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 .
DETAILED ACTION
This action is responsive to the following communication: Amendment filed Sept. 17, 2026. This Action is made Final.
Claims 1, 5, 7-11, 15-26 are pending in the case. Claims 1, 15 and 16 are independent claims.
Claim Rejections - 35 USC § 103
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.
Claims 1, 5, 7-11, 15-26 are rejected under 35 U.S.C. 103 as being unpatentable over Gong et al. (hereinafter Gong) U.S. Patent Publication No. 2020/0372369 in view of Zhang et al. (hereinafter Zhang) U.S. Patent Publication No. 2020/0104702.
With respect to independent claim 1, Gong teaches a computer-implemented method comprising: receiving data from a device, the data comprising a plurality of data points representing sensor values of the device (see e.g., Para [3] – “data are collected from different measurement tools with different recording mechanisms”), each data point comprising a respective value of each of a plurality of features for each data point in the received data, except that for at least one of the plurality of data points a respective value of at least one but not all of the features is missing such that the received data comprises at least one missing value (see e.g., Fig. 1 Para [63][64] – “Input data 102 is received by an attributive proposal network 106, and the input data 102 is typically only partially-observed … input data 102 can also be heterogeneous data … The data is split between observed and unobserved data to reflect the “completeness” of the input data 102”);
encoding, using a first neural network having a first set of parameters (see e.g., Fig. 1 Para [105]-[106]– “VSAE handles partially-observed heterogeneous data by learning individual encoders for observed attributes and a collective encoder for unobserved ones. The latent codes representing each attribute are then aggregated and fed to the decoders to reconstruct the mask and all attributes independently.”), the values of the received data into a distribution of a first plurality of latent vectors, the distribution of the first plurality of latent vectors having a mean and a variance (see e.g., Para [67][90][106]-[119][120] [203] – “All priors p(z.sub.i) can be standard Gaussians and proposal distributions q(z.sub.i|x, m) are diagonal Gaussians with means and covariances inferred by neural networks. This factorization separates encoding of each attribute and yields an efficient distribution for the latent space by assuming the latent variables are conditionally independent given the data and mask.”);
decoding, using a second neural network having a second set of parameters (see e.g., Fig. 1 – 110 Data Generative Network), a random sample of the distribution (see e.g., Para [76][110]-[116] – “The latent variables for all attributes are sampled using the selective proposal distributions in (8). Next, to capture the intricate dependency relationships between observed and unobserved attributes, the variational latent variables are aggregated with an aggregator … “) of the first plurality of latent vectors into a computed vector for each data point (see e.g., Fig. 1 Para [58][76][115]-[118][133]);
inputting the computed vector (see e.g., Para [115]-[120][133] – “The latent variables for all attributes are sampled using the selective proposal distributions … the variational latent variables are aggregated with an aggregator … before providing to the decoders”) for each data point into a third neural network (see e.g., Fig. 1 – 116 Mask Generative Network Para [11]) having a third set of parameters to determine a computed plurality of mask vectors (see e.g., Para [120] – “The decoder, an MLP parameterized by, maps the aggregated latent code to a reconstruction of the binary mask vector assuming each output dimension is independent Bernoulli distribution.”), the plurality of mask vectors comprising a computed mask vector for each data point, wherein each computed mask vector comprises a computed binary value for each feature to indicate whether a value for each feature is missing or not (see e.g., Para [11][99][119][120] – “ An M-dimensional binary mask variable m∈{0, 1}.sup.M is denoted to represent the missingness: for the i-th attribute, m.sub.i=1 if it is observed and 0 otherwise. T “ “The latent codes representing each attribute are then aggregated and provided to decoders to reconstruct a mask (e.g., using a mask generative network) and all attributes independently. The latent codes and the generated mask are provided to a data generative network to sample the attributes. “ “Mask VAE consists of an encoder which encodes the mask vector to stochastic latent variables and a decoder which decodes the latent variables to a reconstruction of the mask.”);
tuning the first set of parameters of the first neural network, the second set of parameters of the second neural network and the third set of parameters of the third neural network by minimizing a loss function (see e.g., Para [121] – the Objective Function), the loss function comprising a sum of (Gong does not expressly show the loss function is a sum of the measure of difference and error. However, based on the teachings of Gong, it would have been obvious for a person of ordinary skill in the art to try the algorithm. It is merely a calculation choice.): a measure of difference between the distribution of the first plurality of latent vectors and the distribution of the second plurality of latent vectors (see e.g., Para [12][144] – “The architecture was compared against benchmark computing architectures on benchmark data sets (e.g., Fashion MNIST+label, MNIST+MNIST, CMU-MOSI), and was found to yield technical benefits and improvements, including improved computational accuracy (e.g., improved mean square error scores, shown both in mean and standard deviations over independent runs).”” Mean-squared error, cross-entropy and binary cross-entropy are used as reconstruction loss for numerical, categorical and mask variables, respectively.”); and
an error determined at least in part by comparing the computed vector for each data point and the computed plurality of mask vectors with and ground truth data for the received data (see e.g., Equation 11 Para [107][121][222]); and
outputting imputed data that diagnoses at least one malfunction in the device (see e.g., Para [6] – “ the architecture can be trained to computationally impute new data to augment the missing data from the observed data (e.g., computationally applying a framework for conditional generation of unobserved attributes). “), the imputed data comprising a computed value for the at least one missing value from the computed vector for each data point (see e.g., Para [26] – “In further variants, any other task that could be solved by representation learning models could be also solved by the model (e.g. feature extraction, dimensionality reduction, anomaly detection, distribution estimation).”).
Gong does not expressly show encoding, using a fourth neural network having a fourth set of parameters, background data into a distribution of a second plurality of latent vectors, the distribution of the second plurality of latent vectors having a mean and a variance, and wherein the background data comprises fully observed data comprising a value for each feature and for each data point in the received data. However, Zhang teaches similar features. Zhang teaches the recited fully observed data (see e.g., Para [89]-[95] – “The aim of the Full VAE is to learn a mapping from observed variables x.sub.o to a latent space variable z, and to learn a mapping from z to estimate variables x. The inference model takes observed variables x.sub.o as inputs and outputs multiple distributions, each distribution representing a respective observed variable … A partial VAE aims to learn a mapping from variables X to a hidden variable z. However, in contrast to a Full VAE, the inference model has to cope with receiving a vector X comprising an arbitrary set of input variables (or values) containing both observed variables x.sub.o and unobserved variables x.sub.u. The partial VAE therefore needs to learn inferences based on an incomplete set of input values. ”) Zhang further teaches encoding background data into a distribution of a second plurality of latent vectors (see e.g., Para [93] – “The inference model takes observed variables x.sub.o as inputs and outputs multiple distributions, each distribution representing a respective observed variable. The distributions are sampled to generate a vector Z. The generative model takes vector Z as an input and outputs estimates of variables x. The Full VAE is trained on sets of variables that each have an observed value.”). Both Gong and Zhang are directed to VAE based latent representations for analyzing input data. Accordingly, it would have been obvious to the skilled artisan before the effective filing date of the claimed invention having Gong and Zhang in front of them to modify the system of Gong to include the above feature. The motivation to combine Gong and Zhang comes from Zhang. Zhang discloses the motivation to incorporated both fully observed data and partially observed data so that additional data can be employed (see e.g. Para [89]-[95]). This motivation for combination also applies to the remaining claims which depend on this combination.
With respect to dependent claim 5, the modified Gong teaches determining the received data by removing, from the ground truth data, values for at least one of the plurality of features outside of a respective range of values for each of the at least one feature (see e.g., Para [7]).
With respect to dependent claim 7, Gong the modified teaches determining the mask vector for each data point in the received data from the at least one sensor value and at least one missing value of the received data (see e.g., Para [66][67] – “a mask 104 can be generated from the input data 102”).
With respect to dependent claim 8, Gong the modified teaches after tuning the first set of parameters of the first neural network, the second set of parameters of the second neural network and the third set of parameters of the third neural network, setting the first set of parameters of the first neural network and the second set of parameters of the second neural network to respective tuned values (see e.g., Para [72] –“ the model is jointly trained following stochastic gradient variational Bayes.”); receiving second received data from the device, the second received data comprising a plurality of data points representing sensor values of the device, each data point comprising a respective value of each of a plurality of features for each data point in the second received data, except that for at least one of the plurality of data points a respective value of at least one but not all of the features is missing such that the second received data comprises at least one missing value; encoding, using the first neural network having the first set of parameters, the second received data into a distribution of a third plurality of latent vectors; sampling a random sample of latent vectors from the distribution of the third plurality of latent vectors; decoding, using the second neural network having the second set of parameters, the random sample of latent vectors from the distribution of the third plurality of latent vectors to obtain the at least one missing value in the second received data (see e.g., Para [5][50][55] – “the trained model can be deployed for later use in both data generation and imputation. In particular, once trained, the model can impute missing data from any combination of observed variables while being trained efficiently with a single variational objective.”) and diagnosing a fault in the device using the obtained at least one missing value (see e.g., Para [26] – “In further variants, any other task that could be solved by representation learning models could be also solved by the model (e.g. feature extraction, dimensionality reduction, anomaly detection, distribution estimation).”).
With respect to dependent claim 9, the modified Gong teaches the background data is input by a user (see e.g., Para [22] – “The system can be interoperated with, for example, through various application programming interfaces (APIs), whereby data sets are provided through upload or data stream and an output data set is returned.”).
With respect to dependent claim 10, the modified Gong teaches tuning the first set of parameters of the first neural network, the second set of parameters of the second neural network and the third set of parameters of the third neural network by minimizing a loss function comprises: using a gradient descent algorithm for each of the first set of parameters of the first neural network, the second set of parameters of the second neural network and the third set of parameters of the third neural network (see e.g., Para [72][137] – “the model is jointly trained following stochastic gradient variational Bayes.” “using neural networks and optimizing the parameters via backpropagation techniques, according to some embodiments.”).
With respect to dependent claim 11, the modified Gong teaches the at least one missing value in the received data comprises Missing Not At Random data (see e.g., Para [87][155][157] – “Not Missing At Random (NMAR): The mask distribution depends on observed and unobserved attributes. Applicants sample the mask from the probability π(m.sub.i)=sigmoid(x.sub.i), where m.sub.i is i-th element of mask m, x.sub.i is the i-th attribute.” NMAR corresponds to “Missing Not At Random”).
With respect to dependent claim 12, the modified Gong teaches the received data comprises at least one data value representing physical measurements of at least one patient (see e.g., Para [50] – “patient records like gender, age, the history of allergies, symptoms and etc.”).
With respect to dependent claim 13, the modified Gong teaches the method is used to diagnose the at least one patient (Gong does not expressly show this feature. However, Gong expressly indicates “[l]earning from data is an objective of artificial intelligence. Learning algorithms often rely heavily on clean homogeneous data, whereas in the real world, data is filled with noisy heterogeneous data. Heterogeneity is ubiquitous in a variety of applications and platforms from healthcare and finance to social networks and manufacturing systems.” See Para [3]. Therefore, it would have been obvious to apply the method to diagnose of patient.).
With respect to dependent claim 14, the modified Gong teaches the background data comprises an age and a gender of each patient (see e.g., Para [50] – “ there are many different types of attributes in patient records like gender, age, the history of allergies, symptoms and etc.”).
Claim 15 is rejected for the similar reasons discussed above with respect to claim 1.
Claim 16 is rejected for the similar reasons discussed above with respect to claim 1.
With respect to dependent claim 17, the modified Gong teaches the background data comprises an age and a gender of each patient (see e.g., Para [50] – “ there are many different types of attributes in patient records like gender, age, the history of allergies, symptoms and etc.”).
With respect to dependent claim 18, the modified Gong teaches determining the received data by removing, from the ground truth data, values for at least one of the plurality of features outside of a respective range of values for each of the at least one feature (see e.g., Para [7]).
With respect to dependent claim 19, the modified Gong teaches the operations further include: determining the mask vector for each data point in the received data from at least one physical measurement and the at least one missing value of the received data (see e.g., Para [50] – “patient records like gender, age, the history of allergies, symptoms and etc.” Also see Zhang Para [47]).
With respect to dependent claim 20, the modified Gong teaches the operations further include: after tuning the first set of parameters of the first neural network, the second set of parameters of the second neural network and the third set of parameters of the third neural network, setting the first set of parameters of the first neural network and the second set of parameters of the second neural network to respective tuned values (see e.g., Para [72] –“ the model is jointly trained following stochastic gradient variational Bayes.”); receiving second received data, the second received data comprising a plurality of data points representing physical measurements of the at least one patient, each data point comprising a respective value of each of a plurality of features for each data point in the second received data, except that for at least one of the plurality of data points a respective value of at least one but not all of the features is missing such that the second received data comprises at least one missing value; encoding, using the first neural network having the first set of parameters, the second received data into a distribution of a third plurality of latent vectors; sampling a random sample of latent vectors from the distribution of the third plurality of latent vectors; decoding, using the second neural network having the second set of parameters, the random sample of latent vectors from the distribution of the third plurality of latent vectors to obtain the at least one missing value in the second received data (see e.g., Para [5][50]-[55] – “the trained model can be deployed for later use in both data generation and imputation. In particular, once trained, the model can impute missing data from any combination of observed variables while being trained efficiently with a single variational objective.”); and diagnosing the at least one patient using the obtained at least one missing value (see e.g., Para [26] – “In further variants, any other task that could be solved by representation learning models could be also solved by the model (e.g. feature extraction, dimensionality reduction, anomaly detection, distribution estimation).”).
With respect to dependent claim 21, the modified Gong teaches determining the received data by removing, from the ground truth data, values for at least one of the plurality of features outside of a respective range of values for each of the at least one feature (see e.g., Para [7]).
With respect to dependent claim 22, the modified Gong teaches the operations further include: determining the mask vector for each data point in the received data from at least one sensor value and the at least one missing value of the received data (see e.g., Para [50] – “patient records like gender, age, the history of allergies, symptoms and etc.” Also see Zhang Para [47]).
With respect to dependent claim 23, the modified Gong teaches the operations further include: after tuning the first set of parameters of the first neural network, the second set of parameters of the second neural network and the third set of parameters of the third neural network, setting the first set of parameters of the first neural network and the second set of parameters of the second neural network to respective tuned values (see e.g., Para [72] –“ the model is jointly trained following stochastic gradient variational Bayes.”); receiving second received data from the device, the second received data comprising a plurality of data points representing sensor values of the device, each data point comprising a respective value of each of a plurality of features for each data point in the second received data, except that for at least one of the plurality of data points a respective value of at least one but not all of the features is missing such that the second received data comprises at least one missing value; encoding, using the first neural network having the first set of parameters, the second received data into a distribution of a third plurality of latent vectors; sampling a random sample of latent vectors from the distribution of the third plurality of latent vectors; decoding, using the second neural network having the second set of parameters, the random sample of latent vectors from the distribution of the third plurality of latent vectors to obtain the at least one missing value in the second received data (see e.g., Para [5][50][55] – “the trained model can be deployed for later use in both data generation and imputation. In particular, once trained, the model can impute missing data from any combination of observed variables while being trained efficiently with a single variational objective.”); and diagnosing a fault in the device using the obtained at least one missing value (see e.g., Para [26] – “In further variants, any other task that could be solved by representation learning models could be also solved by the model (e.g. feature extraction, dimensionality reduction, anomaly detection, distribution estimation).”).
With respect to dependent claim 24, the modified Gong teaches the background data is input by a user (see e.g., Para [22] – “The system can be interoperated with, for example, through various application programming interfaces (APIs), whereby data sets are provided through upload or data stream and an output data set is returned.”).
With respect to dependent claim 25, the modified Gong teaches tuning the first set of parameters of the first neural network, the second set of parameters of the second neural network and the third set of parameters of the third neural network by minimizing a loss function comprises: using a gradient descent algorithm for each of the first set of parameters of the first neural network, the second set of parameters of the second neural network and the third set of parameters of the third neural network (see e.g., Para [72][137] – “the model is jointly trained following stochastic gradient variational Bayes.” “using neural networks and optimizing the parameters via backpropagation techniques, according to some embodiments.”).
With respect to dependent claim 26, the modified Gong teaches the at least one missing value in the received data comprises Missing Not At Random (MNAR) data (see e.g., Para [87][155][157] – “Not Missing At Random (NMAR): The mask distribution depends on observed and unobserved attributes. Applicants sample the mask from the probability π(m.sub.i)=sigmoid(x.sub.i), where m.sub.i is i-th element of mask m, x.sub.i is the i-th attribute.” NMAR corresponds to “Missing Not At Random”).
It is noted that any citation to specific pages, columns, lines, or figures in the prior art references and any interpretation of the references should not be considered to be limiting in any way. “The use of patents as references is not limited to what the patentees describe as their own inventions or to the problems with which they are concerned. They are part of the literature of the art, relevant for all they contain.” In re Heck, 699 F.2d 1331, 1332-33, 216 USPQ 1038, 1039 (Fed. Cir. 1983) (quoting In re Lemelson, 397 F.2d 1006, 1009, 158 USPQ 275, 277 (CCPA 1968)). Further, a reference may be relied upon for all that it would have reasonably suggested to one having ordinary skill the art, including nonpreferred embodiments. Merck & Co. v. Biocraft Laboratories, 874 F.2d 804, 10 USPQ2d 1843 (Fed. Cir.), cert. denied, 493 U.S. 975 (1989). See also Upsher-Smith Labs. v. Pamlab, LLC, 412 F.3d 1319, 1323, 75 USPQ2d 1213, 1215 (Fed. Cir. 2005); Celeritas Technologies Ltd. v. Rockwell International Corp., 150 F.3d 1354, 1361, 47 USPQ2d 1516, 1522-23 (Fed. Cir. 1998).
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 PEIYONG WENG whose telephone number is (571)270-1660. The examiner can normally be reached on Mon.-Fri. 8 am to 5 pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Matthew Ell, can be reached on (571) 270-3264. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://portal.uspto.gov/external/portal. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
/PEI YONG WENG/Primary Examiner, Art Unit 2141
