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 .
Response to Amendment
This Final Rejection is filed in response to Applicant Arguments/Remarks Made in an Amendment filed 09/01/2025.
Claims 1, 5, 9, and 16 are amended.
Claims 1-20 remain pending.
Response to Arguments
Argument 1, Applicant argues in Applicant Arguments/Remarks Made in an Amendment filed 09/01/2025 on pg. 8-11, that prior art Hedge fails to teach the primary claim limitations, “training, by a training engine associated with the one or more processors, a machine learning model using the plurality of mixed input data, wherein training the machine learning model comprises:
generating, using an encoding laver of the training engine, a first encoding for the one or more continuous variables as a multi-dimensional vector, wherein the multi-dimensional vector has a dimension corresponding to a number of the one or more continuous variables;
generating, using the encoding laver, a second encoding for each of the one or more categorical variables, wherein the second encoding is a one-hot encoding for each of the one or more categorical variables;
combining the first and the second encodings with a dense laver of the training engine, wherein the dense layer is configured to determine a mean and a standard deviation across the first encoding for the one or more continuous variables and the second encoding for each of the one or more categorical variables”
Response to Argument 1, applicant’s arguments have been considered, however in light of the amendments a newly found combination of prior art (U.S. Patent Application Publication NO. 20210103822 “Hedge”, in light of U.S. Patent Application Publication NO. 20200034436 “Chen”, and further in light of U.S. Patent Application Publication NO. 20200026257 “Dalal”) is applied to updated rejections.
The examiner notes that Hedge teaches in para. [0032], “inputting a database including real structures and real properties to an encoder network as an input and compressing the input to generate encoded vectors, mean and standard deviation of a distribution of the encoded vectors in a latent space”. Wherein its noted that specifically a suitable loss function is applied to respective continuous and categorical variables being learned and thus Hedge teaches the BRI for the primary claim limitation of, “herein the dense layer is configured to determine a mean and a standard deviation across the first encoding for the one or more continuous variables and the second encoding for each of the one or more categorical variables”, as the examiner notes that each attribute of an event is represented by one or more vectors and that the if an event consists of both continuous and categorical attributes, then a mean and standard deviation would be found across the plurality of dimensions that represent such an event.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
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.
Claim(s) 1-5, 7, 9-11, 13, 15-17, & 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication NO. 20210103822 “Hedge”, in light of U.S. Patent Application Publication NO. 20200034436 “Chen”, and further in light of U.S. Patent Application Publication NO. 20200026257 “Dalal”.
Claim 1:
Hedge teaches a computer-implemented method comprising: obtaining, by one or more processors (i.e. para. [0049], The various components of these devices may be a process or thread, running on one or more processors, in one or more computing devices), a plurality of mixed input data (i.e. para. [0006], “A list of candidate materials (M) may then be gathered in act 102 based on prior knowledge and/or chemical intuition based on similarity with known materials having relevant values of property P”, wherein the BRI for mixed input data encompasses data with at least two variables, such as a structure and a relevant property), wherein at least some of the plurality of mixed input data include one or more categorical variables and one or more continuous variables (i.e. para. [0079], “According to some example embodiments, the system and method may be applied to categorical, ordinal, integer and/or count properties in addition to continuous properties”, wherein it is noted that the pair of variables may be categorical and continuous property variables); and training, by using a training engine associated with the one or more processors (i.e. para. [0067], “FIG. 2 is a schematic illustration of a system and a training process for a generative adversarial network (GAN) according to an embodiment of the present disclosure. The system includes a latent space vector generator (V) 210, a generator network (G) 230, and a discriminator network (D) 250”, wherein the BRI for a training engine encompasses the software executed on the processors that is programmed to perform the training process), a machine learning model using the plurality of mixed input data (i.e. para. [0053], “the term “joint probability distribution p(S, P)” as used herein refers to the probability distribution representing the structure and property relationships, where both the structure S and the target property P, as two separate events”, wherein the BRI for mixed data encompasses data comprising two separate, but related, variables with different properties, such as structure and property variables), wherein training the machine learning model comprises: generating, using an encoding layer of the training engine (i.e. para. [0091], the encoder network is composed of Convolutional Layers, which compress the input and outputs a dense representation of the input, called encodings), a first encoding for the one or more continuous variables as a multi-dimensional vector, wherein the multi-dimensional vector has a dimension (i.e. para. [0070], “latent space refers to an abstract multi-dimensional space containing feature values (i.e., vectors) that encodes a meaningful internal representation of externally observed events (such as the structure and the property of materials)... Each attribute of the event is represented by one or more vectors in the latent space, and an event may be represented with a plurality of dimensions in the latent space”, wherein the continuous property attribute variables may be encoded as multi-dimensional vectors);
generating, using the encoding layer, a second encoding for each of the categorical variables (i.e. para. [0079-0080], “the system and method may be applied to categorical, ordinal, integer and/or count properties in addition to continuous properties…According to an embodiment of the present disclosure, the target property may be any combination of two or more desirable physical quantities, such as resistivity, density of states, etc., while the structure may be an encoded vectorized representation in which elements of the structure vector represent atomic species in specific locations”, where a first and second encoding may be generated for a each of the continuous and categorical property variables),
combining the first and the second encodings with a dense layer of the training engine (i.e. para. [0091-0092], “the encoder network is composed of Convolutional Layers, which compress the input and outputs a dense representation of the input, called encodings… both the real structure and the real properties are fed to the encoder network”, wherein encodings for the structure and property pair are fed into the encoder network), wherein the dense layer is configured to determine a mean and a standard deviation across the first encoding for the one or more continuous variables and the second encoding for each of the one or more categorical variables (i.e. para. [0078, 0092], “The loss function may be any suitable type corresponding to the quantity being learned. For instance, if a continuous variable is being learned, the loss function may be a quantity such as root mean squared error. If a categorical (e.g., qualitative) variable is being learned, the loss may be cross-entropy, Kullback-Leibler divergence or similar measures… The encoder network transforms the input as a vector distribution over the latent space, and learns the mean and standard deviation of the probability distribution”, wherein a loss function learns a mean and standard deviator for any continuous and categorical values being learned in order to regularize the organization of the latent space) ; sampling at least some of the combined encodings (i.e. para. [0095], The encoded vector from the latent distribution is sampled by the decoder network 750 and transformed to a decoded output through the decoding process); obtaining reconstructed data by processing the sampled combined encodings using a decoder (i.e. para. [0095], During the generating process, a point from the latent space (as a sampled vector) 730 is selected and decoded by the decoder network 750 to produce a newly generated structure/property pair); and determining a loss of the reconstructed data (i.e. para. [0074], For instance, if a continuous variable is being learned, the loss function may be a quantity such as root mean squared error. If a categorical (e.g., qualitative) variable is being learned, the loss may be cross-entropy, Kullback-Leibler divergence or similar measures) from the plurality of mixed input data, wherein the loss includes a divergence error and a reconstruction error (i.e. para. [0093] “The loss function is composed of a “reconstruction term” (on the final layer), that tends to make the encoding-decoding scheme as performant as possible, and a “regularization term” (on the latent layer), that tends to regularize the organization of the latent space by making the distributions returned by the encoder close to a standard normal distribution”, wherein the BRI for a divergence error encompasses a regularization term” and the BRI for a reconstruction error encompasses a reconstruction term number that defines the distance between an original and reconstructed data, both of which comprise the loss function); and generating, by the one or more processors running the trained machine learning model, a plurality of mixed synthetic data (i.e. para. [0094], “the learned joint distribution probability, mean and standard deviation are utilized to provide inputs from the latent space to be decoded by the decoder to generate (partake) new structure-property pairs that have the structures suitable to provide the target properties”, wherein the output mixed pair data shares statistical probability values similar and suitable to the input joint property pairs).wherein the plurality of mixed synthetic data
(i) includes one or more categorical variables and one or more continuous variables (i.e. para. [0075], “the structure and/or property may include or consist of ordinal or count data. The discriminator network (D) may output a floating point (e.g., continuous floating point) or discrete count value instead of binary (True/False, 0/1) values”, wherein the generated sample data may include the include categorical binary data and continuous data) and
(ii) shares statistical properties with the plurality of mixed input data (i.e. para. [0077], When the GAN is trained to a point where its simulated structure-property pairs are statistically indistinguishable from the real structure-property pairs, the training is completed).
While Hedge teaches encoding an encoding for each attribute of an event and thus teaches generating a first encoding for a continuous variable attribute that may be represented by a plurality of dimensions and thus be a multidimensional vector having a dimension, Hedge may not explicitly teach that the multidimension vector representative of a continuous variable has a dimension
corresponding to a number of the one or more continuous variables.
However, Chen teaches
a first encoding for the one or more (i.e. para. [0064], “The projection layer 214 outputs an N-dimensional vector 302 (e.g., encoder output vector) from the encoder 205. The projection layer 214 may provide the N-dimensional vector 302 to have a dimension that equals the input dimension of the decoder 207. In some implementations, the operation of the attention network may adjust the dimensionality by producing a vector that has more or fewer values than the output of the projection layer 214”, wherein an encoder may generate a multi-dimensional encoding vector in that an N-dimensional vector may have N dimensions equal to the number of input variables in an input dimension).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to add that multidimension vector representative of a variable has a dimension corresponding to a number of the one or more variables, to the mixed synthetic data generation in which continuous variables have encodings generated with a random number of dimensions by a vector generator of Hedge, with a how categorical variables may specifically be encoding using a one hot encoding technique, as taught by Chen. One would have been motivated to combine the dimensionality preservation of input dimensions during feature encoding of Chen with the mixed synthetic data generation of that has a generator that generates a desired dimensionality for each encoding, where an encoding is generated for each variable including continuous variables of Hedge in order to prevent the loss of potentially important information that might occur during dimensionality reduction and ensures consistency across training and prediction datasets, preventing mismatches and potentially improving model performance.
While Hedge-Chen teaches encoding each attribute which includes encoding categorical attribute properties, Hedge may not explicitly teach that that categorical values are second encodings,
wherein the second encoding is a one-hot encoding for each of the one or more categorical variables.
However, Dalal teaches
wherein the second encoding is a one-hot encoding for each of the one or more categorical variables (i.e. para. [0049], The FFV 306 processes the output from the fault identifier 304 to be one hot encoded in one example. One hot encoding is a technique used to encode categorical features into binary vectors that enable machine learning (ML) algorithms to better predict the outcomes).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to add wherein the second encoding is a one-hot encoding for each of the one or more categorical variables, to the mixed synthetic data generation in which categorical variables are encoded during training of Hedge, with a how categorical variables may specifically be encoding using a one hot encoding technique, as taught by Dalal. One would have been motivated to combine the one hot encoding of categorical variables of Dalal with the mixed synthetic data generation of Hedge-Chen in order to augment the accuracy of machine learning (ML) algorithms to better predict the outcome as one hot encoding can be used to classify the best resolution step.
Claim 2:
Hedge, Chen, and Dalal teach the computer-implemented method of claim 1.
Hedge further teaches wherein training the machine learning model uses one or more variational autoencoders (i.e. para. [0029], a machine learning system to generate a structure (S) for a target property (P) includes a variational autoencoder (VAE), wherein the VAE includes an encoder network, a latent space, and a decoder network).
Claim 3:
Hedge, Chen, and Dalal teach the computer-implemented method of claim 1.
Hedge further teaches wherein the loss is a weighted sum of the divergence error and the reconstruction error, wherein weights combining the divergence and the reconstruction errors are determined heuristically (i.e. para. [0092], both the real structure and the real properties are fed to the encoder network. The encoder network transforms the input as a vector distribution over the latent space, and learns the mean and standard deviation of the probability distribution… The difference between the restructured version of the input and the actual input is backpropagated as the loss function to adjust the weights of the neural networks of the encoder and decoder in an iterative optimization process).
Claim 4:
Hedge, Chen, and Dalal teach the computer-implemented method of claim 3.
Hedge further teaches wherein the divergence error is a Kullback-Leibler divergence of the mean and the standard deviation (i.e. para. [0074], the loss function may be any suitable type corresponding to the quantity being learned... the loss function may be a quantity such as root mean squared error... the loss may be cross-entropy, Kullback-Leibler divergence or similar measure) .
Claim 5:
Hedge, Chen, and Dalal teach the computer-implemented method of claim 3.
Hedge further teaches
wherein the reconstruction error is a combination of a mean absolute error for the one or more continuous variables (i.e. para. [0074], the loss function may be any suitable type corresponding to the quantity being learned … if a continuous variable is being learned, the loss function may be a quantity such as root mean squared error) and a categorical cross entropy for the one or more categorical variables (i.e. para. [0074], “a categorical (e.g., qualitative) variable is being learned, the loss may be cross-entropy, Kullback-Leibler divergence or similar measures”, wherein reconstruction term is a combination of the suitable loss functions for respective continuous and categorical variables).
Claim 7:
Hedge, Chen, and Dalal teach the computer-implemented method of claim 1.
Hedge further teaches wherein training the machine learning model comprises minimizing the loss (i.e. para. [0090], Through training, the loss function gets reduced or minimized and the VAE learns a probability distribution of the data) .
Claim 9:
Claim 9 is the system claim reciting similar limitations to Claim 1 and is rejected for similar reasons.
Claim 10:
Claim 10 is the system claim reciting similar limitations to Claim 3 and is rejected for similar reasons.
Claim 11:
Claim 11 is the system claim reciting similar limitations to Claim 4 and is rejected for similar reasons.
Claim 13:
Claim 13 is the system claim reciting similar limitations to Claim 7 and is rejected for similar reasons.
Claim 15:
Claim 15 is the system claim reciting similar limitations to Claim 2 and is rejected for similar reasons.
Claim 16:
Claim 16 is the medium claim reciting similar limitations to Claim 1 and is rejected for similar reasons.
Claim 17:
Claim 17 is the medium claim reciting similar limitations to Claim 3 and is rejected for similar reasons.
Claim 20:
Claim 20 is the medium claim reciting similar limitations to Claim 2 and is rejected for similar reasons.
Claim(s) 6, 12, & 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication NO. 20210103822 “Hedge”, in light of U.S. Patent Application Publication NO. 20200034436 “Chen”, and further in light of U.S. Patent Application Publication NO. 20200026257 “Dalal”, and further in light of U.S. Patent Application Publication NO. 20210312307 “Hazard”.
Claim 6:
Hedge, Chen, and Dalal teach the computer-implemented method of claim 1.
While Hedge teaches a model for generating mixed synthetic data, Hedge may not explicitly teach further comprising: providing at least some of the plurality of mixed synthetic data as an input to training a second machine learning model.
However, Hazard teaches
providing at least some of the plurality of mixed synthetic data as an input to training a second machine learning model (i.e. para. [160, 0232], Fig. 6, “then the synthetic dataset may be provided 170 for use in one or more computer-based reasoning models and/or used to cause 199 control of a controllable system. Numerous embodiments of providing 170 the dataset for use in a computer-based reasoning system and causing 199 control of a controllable system”, wherein it is noted that the synthetic data may be mixed data generated from dataset a number continuous variables ‘n’ and categorical variables ‘I’).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to add providing at least some of the plurality of mixed synthetic data as an input to training a second machine learning model, to the mixed synthetic data generation of Hedge-Chen-Dalal, with a mixed synthetic data is provided to other machine learning model, as taught by Hazard. One would have been motivated to combine the additional testing steps of Hazard with the mixed synthetic data generation of Hedge-Chen-Dalal in order to increases the augment the accuracy of other machine learning models which may have only few real datasets to train with.
Claim 12:
Claim 12 is the system claim reciting similar limitations to Claim 6 and is rejected for similar reasons.
Claim 18:
Claim 18 is the medium claim reciting similar limitations to Claim 6 and is rejected for similar reasons.
Claim(s) 8, 14, & 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication NO. 20210103822 “Hedge”, in light of U.S. Patent Application Publication NO. 20200034436 “Chen”, and further in light of U.S. Patent Application Publication NO. 20200026257 “Dalal”, and further in light of U.S. Patent Application Publication NO. 20210089903 “Murray”.
Claim 8:
Hedge, Chen, and Dalal teach the computer-implemented method of claim 1.
Hedge may not explicitly teach
wherein at least some of the one or more continuous variables are pixels representing one or more images.
However, Murray teaches
wherein at least some of the one or more continuous variables are pixels representing one or more images (i.e. para. [0037], “the second conditioning variable (y.sub.a) 40 may relate to any other image property that may be represented by a score on a predefined scale. The second conditioning variable (y.sub.a) 40 may be referred to as the continuous conditioning variable”, wherein a continuous variable may be an property related to a pixel)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to add wherein at least some of the one or more continuous variables are pixels representing one or more images, to the continuous variable encoding of Hedge-Chen-Dalal, with a continuous variable may belong to a pixel representing an image property as taught by Murray. One would have been motivated to combine the distance finding of Murray with the mixed data encoding of Hedge-Chen-Dalal in order to increases the accuracy in generating realistic images conforming to the statistics of these datasets.
Claim 14:
Claim 14 is the system claim reciting similar limitations to Claim 8 and is rejected for similar reasons.
Claim 19:
Claim 19 is the medium claim reciting similar limitations to Claim 8 and is rejected for similar reasons.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
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.
U.S. Patent Application Publication NO. 20210209388 “Ciftci”, teaches in para. [0097] Given a training set, this technique learns to generate new data with the same statistics as the training set. For example, a GAN trained on photographs can generate new photographs that look at least superficially authentic to human observers, having many realistic characteristics. Though originally proposed as a form of generative model for unsupervised learning, GANs have also proven useful for semi-supervised learning, fully supervised learning, and reinforcement learning. The core idea of a GAN is based on the “indirect” training through the discriminator, which itself is also being updated dynamically. This basically means that the generator is not trained to minimize the distance to a specific image, but rather to fool the discriminator. This enables the model to learn in an unsupervised manner.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DAVID H TAN whose telephone number is (571)272-7433. The examiner can normally be reached M-F 7:30-4:30.
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, Cesar Paula can be reached at (571) 272-4128. 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./Examiner, Art Unit 2145
/CESAR B PAULA/Supervisory Patent Examiner, Art Unit 2145
Prosecution Insights