DETAILED ACTION
This action is responsive to the Application filed on 02/07/2024. Claims 1-10 are pending in the case. Claims 1 and 10 are independent claims.
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 .
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-10 are rejected under 35 U.S.C. 101 because the claim are directed to an abstract idea without significantly more.
Regarding Claim 1/10:
Under step 1, claim 1 is directed to a method which is directed to a process, one of the statutory categories.
Under step 1, claim 10 is directed to a device for component anomaly detection which is directed to a system, one of the statutory categories.
Under Step 2A Prong 1, the claim recites the following limitations which are considered mental evaluations:
deriving a first abnormal score of input data… extracting a manifold coordinate of the input data; and calculating a distance between the manifold coordinate and an average coordinate of normal data as the first abnormal score; … deriving a second abnormal score of the input data… adding the first abnormal score to the second abnormal score to calculate an abnormal degree of the input data.
Each of these amount to mental evaluation because they describe manipulation of abstract data. Deriving, extracting and calculating are all evaluations of data which can be performed in the mind.
Under step 2A Prong 2, The claim recites the following additional element(s):
by a denoising auto-encoder … by a discriminator; …[from claim 10] a processor; and a computer storage media coupled to the processor, and configured to store computer-readable instructions for instructing the processor to execute the generative adversarial networks-based method (which amounts to descriptions which makes use of or applies the abstract idea because under 2106.05(f)(1) “the claim fails to recite details of how a solution to a problem is accomplished”)
Therefore the claim is directed to a judicial exception.
Under step 2B, the recited additional elements when considered alone or in combination neither integrates the abstract idea into a practical application nor provides significantly more than the abstract idea itself.
Regarding Claim 2:
The rejection of claim 1 is incorporated and further:
Under Step 2A Prong 1, The claim recites the limitations:
deriving a first discrimination value of the input data
converting the first discrimination value into the second abnormal score
Under step 2A Prong 2, The claim recites the following additional element(s):
by the discriminator (which amounts to descriptions which makes use of or applies the abstract idea because under 2106.05(f)(1) “the claim fails to recite details of how a solution to a problem is accomplished”)
Therefore the claim is directed to a judicial exception.
Under step 2B, the recited additional elements when considered alone or in combination neither integrates the abstract idea into a practical application nor provides significantly more than the abstract idea itself.
Regarding Claim 3:
The rejection of claim 2 is incorporated and further:
Each of the limitations described in the claim, under Step 2A Prong 1, only serve to describe the abstract ideas addressed in the parent claim, in particular the limitations describe mental evaluations.
Regarding Claim 4:
The rejection of claim 1 is incorporated and further:
Under Step 2A Prong 1, The claim recites the limitations:
preprocessing real-time data to generate processed data; deleting the processed data that does not match with a feature set; and retaining the processed data that matches with the feature set as the input data.
Furthermore, under step 2A Prong 2 and 2B, the claim(s) do not recite additional elements to consider other than those considered in the independent/parent claim.
Regarding Claim 5:
The rejection of claim 1 is incorporated and further:
Under Step 2A Prong 1, The claim recites the limitations:
A1) deriving second discrimination values of multiple sets of input data …, wherein the multiple sets of input data include at least one real data and at least one generative data;
A2) calculating a first expected value based on the second discrimination values by the initial discriminator;
A3) inputting the first expected value into a first objective function to determine whether an outcome of the first objective function is maximized;
A4) when the outcome of the first objective function has not been maximized, adjusting parameters of the initial discriminator with fixed parameters of an initial denoising auto-encoder according to the first expected value; and
A5) repeating the step A1) to the step A4)
Each of these amount to mental evaluation because they describe manipulation of abstract data.
Under step 2A Prong 2, The claim recites the following additional element(s):
by an initial discriminator … to train the initial discriminator until the outcome of the first objective function has been maximized, to generate the discriminator (which amounts to descriptions which makes use of or applies the abstract idea because under 2106.05(f)(1) “the claim fails to recite details of how a solution to a problem is accomplished” The claims recites a series of mental evaluations performed by named models to produce a trained model.)
Therefore the claim is directed to a judicial exception.
Under step 2B, the recited additional elements when considered alone or in combination neither integrates the abstract idea into a practical application nor provides significantly more than the abstract idea itself.
Regarding Claim 6:
The rejection of claim 5 is incorporated and further:
Under Step 2A Prong 1, The claim recites the limitations:
deriving third discrimination values of the generated data and the at least one real data…
calculating a second expected value based on the third discrimination value…to determine whether an outcome of the second objective function is minimized;… when the outcome of the second objective function has not been minimized, adjusting the parameters of the initial denoising auto-encoder with fixed parameters of the discriminator and according to the second expected value;
Each of these amount to mental evaluation because they describe manipulation of abstract data.
Under step 2A Prong 2, The claim recites the following additional element(s):
by the initial discriminator, … inputting the second expected value into a second objective function … and A10) repeating the step A6) to the step A9) to train the initial denoising auto-encoder until the outcome of the second objective function has been minimized, to generate the denoising auto-encoder (which amounts to descriptions which makes use of or applies the abstract idea because under 2106.05(f)(1) “the claim fails to recite details of how a solution to a problem is accomplished” The claims recites a series of mental evaluations performed by named models to produce a trained model.)
and feeding the third discrimination values back to the denoising auto-encoder; (which amounts to adding insignificant extra-solution activity to the judicial exception, because the limitation describe mere data gathering. See MPEP 2106.05(g) )
Therefore the claim is directed to a judicial exception.
Under step 2B,
The additional element and feeding the third discrimination values back to the denoising auto-encoder is well understood, routine, and conventional activity because it amounts to “transmitting or receiving data over a network" (see MPEP 2106.05(d)(II)(i) )
the recited additional elements when considered alone or in combination neither integrates the abstract idea into a practical application nor provides significantly more than the abstract idea itself.
Regarding Claim 7:
The rejection of claim 5 is incorporated and further:
Under Step 2A Prong 1, The claim recites the limitations:
adding noise to the normal data to generate noisy data; and converting the noisy data into the generative data Under step 2A Prong 2, The claim recites the following additional element(s):
by the initial discriminator, to input the generative data to the discriminator. (which amounts to descriptions which makes use of or applies the abstract idea because under 2106.05(f)(1) “the claim fails to recite details of how a solution to a problem is accomplished”)
Therefore the claim is directed to a judicial exception.
Under step 2B, the recited additional elements when considered alone or in combination neither integrates the abstract idea into a practical application nor provides significantly more than the abstract idea itself.
Regarding Claim 8:
The rejection of claim 5 is incorporated and further:
Under Step 2A Prong 1, The claim recites the limitations:
wherein before the training process, the method further comprises … performing a pre-processing process, and the pre-processing process comprises: B1) selecting multiple influential features from statistical values of a training data set through a feature selection method to generate a feature set; and B2) selecting data and parameters that match with the feature set as the training data set
providing the training data set to the initial discriminator and the initial denoising auto-encoder during the training process (which amounts to adding insignificant extra-solution activity to the judicial exception, because the limitation describe mere data gathering. See MPEP 2106.05(g) )
Therefore the claim is directed to a judicial exception.
Under step 2B,
The additional element providing the training data set to the initial discriminator and the initial denoising auto-encoder during the training process is well understood, routine, and conventional activity because it amounts to “transmitting or receiving data over a network" (see MPEP 2106.05(d)(II)(i) )
the recited additional elements when considered alone or in combination neither integrates the abstract idea into a practical application nor provides significantly more than the abstract idea itself.
Regarding Claim 9:
The rejection of claim 1 is incorporated and further:
Under Step 2A Prong 1, The claim recites the limitations:
determining that the input data is abnormal when the abnormal degree is higher than a user-defined threshold.
Furthermore, under step 2A Prong 2 and 2B, the claim(s) do not recite additional elements to consider other than those considered in the independent/parent claim.
Claim Rejections - 35 USC § 102
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 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 the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1-6 and 9-10 is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Chen et al. “Dual Auto-Encoder GAN-Based Anomaly Detection for Industrial Control System”
Regarding claim 1/10,
Chen teaches, A generative adversarial networks-based method for component anomaly detection, comprising: ( pg 2 “The main contributions of this paper are as follows: …A dual GAN model based on the “encoder–decoder–encoder” architecture is developed to accurately detect the outliers of the industrial control system without depending on any anomalous samples”) [claim 10] A device for component anomaly detection, comprising: a processor; and a computer storage media coupled to the processor, and configured to store computer-readable instructions for instructing the processor to execute the generative adversarial networks-based method ( pg 15 “A normal PC server (ThinkPad L490) is used as the experimental platform. The server is equipped with an 8-core i7 CPU, 16 GB of main memory, and Windows 10 operating system. The DAGAN model is programmed with Python and runs with Spyder as an IDE.”) 1) deriving a first abnormal score of input data by a denoising auto-encoder, comprising: 11) extracting a manifold coordinate of the input data; and
12) calculating a distance between the manifold coordinate and an average coordinate of normal data as the first abnormal score; (pg 13 A test sample ˆx is first fed into the first GAN to obtain its low-dimensional feature z. Second, z is reconstructed into a new sample ˆx1 according to the normal distribution learned by G1. Finally, ˆx1 is encoded again by En1 as a low-dimensional feature z1. Therefore, the first anomaly score measures the similarity between two features z and z1, defined as:
PNG
media_image1.png
89
322
media_image1.png
Greyscale
… When the test sample ˆx is a normal sample, Score1 is small and close to 0. When the test sample ˆx is a marginal sample, Score1 is large. When the test sample ˆx is outlier, Score1 is very large” Score1 is the first abnormal score which is computed based on input to the encoder En. The output of the encoder is the manifold coordinate. The distance between the two features is the distance between the manifold coordinate and a typical or average coordinate of normal data. Examiner, notes an “denoising” is understood to be a label given to the autoencoder, nothing in the disclosure suggests how denoising is performed) 2) deriving a second abnormal score of the input data by a discriminator; and (pg 13 “According to the first anomaly score, marginal samples and outliers cannot be well identified, which may easily lead to the misjudgment of marginal samples. For this, the test sample ˆx is again fed into the second GAN and reconstructed to generate marginal samples ˆx2 according to the marginal distribution learned by G2. To compare ˆx and ˆx2 from the feature space, the output features of an intermediate layer in discriminator D2 are used” a second score is derived from the discriminator.) adding the first abnormal score to the second abnormal score to calculate an abnormal degree of the input data. (pg 13 “Combining the above two scores, the final anomaly score of our DAGAN model is defined as follows:
PNG
media_image2.png
65
334
media_image2.png
Greyscale
” the scores are combined or added together.)
Regarding claim 2
Chen teaches claim 1
Further Chen teaches, 21) deriving a first discrimination value of the input data by the discriminator; and 22) converting the first discrimination value into the second abnormal score. (pg 13 “To compare ˆx and ˆx2 from the feature space, the output features of an intermediate layer in discriminator D2 are used…. defined as:…
PNG
media_image3.png
77
280
media_image3.png
Greyscale
… the final anomaly score of our DAGAN model is defined as follows:
PNG
media_image4.png
47
318
media_image4.png
Greyscale
” G2 contains a discriminator which is derived a discrimination values based on the input x, the normalized score is the converted value, further the score modified by the gamma parameter can also be considered the converted value.)
Regarding claim 3
Chen teaches claim 2
Further Chen teaches, wherein the step 22) further comprises one of the following steps: multiplying the first discrimination value by -1 to obtain the second abnormal score, wherein the first discrimination value is an arbitrary real number; subtracting the first discrimination value from a preset value to obtain the second abnormal score; and when the first discrimination value is greater than zero and less than 1, taking a reciprocal of the first discrimination value as the second abnormal score. (pg 13 “defined as:…
PNG
media_image3.png
77
280
media_image3.png
Greyscale
…” the score is derived via subtracting the first discrimination value from a preset value as shown in the equation.)
Regarding claim 4
Chen teaches claim 1
Further Chen teaches, preprocessing real-time data to generate processed data; deleting the processed data that does not match with a feature set; and retaining the processed data that matches with the feature set as the input data. (pg 14 Section 5 “In this paper, two publicly available industrial control system datasets are selected to train and verify our DAGAN model…. Distributed Smart Space Orchestration System (DS2OS). This dataset contains 357,952 samples… For simplicity, 20,000 normal samples in DS2OS dataset are used as the training set in this paper, 1500 normal samples and 350 abnormal samples are used as the test set (50 samples for each attack).” A subset of data is selected from a set of 357,952, this smaller subset corresponds to the features selected to match with the input data for training. The data not used are the data which does not match, and are considered deleted with respect to the data processing.)
Regarding claim 5
Chen teaches claim 1
Further Chen teaches, performing a training process and the training process comprises… A1) deriving second discrimination values of multiple sets of input data by an initial discriminator, wherein the multiple sets of input data include at least one real data and at least one generative data; A2) calculating a first expected value based on the second discrimination values by the initial discriminator; A3) inputting the first expected value into a first objective function to determine whether an outcome of the first objective function is maximized;(pg 4-5 Section 3.1 “Therefore, the generator G and discriminator D constitute a min–max game. After multiple alternating trainings, the generator G and discriminator D gradually converge, and then the generator G has learned the latent distribution of the real data x… The objective function of GAN model is as follows:
PNG
media_image5.png
62
653
media_image5.png
Greyscale
the discriminator is trained accordining to multiple samples drawn from P_z and P_data corresponding to real and generated data. This training amounts to derivation and calculation of the claimed descrimination values and expected values. Checking for convergence is determining whther an outcome is maximized.) A4) when the outcome of the first objective function has not been maximized, adjusting parameters of the initial discriminator with fixed parameters of an initial denoising auto-encoder according to the first expected value; and A5) repeating the step A1) to the step A4) to train the initial discriminator until the outcome of the first objective function has been maximized, to generate the discriminator. (pg 13 Section 4.2.2 “Based on the above loss functions, the training process of our DAGAN model is an alternate iterative training between two GANs. The detailed process is as follows… According to the loss function (Equation (10)), the parameters of GEn, GDn1 and En1 are updated…. Finally, the first GAN and the second GAN are trained alternately for multiple rounds…. When two GANs converge, the training of our DAGAN model ends.” As shown above the objective functions with respect to the discriminator is maximized.)
Regarding claim 6
Chen teaches claim 5
Further Chen teaches, wherein the training process further comprises:
A6) deriving third discrimination values of the generated data and the at least one real data by the initial discriminator, and feeding the third discrimination values back to the denoising auto-encoder; A7) calculating a second expected value based on the third discrimination values by the denoising auto-encoder; A8) inputting the second expected value into a second objective function to determine whether an outcome of the second objective function is minimized; (pg 4-5 Section 3.1 also see pages 11-13 “Therefore, the generator G and discriminator D constitute a min–max game. After multiple alternating trainings, the generator G and discriminator D gradually converge, and then the generator G has learned the latent distribution of the real data x… The objective function of GAN model is as follows:
PNG
media_image5.png
62
653
media_image5.png
Greyscale
the discriminator is trained accordining to multiple samples drawn from P_z and P_data corresponding to real and generated data. The generator which forms the aotencoder in part is minimized according to the loss function) A9) when the outcome of the second objective function has not been minimized, adjusting the parameters of the initial denoising auto-encoder with fixed parameters of the discriminator and according to the second expected value;
and A10) repeating the step A6) to the step A9) to train the initial denoising auto-encoder until the outcome of the second objective function has been minimized, to generate the denoising auto-encoder. (pg 13 Section 4.2.2 “Based on the above loss functions, the training process of our DAGAN model is an alternate iterative training between two GANs. The detailed process is as follows… According to the loss function (Equation (10)), the parameters of GEn, GDn1 and En1 are updated…. Finally, the first GAN and the second GAN are trained alternately for multiple rounds…. When two GANs converge, the training of our DAGAN model ends.” As shown above the objective functions with respect to the autoencoder generator is minimized.)
Regarding claim 9
Chen teaches claim 1
Further Chen teaches, determining that the input data is abnormal when the abnormal degree is higher than a user-defined threshold (pg 6 “Therefore, the GANomaly model uses an anomaly score to measure this difference, defined as follows:
PNG
media_image6.png
47
297
media_image6.png
Greyscale
When the score of a sample is greater than a preset threshold µ, the sample is regarded as an outlier” pg 14 “Based on the above anomaly score, we can quickly identify whether a test sample is an outlier,… Because when both normal samples and marginal samples are tested, the anomaly score is smaller. Conversely, when anomalous samples are tested, the anomaly score becomes significantly larger” examiner notes the art makes clear that anomaly of a sample is made compared to a threshold. Pg 14 describes the abnormality degree equation when applied to the dualGAN.)
Claim Rejections - 35 U.S.C. § 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 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 of this title, 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) 7 is/are rejected under 35 U.S.C. § 103 as being unpatentable over Chen further in view of Ghojogh et al. “Generative Adversarial Networks and Adversarial Autoencoders: Tutorial and Survey”
Regarding claim 7
Chen teaches claim 1
Chen does not explicitly teach, adding noise to the normal data to generate noisy data; and converting the noisy data into the generative data by the initial discriminator, to input the generative data to the discriminator.
Ghojogh however when addressing adding noise to input data in a GAN system teaches, adding noise to the normal data to generate noisy data; and converting the noisy data into the generative data ( pg 28-29 “Universal approximator posterior: we concatenate the data point x and some noise η, with a fixed distribution such as Gaussian, as input to the encoder” pg 28 “The encoder of the autoencoder (i.e., block B1) is the generator G which generates the latent variable from the posterior distribution:” noise is added to the real data point, thus creating noisy data) by the initial discriminator, to input the generative data to the discriminator. ( pg 28 and Figure 13 pg 29 “The discriminator D (i.e., block B3) has a single output neuron with sigmoid activation function. It classifies the latent variable z to be a real latent variable from the prior distribution p(z) or a generated latent variable by the encoder of autoencoder”)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Gan discriminator system of Chen with the noise conditioned discriminator described by Ghojogh. One would have been motivated to make such a combination because both references relate to adversarial autoencoding of data. Further, as noted by Ghojogh “This adversarial learning makes both autoencoder and adversarial network stronger gradually because the autoencoder tries to generate the latent variable which is very similar to the real la tent variable from the prior distribution.” (pg 28 Ghojogh)
Claim(s) 8 is/are rejected under 35 U.S.C. § 103 as being unpatentable over Chen further in view of Yadav et al. Qualitative and Quantitative Evaluation of Multivariate Time-Series Synthetic Data Generated Using MTS-TGAN: A Novel Approach
Regarding claim 8
Chen teaches claim 5
Chen teaches, selecting data and parameters that match with the feature set as the training data set and providing the training data set to the initial discriminator and the initial denoising auto-encoder during the training process. (pg 14 Section 5 “In this paper, two publicly available industrial control system datasets are selected to train and verify our DAGAN model…. Distributed Smart Space Orchestration System (DS2OS). This dataset contains 357,952 samples… For simplicity, 20,000 normal samples in DS2OS dataset are used as the training set in this paper, 1500 normal samples and 350 abnormal samples are used as the test set (50 samples for each attack).” A subset of data is selected from a set of 357,952, this smaller subset corresponds to the features selected to match a set of features to be provided for the initial discriminator and auto-encoder during training.)
Chen does not explicitly teach, performing a pre-processing process, and the pre-processing process comprises: B1) selecting multiple influential features from statistical values of a training data set through a feature selection method to generate a feature set
Yadev however when addressing features selection for machine learning pre-processing teaches, performing a pre-processing process, and the pre-processing process comprises: B1) selecting multiple influential features from statistical values of a training data set through a feature selection method to generate a feature set ( pg 2 Introduction “The further novelty of our proposed model (MTS-TGAN) includes the inception of a pre processing layer, namely, a feature selector which outputs the important features …” pg 8 Section 4.2 “Prior to synthesizing the data, pre-processing must be ensured due to the varied nature of the data. The six signals’ values in Google stocks … fall into varying ranges and exhibit different attributes and characteristics. The Sklearns-Feature selector function … is used, which helps in removing the extra noise by outputting important features. Each feature in the dataset is then normalized individually using feature scaling, bringing all values into the range [0, 1], thus normalizing the dataset to ensure that the dataset is consistent. All features have the same format/range. We used the Scikit-MinMaxScaler function …and created rolling windows with overlapping sequences of 24 data points”)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify machine learning system of Chen with the feature preprocessing described by Yadev. One would have been motivated to make such a combination because both references address neural network data pre-processing for improved training. Further, as noted by Yadev “the important features [selected by the preprocessing layer] help in removing extra noise, which hinders the synthetic generation of data, as well as hyper-parameter tuning” (pg 2 Yadev)
Conclusion
Prior art not relied upon:
Tran et al. “Dist-GAN: An Improved GAN using Distance Constraints” described a Gan system with various constraints on the latent space of the autoencoder.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOHNATHAN R GERMICK whose telephone number is (571)272-8363. The examiner can normally be reached M-F 9:30-4:30.
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, Kakali Chaki can be reached on 571-272-3719. 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.
/J.R.G./
Examiner, Art Unit 2122
/KAKALI CHAKI/Supervisory Patent Examiner, Art Unit 2122
Prosecution Insights