DETAILED ACTION
This action is in response to claims filed 21 May 2024 for application 18669844 filed 21 May 2024. Currently claims 1-10 are pending.
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 .
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-4 and 6-10 are rejected under 35 U.S.C. 103 as being unpatentable over Andoni et al. (US 20210182691) in view of Hasegawa et al. (Bezier Curve Parametrization Using a Multiobjective Evolutionary Algorithm).
Regarding claims 1, 9 and 10, Andoni discloses:
A method for determining a machine learning model architecture for detecting anomalies in sensor signals, comprising:
generating a population for an evolutionary algorithm, wherein each individual of the population is a machine learning model obtained based on an encoding of the machine learning model (“The genetic algorithm 110 may automatically assign layer types, layer counts, node counts per layer, activation functions, aggregation functions, biases, connection weights, etc. to each model of the input set 120 for the initial epoch. In some aspects, the connection weights are assigned randomly or pseudo-randomly. In some implementations, a single activation function is used for each node of a particular model or each node of a particular layer. For example, a sigmoid function may be used as the activation function of each node of the particular model or layer. The single activation function may be selected based on configuration data. For example, the configuration data may indicate that a hyperbolic tangent activation function is to be used or that a sigmoid activation function is to be used. Alternatively, the activation function may be randomly or pseudo-randomly selected from a set of allowed activation functions, and different nodes or layers of a model may have different types of activation functions. In other implementations, the activation function assigned to each node may be randomly or pseudo-randomly selected (from the set of allowed activation functions) for each node of the particular model. Aggregation functions may similarly be randomly or pseudo-randomly assigned for the models in the input set 120 of the initial epoch. Thus, the models of the input set 120 of the initial epoch may have different architectures and different connection weights.” [0066]) … of neurons and a number of layers of the respective machine learning model (“During a configuration stage of operation, a user can specify the input data 102 and a particular data field or a set of data fields in the input data 102 to be modeled. The data field(s) to be modeled may correspond to input nodes and output nodes of an autoencoder that is to be generated by the system 100. In particular implementations, the user can also configure other aspects of the model. For example, the user may provide input to indicate a particular data field of the data set that is to be included in the model or a particular data field of the data set that is to be omitted from the model. As another example, the user may provide input to constrain allowed autoencoder architectures. To illustrate, the model may be constrained to include a particular number of nodes in the latent-space layer, an hourglass architecture, or a mirrored architecture. Alternatively, the architecture of the model can be unconstrained, the number of latent-space layer nodes can be unspecified, or both. Various network topologies or architectures of autoencoders are described in more detail below. An advantage of allowing the genetic algorithm to generate unconstrained, non-hourglass, non-mirrored, or other novel architectures is that such architectures can reduce processing resources used to monitor equipment, can detect anomalous behavior earlier, more accurately, or both.” [0060]);
performing detection of anomalies by each individual using a training data set to determine and assign an error value of each individual to the individual, wherein the error value determines an accuracy of detection of anomalies of the respective individual based on a complexity of the respective individual (“In FIG. 3, to magnify the distinction between the reconstruction loss for normal operational data and the reconstruction loss for anomalous operational data, the genetic algorithm 110 and the optimization trainer 160 use different portions 104, 106 of the input data 102 and have different goals for the respective reconstruction losses 316, 336. For example, the genetic algorithm selects as fitter those autoencoders 134 of the input set 120 that have higher reconstruction loss 316 based on the first portion 104 of the input data 102, where the first portion 104 of the input data 102 corresponds to abnormal-state data. In contrast, the optimization trainer 160 trains one or more trainable autoencoders received from the genetic algorithm 110 to reduce reconstruction loss 336 based on the second portion 106 of the input data 102, where the second portion 106 corresponds to normal-state operational data.” [0051], “Input data samples 332 from the second portion 106 of the input data 102 are provided as input to one or more trainable autoencoders 132. Each of the one or more trainable autoencoders 132 generates an output data sample 334 corresponding to each input data sample 332. The loss function evaluation 170 calculates the reconstruction loss 336 for each input data sample 332/output data sample 334 pair. The loss function evaluation 170 may aggregate or average a set of reconstruction losses 336 for various input data sample 332/output data sample 334 pairs to generate an aggregate or average reconstruction loss 336 for each of the one or more trainable autoencoders 132.” [0055]);
performing a modification of the individuals according to the evolutionary algorithm (“The selection operation 318 selects a subset (e.g., one or several) of the one or more autoencoders 134 on which to perform one or more of the evolutionary operations 150. The selection is based on relative fitness of each of the one or more autoencoders 134. In FIG. 3, the selection operation 318 uses the reconstruction loss 316 as a fitness metric (e.g., as the fitness function value 144). In general, fitter autoencoders are selected for mutation and crossover operations 152, 154. Additionally, less fit autoencoders may be selected for extinction 156. As explained above, in the example illustrated in FIG. 3, the selection operation 318 considers a first autoencoder to be fitter than a second autoencoder if the first autoencoder has higher reconstruction loss 316 than the second autoencoder based on the first portion 104 of the input data 102. The selected subset of the input set 120 is subjected to the evolutionary operations 150, which are described in greater detail below, and one or more trainable autoencoders 132 resulting from the evolutionary operations 150 are provided to the optimization trainer 160.” [0054]); and
selecting one of the individuals based on repeatedly performing the detection and modification based on the error value (“When the termination criterion is satisfied, at 1320, the method 1300 may include selecting and outputting a fittest model, at 1322, and the method 1300 may end, at 1324. In some implementations, the selected model may be subjected to a final training operation, e.g., by the optimization trainer or by another trainer, before being output.” [0119]).
Andoni does not explicitly disclose: using a parametrically modeled curve, wherein the respective parametrically modeled curve represents a relationship between a number. Hasegawa teaches: using a parametrically modeled curve, wherein the respective parametrically modeled curve represents a relationship between a number (“This work proposed a parameter optimization method for the Bezier Curve Fitting problem. This approach uses a multiobjective Evolutionary Algorithm with two objective functions, minimizing error and length to find the best fit for a set of data points. Test results showed that the proposed method can produce better looking solutions than single objective methods when working with small number of data points. For sets of tightly-packed data points, the proposed method can perform similarly to others methods.” P17 §6 ¶1).
Andoni and Hasegawa are in the same field of endeavor of genetic algorithms and are analogous. Andoni discloses genetic algorithms for optimizing autoencoders. Hasegawa teaches the use of parametrically modeled Bezier curve for guiding a genetic algorithm. It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the genetic algorithm autoencoder optimization of Andoni with the Bezier curve guided genetic algorithm to yield predictable results of better solutions.
Regarding claim 2, Andoni discloses: The method according to claim 1, wherein:
the machine learning model comprises an autoencoder (Fig 9a),
the autoencoder comprises an encoder module and a decoder module (Fig 9a),
encoding is performed by the encoder module and the decoder module is configured to perform decoding (Fig 9a).
Regarding claim 3, Andoni discloses: The method according to claim 2, wherein:
the encoding comprises a dimensional reduction of training data of the training data set in order to obtain a compressed representation of the training data (“The autoencoder is configured to be relatively accurate (e.g., to have low reconstruction loss) when reconstructing input data during normal operation of the device and to be relatively inaccurate (e.g., to have high reconstruction loss) when reconstructing input data during anomalous operation of the device. Thus, the reconstruction loss associated with particular sensor data can be used as an indication of whether the device is operating in a normal operational mode or an anomalous operational mode. An advantage of monitoring the device in this manner is that all operational states of the device do not need to be characterized in advance, which can be time consuming and expensive. Rather, it is sufficient to identify from among a set of sensor data, one or more subsets representing abnormal or anomalous operations. Such data is often easy to isolate from historical data by selecting data immediately preceding a known device failure or fault condition.” [0050]),
the decoding comprises reconstruction of the dimension-reduced training data to obtain a reconstruction of the compressed representation of the training data [0050], and
the detection of anomalies is performed based on a comparison between the training data and the reconstruction of the compressed representation of the training data [0050].
Regarding claim 4, Andoni does not explicitly disclose, however, Hasegawa teaches: The method according to claim 1, wherein the parametrically modeled curve is a quadratic Beziér curve (“This work proposed a parameter optimization method for the Bezier Curve Fitting problem. This approach uses a multiobjective Evolutionary Algorithm with two objective functions, minimizing error and length to find the best fit for a set of data points. Test results showed that the proposed method can produce better looking solutions than single objective methods when working with small number of data points. For sets of tightly-packed data points, the proposed method can perform similarly to others methods.” P17 §6 ¶1, “The curve for the first example was constructed with two straight line segments (in the shape of a ‘7’) followed by a quadratic curve segment. Another particularity is that the number of data points in this example is very low. This test fits a single 6th-Order Bezier Curve segment and aims to demonstrate how the proposed method performs with unevenly spaced data points.” P11 §5 ¶4).
Regarding claim 6, Andoni discloses: The method according to claim 1, wherein:
the modification of the individuals according to the evolutionary algorithm comprises a gene transfer with the following steps:
dividing the population into two halves, wherein the individuals of a superior half have a lower error value than the individuals of an inferior half (Fig 5&6), and
transferring the parameters of the parametrically modeled curve of a randomly selected individual of the superior half to a randomly selected individual of the inferior half (Fig 5&6).
Regarding claim 7, Andoni discloses: The method according to claim 1, wherein:
the error value is determined based on a deviation and a hyperparameter and is calculated according to the following formula, wherein λ is the hyperparameter:
error value=deviation+λ*complexity
the deviation represents the accuracy of the detection of anomalies and the complexity represents a ratio between a number of trainable parameters of the machine learning model and a maximum possible number of trainable parameters of the machine learning model (“Continuing to a second stage 450 of operation, each model of the input set 120 may be tested based on the first portion 104 of input data 102 to determine model fitness. For example, the first portion 104 of input data 102 may be provided as input data samples 312 to each model, which processes the input data samples 312 (according to the architecture, connection weights, activation function, etc., of the respective model) to generate corresponding output data samples 314. Input data sample/output data sample pairs are evaluated to determine how well the autoencoder of the model reconstructed the first portion 104 of the input data 102. The fitness evaluation operation 140 evaluates the fitness of each model based on performance (e.g., reconstruction loss) of the model, complexity (or sparsity) of the model, or a combination thereof.” [0067], hyperparameters [0039]).
Regarding claim 8, Andoni discloses: The method according to claim 1, wherein a computer program comprises instructions that, when the computer program is executed by a computer, cause the computer to carry out the method (Fig 19).
Claim 5 is/are rejected under 35 U.S.C. 103 as being unpatentable over Andoni in view of Hasegawa and further in view of Rudland et al. (US 20120021768).
Regarding claim 5, Andoni does not explicitly disclose, however, Rudland teaches: The method according to claim 1, wherein:
the modification of the individuals according to the evolutionary algorithm comprises a mutation with the following steps:
generating at least two clones for each original individual (“A predetermined number of high fitness individuals are selected for cloning, and are cloned according to their fitness. The better the fitness of the individual, the more clones of that individual are produced. For example, the highest fitness individual can be cloned into five clones, the next highest into three clones, and the next highest into one clone. Those skilled in the art will appreciate that other cloning strategies to determine the number of clones can be used as desired for a particular application.” [0065]),
modifying at least one parameter of the parametrically modeled curve for at least one clone, wherein one of the clones remains unchanged (“The clones are mutated, with the degree of mutation inversely proportional to the fitness of the clone. The better the fitness of the clone, the less mutation that is needed to move the clone toward the optimum solution. The distance from the clone to the mutated node depends on the fitness of the clone. In one embodiment, the mutated node is selected a distance away from the clone that is inversely proportional to the fitness of the clone. The fitter the clone, the closer it is to the desired solution, and so the less mutation needed to get to the solution. In another embodiment, the constituent components of the FNerror calculation such as the error from each of the field devices are analyzed and the mutated node is shifted in distance and direction to reduce the FNerror value.” [0066]),
performing the detection of anomalies by each clone using the training data set to determine the error value of each clone and assign it to the clone [0066],
transferring the at least one modified parameter of the clone with the lowest error value to the at least one further clone [0066],
repeating the modification, performing, detection, and transfer until each parameter of the parametrically modeled curve has been modified once to determine an optimized clone [0066], and
replacing the respective original individual with the optimized clone [0066].
Andoni, Hasegawa and Rudland are in the same field of endeavor of genetic algorithms and are analogous. Andoni discloses genetic algorithms for optimizing autoencoders. Hasegawa teaches the use of parametrically modeled Bezier curve for guiding a genetic algorithm. Rudland discloses genetic algorithms with cloning. It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the genetic algorithm autoencoder optimization based on parametric curve of Andoni and Hasegawa with the clone based genetic algorithm as taught by Rudland to yield predictable results of fitter solutions.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ERIC NILSSON whose telephone number is (571)272-5246. The examiner can normally be reached M-F: 7-3.
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, James Trujillo can be reached at (571)-272-3677. 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.
/ERIC NILSSON/ Primary Examiner, Art Unit 2151
Prosecution Insights