DETAILED ACTION
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
2. Applicant's submission filed on 11 June 2026 [hereinafter Response] has been entered, where:
Claims 1, 2, 9, 14, 15, 17, and 20 have been amended.
Claims 1-20 are pending.
Claims 1-20 are rejected.
Claim Rejections – 35 U.S.C. § 112
3. The following is a quotation of 35 U.S.C. § 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
4. The rejection to claims 2, 14, 15, and 20 under 35 U.S.C. § 112(b) as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor regards as the invention is WITHDRAWN in view of the Applicant’s amendments to the claims.
Claim Rejections - 35 U.S.C. § 102
5. 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.
6. Claims 1, 3-5, 9, 11-13, and 17 are rejected under 35 U.S.C. § 102(a)(1) as being anticipated by Bazarbaev et al., "Generation of Time-Series Working Patterns for Manufacturing High-Quality Products through Auxiliary Classifier Generative Adversarial Network," MDPI (2021) [hereinafter Bazarbaev].
Regarding claims 1, 9 and 17, Bazarbaev teaches [a] method (Bazarbaev, Abstract), [a] system (Bazarbaev at p. 22, “5. Conclusions,” second paragraph), and [a] non-transitory machine-readable storage medium (Bazarbaev at p. 3, “1. Introduction,” third paragraph, teaches “We used the AC-GAN method to generate the TSWP based on the historical data from the melting and casting processes [(that is, “historical data” is inherently stored in memory, which is a non-transitory machine-readable storage medium)]”), comprising:
providing input data for inference operations of a first trained machine learning model (Bazarbaev, Fig. 4, teaches input data to a generator [Examiner annotations in dashed-line text boxes]:
PNG
media_image1.png
744
1203
media_image1.png
Greyscale
Bazarbaev at p. 7, “3.3. Problem Statement,” first paragraph, teaches “The generator model of the AC-GAN of the melting process has the following inputs: Auxiliary class input, product type of working cycle data; Auxiliary continuous input from raw material data; Latent space data (random normal) of length 150 [(that is, providing input data for inference operations of a first trained machine learning model )]”), the input data comprising:
first data comprising a random or pseudo-random input (see above, Bazarbaev, Fig. 4; Bazarbaev at p. 7, “3.3 Problem Statement,” first paragraph, teaches “[t]he generator model takes . . . random points from the latent space . . . to generate new synthetic data [(that is, first data comprising a random or pseudo-random input )]”); and
second data indicative of one or more attributes of target synthetic sensor time series data ((see above, Bazarbaev, Fig. 4; Bazarbaev at p. 7, “3.3 Problem Statement,” first paragraph, teaches “[t]he generator model takes class labels, . . . , and existing real data to generate new synthetic data [(that is, second data indicative of one or more attributes of target synthetic sensor time series data )]”)
, the second data comprising one or more labels identifying one or more of a data source of interest comprising a sensor type, sensor location, or processing recipe associated with a processing chamber, or a state of interest of the processing chamber (see above, Bazarbaev, Fig. 4; Bazarbaev at p. 10, “3.4.1 Data Preparation,” first paragraph, teaches “we use several raw data types as input to our model, including real-time data, working cycle data, raw material data, ingredient data, and milling data [(that is, the second data comprising one or more labels identifying one or more of a data source of interest comprising . . . processing receipt associated with a processing chamber, or a state of interest of the processing chamber)]”), associated with a rare event for which there are insufficient examples of real world data (Bazarbaev at p. 19, “4.2 Visual Comparison of Results,” second paragraph, teaches that for “both processes, the AC-GAN methods yield better results for different inputs, and the performance metrics indicate that the AC-GAN methods are not overfitted and can generate valuable TSWP even for the new input [(that is, an attribute of “not overfitted” is the training being associated with a rare event for which there are insufficient examples of real world data )]”);;
receiving an output from the first trained machine learning model based on the first data and the second data (see above Bazarbaev, Fig. 4; Bazarbaev at p. 15, “3.5 Generation of Time-Series Working Patterns using AC-GAN,” first paragraph, teaches the “input of the generator model for the melting process consists of a latent space (random normal noise), product type, and one row from the raw material data. The output of the generator is a synthetic TSWP of the melting process with the same shape [(that is, receiving an output from the first trained machine learning model based on the first data and the second data)]”),
wherein the output comprises synthetic sensor time series data associated with a processing chamber and the data source of interest or the state of interest (see above Bazarbaev, Fig. 4; Bazarbaev at p. 15, “3.5 Generation of Time-Series Working Patterns using AC-GAN,” first paragraph, teaches the “input of the generator model for the melting process consists of a latent space (random normal noise), product type, and one row from the raw material data. The output of the generator is a synthetic TSWP of the melting process with the same shape [(that is, wherein the output comprises synthetic sensor time series data associated with a processing chamber and the data source of interest)]“),
wherein the output is generated in view of the second data indicative of one or more attributes (see above Bazarbaev, Fig. 4; Bazarbaev at p. 15, “3.5 Generation of Time-Series Working Patterns using AC-GAN,” first paragraph, teaches the “output of the generator is a synthetic TSWP of the melting process with the same shape [(that is, “with the same shape” is wherein the output is generated in view of the second data indicative of one or more attributes)]“); and
utilizing the synthetic sensor time series data to train a second machine learning model to control operation of the processing chamber (Bazarbaev, Fig. 5, teaches a discriminator receiving a synthetic TSWP from the generator [Examiner annotations in dashed-line text boxes]:
PNG
media_image2.png
527
1189
media_image2.png
Greyscale
Bazarbaev at p. 9, “3.3 Problem Statement,” third paragraph, teaches the “discriminator [(that is, second machine learning model)] receives both real and generated data during training [(that is, utilizing the synthetic sensor time series data to train a second machi ne learning model to control operation of the processing chamber)], with a label of 1 for real data and 0 for generated data, as shown in Figure 5. The goal of the generator model is to generate [synthetic] TSWP data that cannot be distinguished from real data by the discriminator. If the discriminator cannot distinguish the generated data from the real data, it can be concluded that the generator generates similar data to real data”),
wherein training the second machine learning model comprises adjusting one or more parameters of the second machine learning model which cause the second machine learning model to provide target control signals when presented with current data associated with the rare event (Bazarbaev, Algorithm 2, teaches adjusting parameters of the second machine learning model through training [Examiner annotations in dashed-line text boxes]:
PNG
media_image3.png
720
1028
media_image3.png
Greyscale
Bazarbaev, Algorithm 2, step 9, teaches “generated_TWSP, Initialize: Discriminator D with parameter θd, where steps 14 & 15 teach “update [parameter] θd by minimizing Lc+Ls [(that is, wherein training the second machine learning model comprises adjusting one or more parameters of the second machine learning model which cause the second machine learning model to provide target control signals when presented with current data associated with the rare event)]”).
Regarding claims 3 and 11, Bazarbaev teaches all of the limitations of claims 1 and 9, respectively, as described above in detail.
Bazarbaev teaches –
wherein the first trained machine learning model comprises a generator of a generative adversarial network (see above Bazarbaev, Fig. 4; Bazarbaev, Abstract, teaches “We used an auxiliary classifier generative adversarial network (AC-GAN) model to generate time-series working patterns of two processes depending on product type and additional material data [(that is, wherein the first trained machine learning model comprises a generator of a generative adversarial network)]”).
Regarding claims 4 and 12, Bazarbaev teaches all of the limitations of claims 1 and 9, respectively, as described above in detail.
Bazarbaev teaches -
wherein the first trained machine learning model comprises a recurrent neural network model (Bazarbaev at pp. 4-5, “2.3 Deep Learning Methods,” first paragraph, teaches “a spatial and sequential deep learning approach to predict temperature distribution during the casting process. The authors claimed that casting quality is highly dependent on temperature distribution, and they proposed a model for predicting this distribution by integrating a CNN and a recurrent neural network (RNN) [(that is, wherein the first trained machine learning model comprises a recurrent neural network model)]”).
Regarding claim 5, Bazarbaev teaches all of the limitations of claim 1, as described above in detail.
Bazarbaev teaches –
wherein the synthetic sensor time series data comprises data corresponding to one or more of:
power, voltage, or current supplied to a component of the processing chamber;
pressure; or
temperature (Bazarbaev, Fig. 5; teaches a synthetic time series work pattern that includes electric power, amperes, voltage, frequency, and molten metal temperature [Examiner annotations in dashed-line text boxes]:
PNG
media_image4.png
528
1269
media_image4.png
Greyscale
Bazarbaev at p. 10, “3.4.1 Data Preparation,” first paragraph, teaches “The electric power variable corresponds to the primary user control variable of the melting process, and the current, voltage, and frequency variables correspond to machine variables related to electric induction in the induction furnace at the indicated time. Molten metal temperature is the temperature of the melted metal inside the furnace [(that is, wherein the synthetic sensor time series data comprises data corresponding to one or more of: power, voltage, or current supplied to a component of the processing chamber; . . . or temperature)]”).
Regarding claim 13, Bazarbaev teaches all of the limitations of claim 9, as described above in detail.
Bazarbaev teaches -
wherein the synthetic sensor time series data comprises data corresponding to one or more of:
power, voltage, or current supplied to one or more of:
a radio frequency plasma generation component;
a heater; or
a substrate support,
pressure; or
temperature (Bazarbaev, Fig. 5; teaches a synthetic time series work pattern that includes electric power, amperes, voltage, frequency, and molten metal temperature [Examiner annotations in dashed-line text boxes]:
PNG
media_image4.png
528
1269
media_image4.png
Greyscale
Bazarbaev at p. 10, “3.4.1 Data Preparation,” first paragraph, teaches “The electric power variable corresponds to the primary user control variable of the melting process, and the current, voltage, and frequency variables correspond to machine variables related to electric induction in the induction furnace at the indicated time [(that is, a heater)]. Molten metal temperature is the temperature of the melted metal inside the furnace [(that is, wherein the synthetic sensor time series data comprises data corresponding to one or more of: power, voltage, or current supplied to one or more of: . . . ; a heater; . . . or temperature)]”).
Claim Rejections - 35 U.S.C. § 103
7. 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.
8. 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.
9. Claims 2, 6-8, 10, 14-16, and 18-20 are rejected under 35 U.S.C. § 103 as being unpatentable over Bazarbaev et al., "Generation of Time-Series Working Patterns for Manufacturing High-Quality Products through Auxiliary Classifier Generative Adversarial Network," MDPI (2021) [hereinafter Bazarbaev] in view of US Published Application 20230104028 to Wang et al. [hereinafter Wang].
Regarding claim 2, Bazarbaev teaches all of the limitations of claim 1, as described above in detail.
Though Bazarbaev teaches training a model with synthetic time series working patterns (TSWP) data, Bazarbaev, however, does not explicitly teach –
training the first machine learning model, wherein training the first machine learning model comprises:
causing the first machine learning model to generate synthetic sensor time series data;
providing the synthetic sensor time series data to a third machine learning model comprising a discriminator;
providing measured sensor time series data to the third machine learning model,
wherein the third machine learning model is configured to distinguish between synthetic sensor time series data and measured sensor time series data;
providing feedback data to the first machine learning model, indicative of how accurately the third machine learning model distinguished synthetic from measured sensor time series data; and
updating the first machine learning model to generate synthetic sensor time series data that the third machine learning model less accurately distinguishes from measured sensor time series data.
But Wang teaches -
further comprising:
training the first machine learning model (Wang ¶ 0038 teaches “a functional generator 300 configured to generate multivariate continuous sensor curves 312 from training with arbitrary multivariate sensor data with irregular timestamps 310 received from one or more apparatuses [(that is, training the first machine learning model)]”), wherein training the first machine learning model comprises:
causing the first machine learning model to generate synthetic sensor time series data (Wang ¶ 0045 teaches “a functional processor [403] first deploys the fully connected neural network to map the random noises into random variables following a complex statistical distribution with tunable parameters. Next, the functional processor [403] combines the achieved random variable and the extracted patterns from 401 to produce new realizations of continuous time series that resemble the real sensor data corresponding to failures [(that is, causing the first machine learning model to generate synthetic sensor time series data)]”);
providing the synthetic sensor time series data to a third machine learning model (Wang, Fig. 4(a), teaches “flow of the F-GAN building module 200 [Examiner annotations in dashed-line text boxes]:”
PNG
media_image5.png
575
862
media_image5.png
Greyscale
Wang ¶ 0046 teaches “the functional discriminator 301 [(that is, a third machine learning model)] distinguishes the generated data [(that is, the synthetic sensor time series data)] from the actual data, given the time series with arbitrary granularities”) comprising a discriminator (Wang ¶ 0016 teaches “a functional generator and a functional discriminator, where the generator produces synthetic sensor data corresponding to failure events and the discriminator detects the fake data from the sensor data of actual failure [(that is, a third machine learning model comprising a discriminator)]”) ;
providing measured sensor time series data to the third machine learning model, wherein the third machine learning model is configured to distinguish between synthetic sensor time series data and measured sensor time series data (Wang ¶ 0046 teaches “the functional discriminator 301 distinguishes the generated data [(that is, synthetic sensor time series data)] from the actual data [(that is, measured sensor time series data)], given the time series with arbitrary granularities [(that is, providing measured sensor time series data to the second machine learning model, wherein the second machine learning model is configured to distinguish between synthetic sensor time series data and measured sensor time series data)]”; Wang ¶ 0047 teaches “the synthetic and real sensor data corresponding to failure events are provided into the MPFNN-based functional discriminator [301] that attempts to sort out the synthetic failure data;
providing feedback data to the first machine learning model (Wang ¶ 0018 teaches “providing feedback to the functional generator [300] to retrain the functional generator [300] [(that is, providing feedback data to the first machine learning model)]”), indicative of how accurately the third machine learning model distinguished synthetic from measured sensor time series data (Wang ¶ 0047 & Fig. 4(A) teaches “all the parameters in the above procedures are trained to solve the following min-max problem with objective function
PNG
media_image6.png
101
433
media_image6.png
Greyscale
where ‘FG’ and ‘FD’ are respectively the functional generator 300 and functional discriminator 301 [(that is, ”probability of being real” via a “min-max problem” is indicative of how accurately the second machine learning model distinguished synthetic from measured sensor time series data)]”); and
updating the first machine learning model (Wang ¶ 0038 teaches “providing feedback 302 to the functional generator to retrain the functional generator 300 [(that is, updating the first machine learning model)]”) to generate synthetic sensor time series data that the third machine learning model less accurately distinguishes from measured sensor time series data (Wang ¶ 0017 teaches “[Multi-Projection Functional Neural Network (MPFNN)] is a proposed failure predictive model building technique that is capable of handling the irregularity and temporal aspects within sensor data through the idea of basis projection and the BLUE [Best Linear Unbiased Estimation (BLUE)] technique. [Multi-Projection Functional Neural Network (MPFNN)] tends to have improved failure prediction accuracy (i.e., generate failure warning alerts when and only when a failure is approaching) due to the usage of multiple types of basic functions to more comprehensively represent the failure and non-failure sensor data [(that is, the “feedback” is the second machine learning model less accurately distinguishes from measured sensor time series data)]”).
Bazarbaev and Wang are from the same or similar field of endeavor. Bazarbaev teaches generating time-series working patterns of control variables for metal-melting induction furnaces and continuous casting machines. Wang teaches a failure prediction system for incoming failures (i.e., production failure that results in higher than expected product defect rate, industrial equipment failure) before they occur.
Thus, it would have been obvious to a person having ordinary skill in the art as of the effective filing date of the Applicant’s invention to modify Bazarbaev pertaining to synthetic time series work patterns for machine learning with the multiple machine learning prediction system of Wang.
The motivation to do so is to “improve the automation in effective information capturing when conducting functional data analysis, the [Multi-Projection Functional Neural Network (MPFNN)] utilizes multiple types of basis functions that cover various sorts of sensor data.” (Wang ¶ 0014).
Regarding claims 6, 14, and 20, Bazarbaev teaches all of the limitations of claims 1, 9, and 17, respectively, as described above in detail.
Though Bazarbaev teaches training a model with synthetic time series working patterns (TSWP) data, Bazarbaev, however, does not explicitly teach –
training a second machine learning model, wherein training the second machine learning model comprises:
providing the output synthetic sensor time series data to the second machine learning model as training input; and
providing first data indicative of one or more attributes associated with the output synthetic sensor time series data to the second machine learning model as target output,
wherein the second machine learning model is configured to predict attributes of the processing chamber based on measured sensor time series data of the processing chamber.
But Wang teaches -
wherein training the second machine learning model comprises:
providing the output synthetic sensor time series data to the second machine learning model as training input; and
providing third data indicative of one or more attributes associated with the output synthetic sensor time series data to the second machine learning model as target output (Wang ¶ 0016 teaches “[t]hese two components [(that is, first machine learning model and second machine learning model)] are trained simultaneously against the error of the discriminator in distinguishing fake data [(that is, providing the output synthetic sensor time series data to the second machine learning model as training input)] from real data [(that is, “real data” is providing first data . . . as target data)], until the error is maximized (i.e., the discriminator cannot tell the difference between fake and real data.” . . . The functional discriminator [(that is, the second machine learning model)] enables F-GAN to generate high-quality sensor time series, as it uses the MPFNN to enhance the discriminator's capability of detecting various sorts of differences between real and fake failure data. This forces the functional generator to improve its capacity in resembling patterns among the real failure data”),
wherein the second machine learning model is configured to predict attributes of the processing chamber based on measured sensor time series data of the processing chamber (Wang ¶ 0016 teaches “[t]he functional discriminator [(that is, the second machine learning model)] is capable of handling the irregularity and temporal aspects within sensor data through the idea of basis projection and the [Best Linear Unbiased Estimation (BLUE)] technique [(that is, “irregularity and temporal aspects” is to predict attributes of the processing chamber based on measured sensor time series data of the processing chamber)]”).
Bazarbaev and Wang are from the same or similar field of endeavor. Bazarbaev teaches generating time-series working patterns of control variables for metal-melting induction furnaces and continuous casting machines. Wang teaches a failure prediction system for incoming failures (i.e., production failure that results in higher than expected product defect rate, industrial equipment failure) before they occur.
Thus, it would have been obvious to a person having ordinary skill in the art as of the effective filing date of the Applicant’s invention to modify Bazarbaev pertaining to synthetic time series work patterns for machine learning with the multiple machine learning prediction system of Wang.
The motivation to do so is to “improve the automation in effective information capturing when conducting functional data analysis, the [Multi-Projection Functional Neural Network (MPFNN)] utilizes multiple types of basis functions that cover various sorts of sensor data.” (Wang ¶ 0014).
Regarding claims 7 and 15, the combination of Bazarbaev and Wang teaches all of the limitations of claims 6 and 15, respectively, as described above in detail.
Wang teaches -
wherein the second machine learning model is configured to detect one or more anomalies associated with measured sensor time series data of the processing chamber (Wang ¶¶ 0013-14 teaches “a failure prediction system equipped with a new AI architecture that effectively and efficiently addresses these challenges [(that is, “failure prediction system” is a second machine learning model is configured to detect one or more anomalies)]. . . . {A] proposed AI model that serves as the core algorithm within failure prediction systems in industrial IoTs [(that is, “industrial IoTs” is measured sensor time series data of the processing chamber)]. The proposed AI model consists of an innovative time series data balancing technique called the Functional Generative Adversarial Network (F-GAN) and a new failure predictive model called the Multi-Projection Functional Neural Network (MPFNN) [(that is, the “F-GAN” is the second machine learning model is configured to detect one or more anomalies)]”).
Regarding claims 8, 16, and 19, Bazarbaev teaches all of the limitations of claims 1, 16, and 17, respectively, as described above in detail.
Though Bazarbaev teaches producing synthetic time series working patterns (TSWP) data to train a model, Bazarbaev, however, does not explicitly teach –
wherein an attribute of target synthetic sensor time series data comprises one or more of:
time since installation of the processing chamber;
time since a previous maintenance event of the processing chamber; or
a fault present in the processing chamber/
But Wang teaches -
wherein an attribute of target synthetic sensor time series data comprises one or more of:
time since installation of the processing chamber;
time since a previous maintenance event of the processing chamber; or
a fault present in the processing chamber (Wang ¶ 0016 teaches “where the generator produces synthetic sensor data corresponding to failure events [(that is, an attribute of target synthetic sensor time series)] and the discriminator detects the fake data from the sensor data of actual failures [(that is, an attribute of target synthetic sensor time series data comprises one or more of: . . . a fault present in the processing chamber)]”).
Bazarbaev and Wang are from the same or similar field of endeavor. Bazarbaev teaches generating time-series working patterns of control variables for metal-melting induction furnaces and continuous casting machines. Wang teaches a failure prediction system for incoming failures (i.e., production failure that results in higher than expected product defect rate, industrial equipment failure) before they occur.
Thus, it would have been obvious to a person having ordinary skill in the art as of the effective filing date of the Applicant’s invention to modify Bazarbaev pertaining to synthetic time series work patterns for machine learning with the multiple machine learning prediction system of Wang.
The motivation to do so is to “improve the automation in effective information capturing when conducting functional data analysis, the [Multi-Projection Functional Neural Network (MPFNN)] utilizes multiple types of basis functions that cover various sorts of sensor data.” (Wang ¶ 0014).
Regarding claims 10 and 18, Bazarbaev teaches all of the limitations of claims 9 and 17, respectively, as described above in detail.
Though Bazarbaev teaches producing synthetic time series working patterns (TSWP) data to train a model, Bazarbaev, however, does not explicitly teach –
train the first machine learning model, wherein training the model comprises:
causing the first machine learning model to generate synthetic sensor time series data;
providing the synthetic sensor time series data to a second machine learning model;
providing measured sensor time series data to the second machine learning model,
wherein the second machine learning model is configured to distinguish between synthetic sensor time series data and measured sensor time series data;
providing feedback data to the first machine learning model indicative of how accurately the second machine learning model distinguished synthetic from measured sensor time series data; and
updating the first machine learning model to generate synthetic sensor time series data that the second machine learning model less accurately distinguishes from measured sensor time series data.
But Wang teaches -
further comprising:
training the first machine learning model (Wang ¶ 0038 teaches “a functional generator 300 configured to generate multivariate continuous sensor curves 312 from training with arbitrary multivariate sensor data with irregular timestamps 310 received from one or more apparatuses [(that is, training the first machine learning model)]”), wherein training the first machine learning model comprises:
causing the first machine learning model to generate synthetic sensor time series data (Wang ¶ 0045 teaches “a functional processor [403] first deploys the fully connected neural network to map the random noises into random variables following a complex statistical distribution with tunable parameters. Next, the functional processor [403] combines the achieved random variable and the extracted patterns from 401 to produce new realizations of continuous time series that resemble the real sensor data corresponding to failures [(that is, causing the first machine learning model to generate synthetic sensor time series data)]”);
providing the synthetic sensor time series data to a third machine learning model (Wang, Fig. 4(a), teaches “flow of the F-GAN building module 200 [Examiner annotations in dashed-line text boxes]:”
PNG
media_image5.png
575
862
media_image5.png
Greyscale
Wang ¶ 0046 teaches “the functional discriminator 301 [(that is, a second machine learning model)] distinguishes the generated data [(that is, the synthetic sensor time series data)] from the actual data, given the time series with arbitrary granularities”);
providing measured sensor time series data to the third machine learning model, wherein the third machine learning model is configured to distinguish between synthetic sensor time series data and measured sensor time series data (Wang ¶ 0046 teaches “the functional discriminator 301 distinguishes the generated data [(that is, synthetic sensor time series data)] from the actual data [(that is, measured sensor time series data)], given the time series with arbitrary granularities [(that is, providing measured sensor time series data to the second machine learning model, wherein the second machine learning model is configured to distinguish between synthetic sensor time series data and measured sensor time series data)]”; Wang ¶ 0047 teaches “the synthetic and real sensor data corresponding to failure events are provided into the MPFNN-based functional discriminator [301] that attempts to sort out the synthetic failure data;
providing feedback data to the first machine learning model (Wang ¶ 0018 teaches “providing feedback to the functional generator [300] to retrain the functional generator [300] [(that is, providing feedback data to the first machine learning model)]”), indicative of how accurately the third machine learning model distinguished synthetic from measured sensor time series data (Wang ¶ 0047 & Fig. 4(A) teaches “all the parameters in the above procedures are trained to solve the following min-max problem with objective function
PNG
media_image6.png
101
433
media_image6.png
Greyscale
where ‘FG’ and ‘FD’ are respectively the functional generator 300 and functional discriminator 301 [(that is, ”probability of being real” via a “min-max problem” is indicative of how accurately the second machine learning model distinguished synthetic from measured sensor time series data)]”); and
updating the first machine learning model (Wang ¶ 0038 teaches “providing feedback 302 to the functional generator to retrain the functional generator 300 [(that is, updating the first machine learning model)]”) to generate synthetic sensor time series data that the third machine learning model less accurately distinguishes from measured sensor time series data (Wang ¶ 0017 teaches “[Multi-Projection Functional Neural Network (MPFNN)] is a proposed failure predictive model building technique that is capable of handling the irregularity and temporal aspects within sensor data through the idea of basis projection and the BLUE [Best Linear Unbiased Estimation (BLUE)] technique. [Multi-Projection Functional Neural Network (MPFNN)] tends to have improved failure prediction accuracy (i.e., generate failure warning alerts when and only when a failure is approaching) due to the usage of multiple types of basic functions to more comprehensively represent the failure and non-failure sensor data [(that is, the “feedback” is the second machine learning model less accurately distinguishes from measured sensor time series data)]”).
Bazarbaev and Wang are from the same or similar field of endeavor. Bazarbaev teaches generating time-series working patterns of control variables for metal-melting induction furnaces and continuous casting machines. Wang teaches a failure prediction system for incoming failures (i.e., production failure that results in higher than expected product defect rate, industrial equipment failure) before they occur.
Thus, it would have been obvious to a person having ordinary skill in the art as of the effective filing date of the Applicant’s invention to modify Bazarbaev pertaining to synthetic time series work patterns for machine learning with the multiple machine learning prediction system of Wang.
The motivation to do so is to “improve the automation in effective information capturing when conducting functional data analysis, the [Multi-Projection Functional Neural Network (MPFNN)] utilizes multiple types of basis functions that cover various sorts of sensor data.” (Wang ¶ 0014).
Response to Arguments
10. Examiner has fully considered Applicant’s arguments, and responds below accordingly.
35 U.S.C. § 101
11. “Applicants respectfully submit that pursuant to the subject matter eligibility analysis under the PEG, the claims are not directed to a judicial exception and are patent eligible for at least the following reasons. While it is not contested that the claims are respectively directed to a process and machine, the Examiner asserts that claims 2, 6-7, 10, 14-15, 18, and 20 are directed to the judicial exception of an abstract idea. However, claims 2, 6-7, 10, 14-15, 18, and 20 do not recite concepts that explicitly fall into the abstract idea exception categories of (a) mathematical concepts, (b) certain methods of organizing human activity, or (c) mental processes. Therefore for at least this reason, the eligibility analysis should end at the first prong of Revised Step 2A. Even assuming arguendo that it could be determined that the claims somehow do explicitly recite a judicial exception, each of the claims integrates such recitation into a practical application and, therefore, should be deemed to be not "directed to" the patent ineligible judicial exception in accordance with the second prong of the Revised Step 2A. Specifically, the practical application of the claims lies in generating a model that provides for improved outcomes for manufacturing, and that in particular is able to identify and mitigate uncommon negative events for which there is an insufficient amount of real world training data to successfully train a machine learning model, as described for example in paragraphs [0025], [0048], and [0059] of the current specification. The techniques described in the Specification enable generation of a model for predictive, control, or other applications with respect to manufacturing equipment without an undue burden of collecting a large amount of live data for model training. For example, claim 1, as amended, is directed to:
A method, comprising:
providing input data for inference operations of a first trained machine learning model, the input data comprising:
first data comprising a random or pseudo-random input, and
second data indicative of one or more attributes of target synthetic sensor time series data, the second data comprising one or more labels identifying one or more of a data source of interest comprising a sensor type, sensor location, or processing recipe associated with a processing chamber, or a state of interest of the processing chamber, associated with a rare event for which there are insufficient examples of real world data;
receiving an output from the first trained machine learning model based on the first data and the second data, wherein the output comprises synthetic sensor time series data associated with a processing chamber and the data source of interest or the state of interest, wherein the output is generated in view of the second data indicative of one or more attributes; and
utilizing the synthetic sensor time series data to train a second machine learning model to control operation of the processing chamber, wherein training the second machine learning model comprises adjusting one or more parameters of the second machine learning model which cause the second machine learning model to provide target control signals when presented with current data associated with the rare event.
[(claim 1, lines 21-23 (emphasis added by Examiner))]. Applicant respectfully submits that the claims of the present application are analogous to the claims in Ex Parte Desjardins. As outlined in MPEP § 2106.04(d), subsection III, in Ex Parte Desjardins, the Appeals Review Panel determined that "improvements as to how the machine learning model itself operates, including training a machine learning model to learn new tasks" include "at least the limitation of' adjust the first values of the plurality of parameters to optimize performance of the machine learning model"' reflects an improvement, and constitutes integration into a practical application at Step 2A Prong Two. The independent claims of the current application, as amended, similarly recite
[wherein training the second machine learning model comprises] adjusting one or more parameters of the second machine learning model which cause the second machine learning model to provide target control signals when presented with current data associated with the rare event."
Similar to the situation in Ex Parte Desjardins, the claims reflect an improvement that constitutes integration into a practical application of Step 2A, Prong Two because like in Ex Parte Desjardins, that claims training a machine learning model to perform a series of tasks while protecting knowledge about previous tasks, the current claims enable the model to function as intended, even when presented with current data corresponding to an event that real-world training data describes in a volume insufficient to provide adjustments to model parameters during training operations to enable target performance of the model. Accordingly, Applicant respectfully requests that the rejections of claims 2, 6-7, 10, 14- 15, 18, and 20 under 35 U.S.C. § 101 be withdrawn.
In view of the above, claim 1 is not directed to an abstract idea. Therefore, Applicant respectfully submits that independent claim 1 is directed to patent eligible subject matter at least under the second prong of Step 2A. Dependent claims 2, 6-7, 10, 14-15, 18, and 20 are also directed to patent eligible subject matter at least by virtue of their respective dependencies from claim 1, and the similarly amended claims 9 and 17.” (Response at pp. 11-13).
Examiner Response:
Examiner finds Applicant’s arguments and amendments persuasive, and accordingly, the rejection under Section 101 is WITHDRAWN.
35 U.S.C. § 103
12. Applicant submits that “Wang is directed to a system for failure prediction for industrial systems (Wang, Abstract). The Office Action recognizes that Wang does not teach or suggest "the second data comprising one or more labels identifying one or more of a data source of interest comprising a sensor type, sensor location, or processing recipe associated with a processing chamber, or a state of interest of the processing chamber" (Office Action, Page 22).
The Office Action, however, asserts that Soni alleviates deficiencies of Wang. Applicant respectfully disagrees.
Soni does not remedy the shortcomings of Wang with respect to claim 1, as amended. Soni is directed to medical machine synthetic data and corresponding event generation (Soni, Abstract). Soni teaches use of a ‘generative model’ to ‘generate additional synthetic data’ (Soni, paragraph [0091]). However, Soni does not teach ‘providing input data’ to a model comprising ‘one or more attributes of target synthetic sensor time series data,’ with output of the model being ‘based on’ this data, as recited in amended claim 1. Specifically, Soni teaches a method where a model outputs the ‘synthetic data,’ where the output ‘can include a channel in which synthetic events, labels, and/or other annotations associated with the series data is also generated' (Soni, para. [0091], Emphasis added). Soni teaches a system which outputs synthetic data and outputs associated labels. In contrast, the claimed invention provides ‘one or more labels identifying one or more of a data source of interest comprising a sensor type, sensor location, or processing recipe associated with a processing chamber, or a state of interest of the processing chamber’ are input to the synthetic data generation model, which then outputs synthetic data associated with the input labels, as recited in amended claim 1. Thus, Soni does not cure the deficiencies of Wang.
Similar language is also included in independent claims 9 and 17. Thus, the combination of cited references does not teach or suggest all the features of the independent claims 1, 9 and 17, and corresponding dependent claims. Applicant respectfully requests the rejection of claims 1-20 under 35 U.S.C. § 103 be withdrawn.” (Response at pp. 14-15).
Examiner’s Response:
Examiner respectfully submits that the Applicant’s arguments to the cited prior art of Soni is moot in view of the new grounds of rejection the presented in this Office action are necessitated by the Applicant’s amendments to the claims.
Conclusion
13. 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.
14. The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
(US Published Application 20180260697 to Sun et al.) teaches facilitating machine learning using multi-dimensional time series data are provided. In one example, a system includes a snapshot component and a machine learning component. The snapshot component generates a first sequence of multi-dimensional time series data and a second sequence of multi-dimensional time series data from multi-dimensional time series data associated with at least two different data types generated by a data system over a consecutive period of time. The machine learning component that analyzes the first sequence of multi-dimensional time series data and the second sequence of multi-dimensional time series data using a convolutional neural network system to predict an event associated with the multi-dimensional time series data.
(Dixit et al., “Intelligent Fault Diagnosis of Rotary Machines: Conditional Auxiliary Classifier GAN Coupled With Meta Learning Using Limited Data,” IEEE (2021)) teaches generative adversarial networks (GANs) are capable to generate high-quality synthetic samples. However, the problem still persists with the training of GAN using limited fault samples that are present in practical conditions. This article proposes a novel conditional auxiliary classifier GAN framework coupled with model agnostic meta learning (MAML) to resolve this problem. The objective is to initialize and update the network parameters using MAML instead of regular stochastic gradient learning. This modification enables GAN to learn the task of synthetic sample generation using the limited training dataset.
15. Any inquiry concerning this communication or earlier communications from the Examiner should be directed to KEVIN L. SMITH whose telephone number is (571) 272-5964. Normally, the Examiner is available on Monday-Thursday 0730-1730.
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 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://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/K.L.S./
Examiner, Art Unit 2122
/KAKALI CHAKI/Supervisory Patent Examiner, Art Unit 2122
Prosecution Insights