning the temporal degradation relationship between input data and the target variable.)
wherein the generated mixed variable type multivariate temporal synthetic data of the industrial assets and an equipment-level sensory information, incorporates condition and constraint aware knowledge, and serves as a virtual simulation capturing a process level data and an equipment level data in deep learning models to aid in prognostics and health management of the industrial assets (Zhang Pages 6820-6822, and 6824-6825, Abstract, Introduction and Sections II, III.C : Describes generated multivariate temporal synthetic data for industrial assets because it is directed to RUL estimation for industrial equipment/components and proposes generating realistic like multivariate time series using CRGAN to enhance RUL methods. It further teaches that industrial time series are multivariate and include measured process/sensor variables over time. It’s conditional aware generation by using a conditioned vector Y and feeding condition information into the generator and discriminator. It also shows that modeling degradation nature can construct a realistic simulation to mimic it, and that generated degradation trajectories are mixed with real run to failure sequences to train RUL estimation models, thereby aiding health prognostic slash condition based maintenance of industrial equipment.);
Although Yoon in view of Xu, in view of Zhang teaches providing, via one or more hardware processors, mixed variable type multivariate temporal real time data related to functioning and operation of industrial assets as an input data, wherein the mixed variable type comprises continuous variables and discrete variables, wherein the continuous variables comprise independent variables and dependent variables; providing the training dataset as an input to the embedding neural network to generate high dimensional real latent temporal embeddings, by mapping low dimensional temporal sequences to high dimensional temporal latent variables using an autoregressive neural-network model realized with an unidirectional recurrent neural network with an extended memory; providing the high dimensional real latent temporal embeddings as an input to the recovery neural network to get a reconstructed input training dataset by transforming the high dimensional temporal latent variables to the low dimensional temporal sequences using an autoregressive, casual-ordering driven neural-network model, realized with the unidirectional recurrent neural network with the extended memory, wherein the embedding and the recovery neural network is jointly trained using a supervised learning approach for reconstructing the training dataset; the autoregressive neural-network model implemented with the unidirectional recurrent neural network with the extended memory, and wherein the critic neural network is initially trained on the input data to map the independent variables to the target variable to minimize a supervised loss in prediction of the target variable and to preserve temporal dynamics between the input data and the target variable; and wherein the generated mixed variable type multivariate temporal synthetic data of the industrial assets and an equipment-level sensory information, incorporates condition and constraint aware knowledge, and serves as a virtual simulation capturing a process level data and an equipment level data in deep learning models to aid in prognostics and health management of the industrial assets. They do not teach training, via the one or more hardware processors, a joint neural network of an autoencoding-decoding component of a Constraint- Condition-Generative Adversarial Network (ccGAN), a supervisor neural network and a critic neural network utilizing the training dataset, wherein the autoencoding-decoding component comprises an embedding neural network and a recovery neural network, the training comprises; providing the high dimensional real latent temporal embeddings as an input to the critic neural network to predict a target variable, wherein the critic neural network is trained using the supervised learning approach; and providing the high dimensional synthetic latent temporal embeddings to the trained critic neural network to predict the synthetic target variable.
Chang teaches, training, via the one or more hardware processors, a joint neural network of an autoencoding-decoding component of a Constraint- Condition-Generative Adversarial Network (ccGAN), a supervisor neural network and a critic neural network utilizing the training dataset, wherein the autoencoding-decoding component comprises an embedding neural network and a recovery neural network, the training comprises (Yoon, Page 3-4 Section 4: Describes autoencoding-decoding (embedding and recovery), joint training with adversarial parts, and adding a supervised (single-step) objective. ;Xu, Page 4-5, Section 4.3: Describes conditional/constraint (ccGAN) aspect, a conditional generator, condition vector, and generator/critic training setup.; Chang [0069]: Describes a predictive head/model operating on embeddings to output a target/candidate output.):
providing the high dimensional real latent temporal embeddings as an input to the critic neural network to predict a target variable, wherein the critic neural network is trained using the supervised learning approach(Chang, [0051-0058]: Describes a machine learning system that includes both an embedding model and a prediction model. The embedding model producers numerical embedding vectors from the input data, and those embeddings are supplied to the prediction model, which is trained to predict a candidate output (target variable) such as a class or value for each data item.; [0062]: Describes that once the embedding model is trained, the prediction model is trained on labeled training data to learn the mapping form embeddings to target outputs. These correspond to critic network trained to predict a target variable under supervised learning.);
providing the high dimensional synthetic latent temporal embeddings to the trained critic neural network to predict the synthetic target variable (Chang, [0051-0058]: Describes a separate neural network which takes embeddings as input and predicts a labeled target under supervised learning.)
It would have been obvious to one of ordinary skill In the art at the time of the claimed invention to combine Yoon in view of Xu, in view of Zhang, with Chang’s trained supervised model. Doing so would enable the system to enhance the network’s ability to predict score target variables within the same adversarial training environment.
Although Yoon in view of Xu, in view of Zhang, in view of Chang teaches training, via the one or more hardware processors, a joint neural network of an autoencoding-decoding component of a Constraint- Condition-Generative Adversarial Network (ccGAN), a supervisor neural network and a critic neural network utilizing the training dataset, wherein the autoencoding-decoding component comprises an embedding neural network and a recovery neural network, the training comprises; providing the high dimensional real latent temporal embeddings as an input to the critic neural network to predict a target variable, wherein the critic neural network is trained using the supervised learning approach; and providing the high dimensional synthetic latent temporal embeddings to the trained critic neural network to predict the synthetic target variable. They do not teach, providing, via the one or more hardware processors, the high dimensional real latent temporal embeddings and the high dimensional synthetic latent temporal embeddings as an input to the sequence discriminator neural network to classify them as one of a real or a fake, and predict cluster labels for synthetic data; and providing the validation dataset as an input to the trained ccGAN to tune a set of hyperparameters and drives a model selection to avoid over-fitting of a ccGAN architecture
Odena teaches, providing, via the one or more hardware processors, the high dimensional real latent temporal embeddings and the high dimensional synthetic latent temporal embeddings as an input to the sequence discriminator neural network to classify them as one of a real or a fake, and predict cluster labels for synthetic data (Odena, Page 3, Section 3: Describes a discriminator which preforms multi-tasks, by outputting real/fake and a class/label prediction for the same input.);
providing the validation dataset as an input to the trained ccGAN to tune a set of hyperparameters and drives a model selection to avoid over-fitting of a ccGAN architecture (Odena 4.4 4.5 Appendix A: Describes using evaluation scores to select the best ACGAN model and performing a grid search across 27 hyperparameter configurations. It searches for signatures of overfitting in the ACGAN by checking whether the network memorized training data. Thus, corresponding to hyperparameter tuning/ model selection for an ACGAN architecture and evaluation to avoid or detect overfitting.).
It would have been obvious to one of ordinary skill In the art at the time of the claimed invention to combine Yoon in view of Xu, in view of Zhang, in view of Chang with Odena’s discriminator. Doing so would enable the system to improve functionality by distinguishing real data form synthetic sequences, enhancing the correctness of each generated sequence’s associated condition or cluster label. This would ensure a more stable training and label consistency within the conational time-series GAN framework.
Although Yoon in view of Xu, in view of Zhang, in view of Chang, in view of Odena teaches, providing, via the one or more hardware processors, the high dimensional real latent temporal embeddings and the high dimensional synthetic latent temporal embeddings as an input to the sequence discriminator neural network to classify them as one of a real or a fake, and predict cluster labels for synthetic data; and providing the validation dataset as an input to the trained ccGAN to tune a set of hyperparameters and drives a model selection to avoid over-fitting of a ccGAN architecture. They do not teach, extracting relationships in the input data by matching a first-order moments and second-order moments of the input data and the mixed variable type multivariate temporal synthetic data in absence of the pre-determined cluster labels, using an unsupervised learning approach
Raimbault teaches, extracting relationships in the input data(Raimbault Page 2 Synthetic data and dependency structures and Synthetic data and socio-spatial systems: Teaches extracting relationships in data because second order structure corresponds to controlling covariance structure between generated variables.)
It would have been obvious to one of ordinary skill In the art at the time of the claimed invention to combine Yoon in view of Xu, in view of Zhang, in view of Chang, in view of Odena’s with Raimbault’s relationship extraction teachings. Doing so would enable the system to identify relationships and dependencies within the input data to Beter preserve underlying structure of the original data.
Although Yoon in view of Xu, in view of Zhang, in view of Chang, in view of Odena, in view of Raimbault teaches, extracting relationships in the input data. They do not teach, matching a first-order moments and second-order moments of the input data and the mixed variable type multivariate temporal synthetic data in absence of the pre-determined cluster labels, using an unsupervised learning approach.
Bogle teaches, matching a first-order moments and second-order moments of the input data and the mixed variable type multivariate temporal synthetic data in absence of the pre-determined cluster labels, using an unsupervised learning approach(Bogle Pages 1-2, Introduction and Usage : Describes generating synthetic data by matching statistical marginal and mixed raw moments of real data to a user specified moment order, including moment order two, which corresponds to matching first order and second order moments. It further teaches validating that interactions between variables are captured by comparing covariance matrices of original and synthetic data. The Bowman matching approach does not require predetermined cluster labels, but instead generate synthetic observations by matching statistical moments of the original data.).
It would have been obvious to one of ordinary skill In the art at the time of the claimed invention to combine Yoon in view of Xu, in view of Zhang, in view of Chang, in view of Odena’s in view of Raimbault, with Bogle’s statistical moments of the input data. Doing so would enable the system to preserve statistical properties of the original input data, improving statistical fidelity and representativeness of the generated mixed variable multivariate temporal synthetic data.
The examiner notes that the following limitations do not introduce new structure or functionality from previous limitations, and is recited to describe a training phase. These limitations are rejected thusly as disclosed in the limitations above.
jointly training, via the one or more hardware processors, adversarial neural networks of the Constraint-Condition aware Generative Adversarial Network (ccGAN), a sequence generator neural network, a sequence discriminator neural network, the supervisor neural network and the critic neural network utilizing the condition aware synthetic noise, wherein the training comprises (The examiner notes that this claim is directed to combining the previous limitations, and is taught by Yoon in view of Xu, in view of Zhang, in view of Chang, in view of Odena as disclosed above):
providing, via the one or more hardware processors, a real world condition aware synthetic noise as an input to the trained sequence generator neural network to get real world high dimensional synthetic latent temporal embeddings (The examiner notes that Xu teaches condition aware noise and Yoon teaches a generator producing latent codes at generation time, as discussed above. This limitation is merely the deployment of the trained pieces as described above.);
providing, via the one or more hardware processors, the real world high dimensional synthetic latent temporal embeddings to the trained supervisor neural network to predict real world single-step ahead synthetic temporal latent embeddings(The examiner notes that Yoon teaches a supervised one-step latent predictor, used on the latent sequence at inference, as discussed above. This limitation is merely the deployment of the trained pieces as described above.); and
providing, via the one or more hardware processors, the real world predicted single-step ahead synthetic temporal latent embeddings as an input to the trained recovery neural network to generate the mixed variable type multivariate temporal synthetic data(The examiner notes that Xu teaches mixed-type rendering and Yoon teaches recovery of latent to feature sequences, as discussed above. This limitation is merely the deployment of the trained pieces as described above.).
Regarding claim 2, Yoon further teaches minimize the discrepancy between the real input temporal data and the mixed variable type multivariate temporal synthetic data using the embedding neural network and the recovery neural network modules (Page 5, Section 4-4.3 : Describes embedding and recovery functions trained jointly with adversarial components and a supervised objective, reconstruction that reduces the difference between original and reconstructed sequences which correspond to minimizing discrepancies.).
Regarding claim 3, Xu further teaches, wherein a conditioned knowledge vector is configured to incorporate the condition into the Constraint-Conditional Generative Adversarial Network (ccGAN) framework(Page 4-5, Section 4.3 : Describes a constructed conditional vector that encodes the chosen category/label and concatenates it with Gaussian noise for the generator input, conditioning is enforced in the generator loss and evaluated by the critic.).
Regarding claim 4, Cancelled
Regarding claims 5-7, which recites substantially the same limitations as claims 1-3. Claims 6-7 further recites a system… comprises: an input/output interface; one or more hardware processors; and a memory (It is understood that the methods used in Yoon, Xu and Odena are implemented on computing systems comprising computer hardware components. Chang [0035]: Describes computer hardware components to run the methods.) to perform the method steps of claims 1-4, respectively, and are therefore rejected on the same premise.
Regarding claim 8, Cancelled
Regarding claim 9, which recites substantially the same limitations as claims 1. Claim 9 further recites a non-transitory machine-readable information storage mediums (It is understood that the methods used in Yoon, Xu and Odena are implemented on computing systems comprising computer hardware components. Chang [0080]: Describes a non-transitory computer useable medium to run the methods.) to perform the method steps of claim 1, respectively, and is therefore rejected on the same premise.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DONALD T RODEN whose telephone number is (571)272-6441. The examiner can normally be reached Mon-Thur 8:00-5:00 EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Omar Fernandez Rivas can be reached at (571) 272-2589. 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.
/D.T.R./Examiner, Art Unit 2128
/OMAR F FERNANDEZ RIVAS/Supervisory Patent Examiner, Art Unit 2128
Prosecution Insights