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 Arguments
Applicant’s arguments, see pgs. 7-9, filed 1-26-2026, with respect to 112, 101 rejections have been fully considered and are persuasive. The rejection has been withdrawn.
Applicant's arguments filed 1-26-2026 have been fully considered but they are not persuasive.
In re pg. 10, applicant argues
PNG
media_image1.png
380
825
media_image1.png
Greyscale
In response, the Examiner respectfully disagrees.
Trenholm at least disclose class-level imbalance and reduce such likelihood (i.e. prevention or exclusion) ([0064] In an embodiment, to prevent or reduce the likelihood of imbalanced classification, the ground-truthing system 100 can be provided with only balanced datasets. Imbalanced classification is a supervised learning problem where one class outnumbers another class by a large proportion. Imbalanced classification arises more frequently in binary classification problems than multi-level classification problems. Imbalanced data can occur where classes are not represented equally. While most classification datasets do not have an equal number of instances in each class, small differences are often insignificant. With imbalanced datasets the machine learning algorithm doesn't get the necessary information about the minority class to make an accurate prediction of the data class, which can cause a bias in the performance of classifiers towards a majority class).
In re pg. 10, applicant argues
PNG
media_image2.png
161
853
media_image2.png
Greyscale
In response, the Examiner respectfully disagrees.
Chen disclose adjusting threshold (i.e. how data are divided) and iterations (i.e. repeating process) (e) in response to a detection that a data exclusion error has emerged, changing the threshold value and repeating (b) - (d) ([0011] In embodiments described herein, the cycling can continue until the refined model-filtered subsets satisfy an iteration criterion based, for example, on data quality or maximum cycle numbers. [0036] The computer implemented process accesses the database to retrieve a first Subset A and a second Subset B of the training data set S (101). In one approach, Subset A and Subset B are selected so that the distribution of dirty data elements in the subset is about equal to the distribution in the overall data set S. Also, the Subset A and Subset B can be selected so that the numbers of data elements in each of the subsets is about the same. As it is desirable to maximize the number of clean data elements utilized in a training algorithm, Subset A and Subset B can be selected by dividing the training data set S equally, randomly selecting the elements for Subset A and Subset B so as to at least statistically maintain the distribution of dirty elements relatively equal in the two subsets. [0049] In the case of FIG. 2B, if the sizes are not converging, or other iteration criterion is not met, then the refined model-filtered Subset AnF is used to train the neural network to produce, and store in memory, a refined model MODEL_AnF (157). The process proceeds to increment the indexes n and m (158) and returns to block 153, where the just produced refined model MODEL_A(n−1)F is used to filter Subset B. [0044] The computer implemented process accesses the database to retrieve a first Subset A and a second Subset B of the training data set S (151). In one approach, Subset A and Subset B are selected so that the distribution of dirty data elements in the subset is about equal to this distribution in the overall data set S. Also, the Subset A and Subset B can be selected so that the numbers of data elements in each of the subsets are the same, or about the same. As it is desirable to maximize the number of clean data elements utilized in a training algorithm, Subset A and Subset B can be selected by dividing the training data set S equally, randomly selecting the elements for Subset A and Subset B so as to at least statistically tend to maintain the distribution of dirty elements relatively equal in the two subsets. Other techniques for selecting the elements of Subset A and Subset B can be applied taking into account the numbers of elements in each category, and other data-content-aware selection techniques.); and
In re pg. 11, applicant argues
PNG
media_image3.png
493
883
media_image3.png
Greyscale
In response, the Examiner respectfully disagrees.
Vahdat is not rely upon for triggering a VAE. Even assuming Vahdat is rely upon for “triggering” a VAE, the broadest reason interpretation of “triggering” means using or applying. In instant case, Vahdat disclose [0004]…After training has completed, new data that is similar to data in the original training dataset can be generated using the trained VAE.
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 factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1, 3-6, 8-10, 12, 14-16, 18, 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chen (US 20210383210 A1) in view of and TRENHOLM et al (US 20190019061 A1) in view of VAHDAT et al (US 20220101122 A1)
1. Chen disclose A system for training a machine learning (ML) model, comprising: at least one preprogrammed error detector; at least one processor and a memory coupled thereto, the at least one processor being configured to perform operations comprising ([0035] FIG. 2A is a flowchart illustrating a computer-implemented process for training a neural network ANN starting with “dirty” training data. The flowchart begins with providing a labeled training data set S (100), which may be stored in a database accessible to the processor or processors executing the process. An example labeled training data set can include thousands or tens of thousands (or more) of images, labeled as discussed above, or any other type of training data selected according to the mission function of the neural network to be implemented.):
(a) executing using dirty training data, starts by accessing a labeled training data set that can be dirty. The labeled training data set is divided into a first subset A and a second subset B. The procedure includes cycling between the subsets A and B, including producing refined model-filtered subsets of subsets A and B to provide a cleaned data set. Each refined model-filtered subset can have improved cleanliness and increased numbers of elements. [0033] Images of defects on integrated circuit assemblies taken in a manufacturing assembly line can be classified in many categories, usable as elements of a training data set. These defects vary significantly in counts for a given manufacturing process, and so the training data can have an uneven distribution, and includes large data sizes. Also, the labeling process for images like this may be done by a person, who can make significant numbers of errors. For example, to build up a new neutral network model to classify defect categories or types, first we need to provide a labeled image database for training. The image database includes the defect information. One might have 50,000 defect images in the database, and with each image labeled by human with a classification. So one image in the set might be classified as category 9, and another image in the set might be classified as category 15 . . . , etc. However, human error and ambiguous cases result in mislabeling. For example, one image in the set which should be classified as defect category 7, might be erroneously classified the into category 3. A data set with erroneously classified elements can be referred to as a dirty data set, or a noisy data set.);
(b) marking as erroneous each record that has been identified as including an error, based on a comparison to a threshold value ([0010] In general, a procedure described herein includes accessing a labeled training data set (S) that comprises relatively dirty labeled data elements. The labeled training data set is divided into a first subset A and a second subset B. The procedure includes, in cycle A, using the first subset A to train a model MODEL_A of a neural network, and filtering the second subset B of the labeled training data set using the model MODEL_A. A model-filtered subset B1F of subset Be is provided that has a number of elements that depends on the accuracy of the MODEL_A. Then, the next cycle, cycle AB, includes using the model-filtered subset B1F to train a model MODEL_B1F, and filtering the first subset A of the labeled training data set using the model MODEL_B1F. The model MODEL_B1F may have better accuracy than the model MODEL_A. This results in a refined model-filtered subset A1F of subset A that has a number of elements that depends on the accuracy of MODEL_B1F. Another cycle, cycle ABA, can be executed which includes using the refined model-filtered subset A1F to train a model MODEL_A1F, and filtering the second subset B of the labeled training data set using the model MODEL_A1F. The model MODEL_A1F may have a better accuracy than the model MODEL_A. This results in a refined model-filtered subset of B2F of subset B that has a number of elements that depends on the accuracy of MODEL_A1F, and can have a greater number of elements than model-filtered subset B1F);
(c) dividing the records from the dirty dataset into a clean fraction and a dirty fraction, the dirty fraction including the record(s) marked as erroneous, the clean fraction including the record(s) not marked as erroneous ([0010] In general, a procedure described herein includes accessing a labeled training data set (S) that comprises relatively dirty labeled data elements. The labeled training data set is divided into a first subset A and a second subset B. The procedure includes, in cycle A, using the first subset A to train a model MODEL_A of a neural network, and filtering the second subset B of the labeled training data set using the model MODEL_A. A model-filtered subset B1F of subset Be is provided that has a number of elements that depends on the accuracy of the MODEL_A. Then, the next cycle, cycle AB, includes using the model-filtered subset B1F to train a model MODEL_B1F, and filtering the first subset A of the labeled training data set using the model MODEL_B1F. The model MODEL_B1F may have better accuracy than the model MODEL_A. This results in a refined model-filtered subset A1F of subset A that has a number of elements that depends on the accuracy of MODEL_B1F. Another cycle, cycle ABA, can be executed which includes using the refined model-filtered subset A1F to train a model MODEL_A1F, and filtering the second subset B of the labeled training data set using the model MODEL_A1F. The model MODEL_A1F may have a better accuracy than the model MODEL_A. This results in a refined model-filtered subset of B2F of subset B that has a number of elements that depends on the accuracy of MODEL_A1F, and can have a greater number of elements than model-filtered subset B1F. [0012] A combination of the refined model-filtered subsets from subset A and subset B can be combined to provide a cleaned training data set. The cleaned training data set can be used to train an output model for a target neural network having a level of accuracy improved over training with the original training data set. The target neural network with the output model can be deployed in an inference engine. [0036] The computer implemented process accesses the database to retrieve a first Subset A and a second Subset B of the training data set S (101). In one approach, Subset A and Subset B are selected so that the distribution of dirty data elements in the subset is about equal to the distribution in the overall data set S. Also, the Subset A and Subset B can be selected so that the numbers of data elements in each of the subsets is about the same. As it is desirable to maximize the number of clean data elements utilized in a training algorithm, Subset A and Subset B can be selected by dividing the training data set S equally, randomly selecting the elements for Subset A and Subset B so as to at least statistically maintain the distribution of dirty elements relatively equal in the two subsets.);
(d) detecting whether a data exclusion error emerges in the dividing of the records into the clean fraction and the dirty fraction ([0036] The computer implemented process accesses the database to retrieve a first Subset A and a second Subset B of the training data set S (101). In one approach, Subset A and Subset B are selected so that the distribution of dirty data elements in the subset is about equal to the distribution in the overall data set S. Also, the Subset A and Subset B can be selected so that the numbers of data elements in each of the subsets is about the same. As it is desirable to maximize the number of clean data elements utilized in a training algorithm, Subset A and Subset B can be selected by dividing the training data set S equally, randomly selecting the elements for Subset A and Subset B so as to at least statistically maintain the distribution of dirty elements relatively equal in the two subsets.);
(e) in response to a detection that a data exclusion error has emerged, changing the threshold value and repeating (b) - (d) ([0036] The computer implemented process accesses the database to retrieve a first Subset A and a second Subset B of the training data set S (101). In one approach, Subset A and Subset B are selected so that the distribution of dirty data elements in the subset is about equal to the distribution in the overall data set S. Also, the Subset A and Subset B can be selected so that the numbers of data elements in each of the subsets is about the same. As it is desirable to maximize the number of clean data elements utilized in a training algorithm, Subset A and Subset B can be selected by dividing the training data set S equally, randomly selecting the elements for Subset A and Subset B so as to at least statistically maintain the distribution of dirty elements relatively equal in the two subsets. [0049] In the case of FIG. 2B, if the sizes are not converging, or other iteration criterion is not met, then the refined model-filtered Subset AnF is used to train the neural network to produce, and store in memory, a refined model MODEL_AnF (157). The process proceeds to increment the indexes n and m (158) and returns to block 153, where the just produced refined model MODEL_A(n−1)F is used to filter Subset B. [0044] The computer implemented process accesses the database to retrieve a first Subset A and a second Subset B of the training data set S (151). In one approach, Subset A and Subset B are selected so that the distribution of dirty data elements in the subset is about equal to this distribution in the overall data set S. Also, the Subset A and Subset B can be selected so that the numbers of data elements in each of the subsets are the same, or about the same. As it is desirable to maximize the number of clean data elements utilized in a training algorithm, Subset A and Subset B can be selected by dividing the training data set S equally, randomly selecting the elements for Subset A and Subset B so as to at least statistically tend to maintain the distribution of dirty elements relatively equal in the two subsets. Other techniques for selecting the elements of Subset A and Subset B can be applied taking into account the numbers of elements in each category, and other data-content-aware selection techniques.); and
While Chen discloses cleaning dirty dataset, Chen fails to disclose (d) detecting whether a data exclusion error emerges in the dividing of the records into the clean fraction and the dirty fraction; (e) in response to a detection that a data exclusion error has emerged, changing the threshold value and repeating (b) - (d); and (f) in response to a detection that a data exclusion error has not emerged: generating a new set of data samples by applying a variational autoencoder (VAE) to the clean fraction; augmenting the dirty dataset with the new set of data samples; and providing the augmented dirty dataset for training of the ML model.
TRENHOLM disclose increasing data quality for machine learning (thereby in the same field of endeavor) and specifically disclose wherein the at least one preprogrammed error detector comprises a plurality of different preprogrammed error detectors (([0081] The term “classifier” as used herein means any algorithm, or mathematical function implemented by a classification algorithm, that implements a classification process by mapping input data to a category. The term “classification” as used herein should be understood in a larger context than simply to denote supervised learning. By classification process we convey: supervised learning, unsupervised learning, semi-supervised learning, active/ground truther learning, reinforcement learning and anomaly detection. Classification may be multi-valued and probabilistic in that several class labels may be identified as a decision result; each of these responses may be associated with an accuracy confidence level. Such multi-valued outputs may result from the use of ensembles of same or different types of machine learning algorithms trained on different subsets of training data samples. There are various ways to aggregate the class label outputs from an ensemble of classifiers; majority voting is one method.);
(d) detecting whether a data exclusion error emerges in the dividing of the records into the clean fraction and the dirty fraction ([0064] In an embodiment, to prevent or reduce the likelihood of imbalanced classification, the ground-truthing system 100 can be provided with only balanced datasets. Imbalanced classification is a supervised learning problem where one class outnumbers another class by a large proportion. Imbalanced classification arises more frequently in binary classification problems than multi-level classification problems. Imbalanced data can occur where classes are not represented equally. While most classification datasets do not have an equal number of instances in each class, small differences are often insignificant. With imbalanced datasets, the machine learning algorithm doesn't get the necessary information about the minority class to make an accurate prediction of the data class, which can cause a bias in the performance of classifiers towards a majority class.); wherein detection of the data exclusion error is based on detecting at least one of a class-level data exclusion and an attribute-level data exclusion ([0064] In an embodiment, to prevent or reduce the likelihood of imbalanced classification, the ground-truthing system 100 can be provided with only balanced datasets. Imbalanced classification is a supervised learning problem where one class outnumbers another class by a large proportion. Imbalanced classification arises more frequently in binary classification problems than multi-level classification problems. Imbalanced data can occur where classes are not represented equally. While most classification datasets do not have an equal number of instances in each class, small differences are often insignificant. With imbalanced datasets, the machine learning algorithm doesn't get the necessary information about the minority class to make an accurate prediction of the data class, which can cause a bias in the performance of classifiers towards a majority class);
(e) in response to a detection that a data exclusion error has emerged, changing the threshold value and repeating (b) - (d) ([0064] In an embodiment, to prevent or reduce the likelihood of imbalanced classification, the ground-truthing system 100 can be provided with only balanced datasets. Imbalanced classification is a supervised learning problem where one class outnumbers another class by a large proportion. Imbalanced classification arises more frequently in binary classification problems than multi-level classification problems. Imbalanced data can occur where classes are not represented equally. While most classification datasets do not have an equal number of instances in each class, small differences are often insignificant. With imbalanced datasets, the machine learning algorithm doesn't get the necessary information about the minority class to make an accurate prediction of the data class, which can cause a bias in the performance of classifiers towards a majority class. [0065] When aggregating training data for training a classifier, the rate at which data is acquired may not be the same for all classes. As a result, datasets for a first class may reach a required sample size in a very short period of time, whereas data acquisition for a second class may have to continue for a longer duration to reach the same sample size. The dataset for the first class and dataset for the second class may then be integrated to create a balanced master dataset.); and
It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the data cleansing of Chen to incorporate balanced dataset generation of TRENHOLM.
Given the importance of balanced dataset ([0064] In an embodiment, to prevent or reduce the likelihood of imbalanced classification, the ground-truthing system 100 can be provided with only balanced datasets. Imbalanced classification is a supervised learning problem where one class outnumbers another class by a large proportion. Imbalanced classification arises more frequently in binary classification problems than multi-level classification problems. Imbalanced data can occur where classes are not represented equally. While most classification datasets do not have an equal number of instances in each class, small differences are often insignificant. With imbalanced datasets, the machine learning algorithm doesn't get the necessary information about the minority class to make an accurate prediction of the data class, which can cause a bias in the performance of classifiers towards a majority class), one having ordinary skill in the art would have been motivated to make this obvious modification.
VAHDAT disclose dataset for machine learning (thereby in the same field of endeavor) and specifically disclose (f) in response to a detection that a data exclusion error has not emerged ([0006] One drawback of using VAEs to generate new data is that VAEs oftentimes assign high probabilities to regions within the distribution of data point values generated by the decoder network that actually have low probabilities within the distribution of data points in the training dataset. These regions of erroneously high probabilities within the distribution of data point values generated by the decoder network correspond to regions of erroneously high probabilities within the distribution of latent variables learned by the prior network. The regions of erroneously high probabilities in the distribution of latent variables learned by the prior network result from the inability of the prior network to learn complex or “expressive” distributions of latent variable values. Because the high probability regions within the distribution of data point values generated by the decoder network or within the distribution of latent variables learned by the prior network may not accurately capture the attributes of actual data points in the training set, new data points generated by selecting latent variable values from regions of erroneously high probabilities in the distribution of latent variables learned by the prior network, converting the selected latent variable values via the decoder network into distributions of pixel values that include corresponding regions of erroneously high probabilities, and sampling pixel values from the distributions of pixel values oftentimes do not resemble the data in the training dataset.): generating a new set of data samples by applying a variational autoencoder (VAE) to the clean fraction; augmenting the dirty dataset with the new set of data samples; and providing the augmented dirty dataset for training of the ML model ([0004] A variational autoencoder (VAE) is a type of generative model. A VAE typically includes an encoder network that is trained to convert data points in the training dataset into values of “latent variables,” where each latent variable represents an attribute of the data points in the training dataset. The VAE also includes a prior network that is trained to learn a distribution of the latent variables associated with the training dataset, where the distribution of latent variables represents variations and occurrences of the different attributes in the training dataset. The VAE further includes a decoder network that is trained to convert the latent variable values generated by the encoder network back into data points that are substantially identical to data points in the training dataset. After training has completed, new data that is similar to data in the original training dataset can be generated using the trained VAE, by selecting latent variable values from the distribution learned by the prior network during training, converting those selected values, via the decoder network, into distributions of values of the data points; and selecting values of the data points from the distributions. Each new data point generated in this manner can include attributes that are similar (but not identical) to one or more attributes of the data points in the training dataset.).
It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the dataset used for machine learning of Chen to incorporate dataset generation of VAHDAT.
Given the advantage of VAHDAT’s new data generation using variational autoencoders ([0009] One approach to resolving the mismatch between the distribution of latent variable values learned by the prior network and the actual distribution of latent variable values produced by the encoder network from the training dataset, and the corresponding mismatch between the distribution of data point values generated by the decoder network and the actual distribution of data point values in the training dataset, is to implement an energy-based model trained with an iterative Markov Chain Monte Carlo (MCMC) sampling technique to learn a more complex or “expressive” distribution of latent variable values and/or data point values to represent the training dataset. However, each MCMC sampling step depends on the result of a previous sampling step, which prevents MCMC sampling operations from being performed in parallel. Further, a relatively large number of MCMC sampling steps is typically required for the energy-based model to achieve sufficient accuracy. Performing a larger number of MCMC sampling steps serially is both computationally inefficient and quite time-consuming. [0010] As the foregoing illustrates, what is needed in the art are more effective techniques for generating new data using variational autoencoders.), one having ordinary skill in the art would have been motivated to make this obvious modification.
3. TRENHOLM disclose The system of claim 1, wherein each of the plurality of different preprogrammed error detector is unique compared to each other error detectors in what types of errors it is configured to identify and/or in how it is preprogrammed to identify errors ([0081] The term “classifier” as used herein means any algorithm, or mathematical function implemented by a classification algorithm, that implements a classification process by mapping input data to a category. The term “classification” as used herein should be understood in a larger context than simply to denote supervised learning. By classification process we convey: supervised learning, unsupervised learning, semi-supervised learning, active/ground truther learning, reinforcement learning and anomaly detection. Classification may be multi-valued and probabilistic in that several class labels may be identified as a decision result; each of these responses may be associated with an accuracy confidence level. Such multi-valued outputs may result from the use of ensembles of same or different types of machine learning algorithms trained on different subsets of training data samples. There are various ways to aggregate the class label outputs from an ensemble of classifiers; majority voting is one method.).
4. TRENHOLM disclose The system of claim 1, wherein a first one of the plurality of different preprogrammed error detectors is an ML-based error detector and a second one of the plurality of different preprogrammed error detectors is an ensemble error detector ([0081] The term “classifier” as used herein means any algorithm, or mathematical function implemented by a classification algorithm, that implements a classification process by mapping input data to a category. The term “classification” as used herein should be understood in a larger context than simply to denote supervised learning. By classification process we convey: supervised learning, unsupervised learning, semi-supervised learning, active/ground truther learning, reinforcement learning and anomaly detection. Classification may be multi-valued and probabilistic in that several class labels may be identified as a decision result; each of these responses may be associated with an accuracy confidence level. Such multi-valued outputs may result from the use of ensembles of same or different types of machine learning algorithms trained on different subsets of training data samples. There are various ways to aggregate the class label outputs from an ensemble of classifiers; majority voting is one method.).
5. TRENHOLM disclose The system of claim 1, wherein (b) includes marking as erroneous each record that has been identified as including an error by a number of the plurality of different preprogrammed error detectors that meets or exceeds the threshold value ([0007] In an aspect, there is provided a method for increasing data quality of a dataset for semi-supervised machine learning analysis, the method executed on one or more processors, the method comprising: receiving the dataset for semi-supervised machine learning; receiving known class label information for a portion of the data in the dataset; receiving clustering parameters from a user; determining a data cleanliness factor, and where the data cleanliness factor is below a predetermined cleanliness threshold: assigning data without class label information as a data point to a cluster using the clustering parameters, each cluster having a cluster class label associated with such cluster; and determining a measure of assignment for each data point in each cluster, and where the measure of assignment for each data point is below a predetermined assignment threshold, receiving a class label for such data points, otherwise, assigning the respective cluster class label to each data point with the respective measure of assignment below the predetermined assignment threshold; and otherwise, outputting the dataset with associated class labels for machine learning analysis).
6. TRENHOLM disclose The system of claim 1, wherein (d) includes detecting class-level and attribute- level data exclusion errors ([0064] In an embodiment, to prevent or reduce the likelihood of imbalanced classification, the ground-truthing system 100 can be provided with only balanced datasets. Imbalanced classification is a supervised learning problem where one class outnumbers another class by a large proportion. Imbalanced classification arises more frequently in binary classification problems than multi-level classification problems. Imbalanced data can occur where classes are not represented equally. While most classification datasets do not have an equal number of instances in each class, small differences are often insignificant. With imbalanced datasets, the machine learning algorithm doesn't get the necessary information about the minority class to make an accurate prediction of the data class, which can cause a bias in the performance of classifiers towards a majority class).
8. Chen discloses The system of claim 1, wherein (b) is practiced by maintaining a list of indices of the records that have been identified as including errors (See Fig. 5).
9. Chen discloses The system of claim 1, wherein the at least one processor is configured to perform further operations comprising: identifying as a partially clean record each record newly added to the clean fraction upon a repetition of (b) - (d) triggered by (e); and modifying each partially clean record prior to the generation of the new set of data samples ([0022] FIG. 8 illustrates Subset A of FIG. 6, with an 80% clean data condition. See also Fig. 2B that the process is repeat until converge).
10. TRENHOLM disclose The system of claim 9, wherein the modification of each partially clean record includes replacing at least some of the data in the respective partially clean record with a statistical measure derived from within that data's corresponding class ([0041] Embodiments may apply unsupervised or semi-supervised learning algorithms to develop a clustering and classification system that facilitates cleaning, labeling, or ground-truthing of unlabeled, partially-labeled, or incorrectly labeled data. Embodiments of the present disclosure may include an autonomous data cleaning and labelling system, and may comprise a computer-implemented user interface to receive input from a human operator for ground-truthing data. Ground-truthing can include correctly labeling the data, validating doubtful class labels, or purging invalid data during a cleaning operation or during a training, testing or production run of a neural network. Further embodiments may include conducting ground-truthing locally or remotely on a computer or handheld network capable device by a user, crowd, domain expert or plurality of experts).
Claims 12, 14-16, 18 are method claims having similar limitation as claims 1-5, 9 and are rejected under the same rationale.
Claim 20 are medium claims having similar limitation as claim 1 and is rejected under the same rationale. Note for 20. A non-transitory computer readable storage medium tangibly storing instructions that, when executed by at least one processor of a system for training a machine learning (ML) model, perform operations comprising (See Chen’s [0082] Other implementations of the method described in this section can include a non-transitory computer readable storage medium storing instructions executable by a processor to perform any of the methods described above. Yet another implementation of the method described in this section can include a system including memory and one or more processors operable to execute instructions, stored in the memory, to perform any of the methods described above.)
Claim(s) 11, 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chen (US 20210383210 A1) in view of and TRENHOLM et al (US 20190019061 A1), VAHDAT et al (US 20220101122 A1), and further in view of Odry et al (US 20200020098 A1)
11. VAHDAT disclose The system of claim 1, wherein the VAE includes first and second neural networks, the first neural network being an encoder and the second neural network being a decoder ([0004] A variational autoencoder (VAE) is a type of generative model. A VAE typically includes an encoder network that is trained to convert data points in the training dataset into values of “latent variables,” where each latent variable represents an attribute of the data points in the training dataset. The VAE also includes a prior network that is trained to learn a distribution of the latent variables associated with the training dataset, where the distribution of latent variables represents variations and occurrences of the different attributes in the training dataset. The VAE further includes a decoder network that is trained to convert the latent variable values generated by the encoder network back into data points that are substantially identical to data points in the training dataset. After training has completed, new data that is similar to data in the original training dataset can be generated using the trained VAE, by selecting latent variable values from the distribution learned by the prior network during training, converting those selected values, via the decoder network, into distributions of values of the data points; and selecting values of the data points from the distributions. Each new data point generated in this manner can include attributes that are similar (but not identical) to one or more attributes of the data points in the training dataset.).
VAHDAT fails to disclose feed-forward neural network.
However, Odry disclose encoder/decoder of VAE is feed forward NN ([0033] Variational autoencoders (VAEs) can represent input MR data in a latent space whose parameters are learned during encoding. A VAE can capture shape variability, and has generative capability to synthesize images of tissue (e.g., brain images) given the underlying latent space (or manifold) coordinates. An autoencoder is a feedforward, non-recurrent neural network having an input layer, an output layer and one or more hidden layers connecting the input and output layers. The output layer has the same number of nodes as the input layer.)
As such, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention that VAE is feed forward NN.
Claim 19 having similar limitation as claim 11 and is rejected under the same rationale.
Pertinent Prior Art
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Eduardo et al (“REPAIRING SYSTEMATIC OUTLIERS BY LEARNING CLEAN SUBSPACES IN VAES” Jul 2022) disclose a novel semi-supervised model for detection and automated repair of systematic errors. See abstract.
Lee et al (“A Survey on Data Cleaning Methods for Improved Machine Learning Model Performance” 2021) disclose A Survey on Data Cleaning Methods for Improved Machine Learning Model Performance. See abstract.
Maharana et al (“A review: Data pre-processing and data augmentation techniques” April 2022) disclose A review: Data pre-processing and data augmentation techniques. See abstract.
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 LUT WONG whose telephone number is (571)270-1123. The examiner can normally be reached M-F 10am-6pm 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, Abdullah Al Kawsar can be reached at 5712703169. 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.
/LUT WONG/Primary Examiner, Art Unit 2127
Prosecution Insights