DETAILED ACTION
Remarks
Claims 1-14 have been examined and rejected. This Office Action is responsive to the continued examination request.
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 .
Claims 1-14 are presented for examination.
Respond to Amendment
The amendment filed 10/29/2025 has been entered. Claims 1, 13 and 14 have been amended. Claims 1-14 are pending in the application.
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 1-14 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Independent claims
Step 1
Claim 1 is drawn to a computer-implemented method, claim 13 is drawn to a device and claim 14 is drawn to a non-transitory computer-readable medium storing a computer program for a machine learning model when executed by a computer to perform the method of claim 1. Therefore, each of these claim groups falls under one of four categories of statutory subject matter (process/method, machines/product/apparatus, manufactures, and composition of matter).
Step 2A – Prong 1
Claims 1, 13 and 14 are directed to a judicially recognized exception of an abstract idea without significantly more.
Claims 1, 13 and 14 recite a method of determining a second parameter for assigning the second parameter to the first parameter in a first iteration of learning that under its broadest reasonable interpretation enumerates a mental concept. A human can mentally perform, with the physical aid such as pen and paper, to determine a parameter in an iteration. Therefore, the step of determining a second parameter for assigning the second parameter is nothing more than a mental concept (MPEP 2106.04(a)(2)(III)).
Claims 1, 13 and 14 recite a method of determining a third parameter that under its broadest reasonable interpretation enumerates a mental concept. A human can mentally perform, with the physical aid such as pen and paper, to determine a parameter. Therefore, the step of determining a third parameter is nothing more than a mental concept (MPEP 2106.04(a)(2)(III)).
Claims 1, 13 and 14 recite a method of determining a rate for changing the first parameter in at least one iteration of learning depending on the third parameter and depending on a measure for evaluating the solution of the task that under its broadest reasonable interpretation enumerates a mental concept. A human can mentally perform, with the physical aid such as pen and paper, to determine a parameter for determining a rate for changing a parameter. Therefore, the step of determining a third parameter for determining a rate for changing the first parameter is nothing more than a mental concept (MPEP 2106.04(a)(2)(III)).
Step 2A – Prong 2
Claims 1, 13 and 14 recite further a method of wherein the determining of the second parameter or third parameter includes determining a solution of an initial value problem that depends on a derivative of the measure with respect to the first parameter that fails to integrate the abstract idea into a practical application. The step of determining a solution of an initial value problem is a form of insignificant input and output solution activities, where determining a solution of an initial value problem is necessary for all uses of the judicial exception. This additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (MPEP 2106.05(g)).
Claims 1, 13 and 14 recite further a method of wherein the determining of the solution of the initial value problem includes determining a first part of the solution of the initial value problem depending on an initial value, determining a second part of the solution of the initial value problem depending on the first part, determining for the first part a partial derivative, determining for the second part a partial derivative, and determining the second parameter and/or the third parameter depending on at least one of the partial derivatives those fail to integrate the abstract idea into a practical application. The steps of determining a first part of the solution of the initial value, determining a second part of the initial value, determining a partial derivative for the first part, determining a partial derivative for the second part and determining the second parameter or the third parameter are forms of insignificant input and output solution activities, where determining a first part of the solution of the initial value, determining a second part of the initial value, determining a partial derivative for the first part, determining a partial derivative for the second part and determining the second parameter or the third parameter are necessary for all uses of the judicial exception. These additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea (MPEP 2106.05(g)).
Claims 1, 13 and 14 recite further a method of wherein the machine learning is applied to adapt an anomaly detection model to a given set of anomaly detection tasks of audio recordings from a set of similar microphones, wherein the anomaly detection model is an autoencoder, wherein the method of machine learning is used to adapt a learning rule used to train the autoencoder such that autoencoder learns to detect anomalies in recordings from the similar microphones those fail to integrate the abstract idea into a practical application. The steps of adapt an anomaly detection model and adapt a learning rule used to train the autoencoder are forms of insignificant input and output solution activities, where adapting an anomaly detection model and adapting a learning rule to the autoencoder are necessary for all uses of the judicial exception. These additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea (MPEP 2106.05(g)).
Step 2B
The additional elements in step 2A-Prong 2 those are forms of insignificant extra-solution activities, do not amount to significantly more than an abstract idea because the court decision have determined that these additional elements of determining a solution of an initial value problem, determining a first part of the solution of the initial value, determining a second part of the initial value, determining a partial derivative for the first part, determining a partial derivative for the second part and determining the second parameter or the third parameter; adapting an anomaly detection model and adapting a learning rule to the autoencoder to be well-understood, routine, and conventional when claimed in a merely generic manner (MPEP 2106.05(d)(II)).
As such, claims 1, 13 and 14 are not patent eligible.
Dependent claims
Claims 2-12 merely narrow the previously recited abstract idea limitations. For the reasons described above with respect to claim 1, this judicial exception is not meaningfully integrated into a practical application, or significantly more than the abstract idea. The claims disclose similar limitations described for the independent claims above and do not provide anything more than the mental processes that are practically capable of being performed in the human mind with the assistance of pen and paper and mathematical concepts that are achievable through mathematical computation. Therefore, claims 2-12 also recite abstract ideas that do not integrate into a practical application or amount to significantly more than the judicial exception, and are rejected under U.S.C. 101.
Step 1
Claim 2-12 are drawn to a computer-implemented method. Therefore, this claim group falls under one of four categories of statutory subject matter (process/method, machines/product/apparatus, manufactures, and composition of matter).
Step 2A – Prong 1
Dependent claim 2 recites further the mental process by sampling the task from a distribution that based on one or more features of the ML project (MPEP 2106.04(a)(2)(III)).
Dependent claim 3 recites further the mental process by sampling a batch of tasks from the distribution; determining a plurality of partial derivatives for the batch of tasks; and determining the second parameter or the third parameter depending on the plurality of partial derivatives those based on one or more features of the ML project (MPEP 2106.04(a)(2)(III)).
Dependent claim 4 recites further the mental process by determining the partial derivatives with respect to the first parameter, and determining a change of the second parameter depending on a function, the function being a sum of the partial derivatives; or determining the partial derivative with respect to the third parameter and determining a change of the third parameter depending on a function, the function being a sum of the partial derivatives those based on one or more features of the ML project (MPEP 2106.04(a)(2)(III)).
Dependent claim 8 recites further the mental process by determining for the task a plurality of parts of the solution of the initial value problem including the first part and the second part, and storing at least a part of the plurality of parts in memory derivatives those based on one or more features of the ML project (MPEP 2106.04(a)(2)(III)).
Dependent claim 11 recites further the mental process by determining in iterations different second parameter and/or third parameter for different batches of tasks sampled from the distribution; wherein the method further comprises changing the first parameter after at least one of the iterations according to the second parameter and/or third parameter of the at least one of the iterations those based on one or more features of the ML project (MPEP 2106.04(a)(2)(III)).
Dependent claim 12 recites further the mental process by assigning the second parameter to the first parameter in a first iteration, determining the rate for changing the first parameter in the first iteration depending on the third parameter, and changing the first parameter in the first iteration and/or a second iteration after the first iteration with the rate those based on one or more features of the ML project (MPEP 2106.04(a)(2)(III)).
Step 2A – Prong 2
Dependent claim 5 recites further the insignificant extra solution activities by wherein the determining of the second parameter includes randomly initializing the initial value in a first step of a plurality of steps of solving the initial value problem. This additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (MPEP 2106.05(g)).
Dependent claim 6 recites further the insignificant extra solution activities by wherein the third parameter is initialized in the first step of the plurality of steps as a positive scalar or a vector or matrix of positive scalars. This additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (MPEP 2106.05(g)).
Dependent claim 7 recites further the insignificant extra solution activities by wherein the determining of the solution of the initial value problem includes solving the initial value problem with an ordinary differential equation solver according to an explicit Runge-Kutta method other than the Euler method. This additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (MPEP 2106.05(g)).
Dependent claim 9 recites further the insignificant extra solution activities by wherein the determining of the at least one of the plurality of partial derivatives for the task depending on the plurality of parts of the solution of the initial value problem includes reading a first subset of the plurality of parts of the solution of the initial value problem from memory and determining a second subset of the plurality of parts of the solution of the initial value problem depending on at least one part of the solution of the initial value problem of the first subset. These additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea (MPEP 2106.05(g)).
Dependent claim 10 recites further the insignificant extra solution activities by wherein the rate is defined depending on an ordinary differential equation including a derivative of a temporal course of the first parameter with respect to time and a partial derivative of a temporal course of the measure with respect to the temporal course of the first parameter. This additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (MPEP 2106.05(g)).
As such, dependent claims 2-12 are not patent eligible.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-14 are rejected under 35 U.S.C. 103 as being unpatentable over Finn et al (“Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks”) hereafter Finn, in view of Xu et al (“Meta Learning in the Continuous Time Limit”) hereafter Xu, further in view of Xu et al (US 20190354859 A1) hereafter Xu Z, and further in view of Yamaguchi et al (US 20210327456 A1) hereafter Yamaguchi.
Finn was cited in the IDS filed on 08/19/2021.
Xu was cited in the IDS filed on 08/19/2021.
With respect to claim 1, Finn teaches a computer implemented method of performing machine learning for a model, the model mapping a dataset to a solution of a task depending on a first parameter (a deep neural network is introduced with a model-agnostic meta-learning (MAML) algorithm that includes classification, regression and policy gradient reinforcement learning (RL). The goal of the trained model is to learn a new task quickly from a small amount of data and to be able to learn a large number of tasks. The key here is to train the initial parameters of the model to obtain a maximal performance on a new task. The first parameter may be a model parameter θ [page 1, 1. Introduction and page 3, 2.2. A Model-Agnostic Meta-Learning Algorithm]), the method comprising the following steps:
determining a second parameter for assigning the second parameter to the first parameter in a first iteration of learning (when the model parameter θ is adapted to a new task Ti, θ becomes an updated parameter θi which is computed using one or more gradient descent updates on task Ti [page 3, 2.2. A Model-Agnostic Meta-Learning Algorithm]); and
determining a third parameter (when adapting to a new task, a model parameter θ may become parameter θ’, wherein the updated parameter is computed using one or more gradient descent updates on the task [par 3, 2.2. A Model-Agnostic Meta-Learning Algorithm]); and
determining a rate for changing the first parameter in at least one iteration of learning depending on the third parameter and depending on a measure for evaluating the solution of the task (a hyperparameter or a MAML parameter is included with the function along with the model parameter to act as a step size, for instance, α is represented to determine a rate of changing of the model parameter θ [par 3, 2.2. A Model-Agnostic Meta-Learning Algorithm]).
However, Finn does not disclose wherein the determining of the second parameter or third parameter includes determining a solution of an initial value problem that depends on a derivative of the measure with respect to the first parameter; wherein the determining of the solution of the initial value problem includes determining a first part of the solution of the initial value problem depending on an initial value, determining a second part of the solution of the initial value problem depending on the first part, determining for the first part a partial derivative, determining for the second part a partial derivative, and determining the second parameter or the third parameter depending on at least one of the partial derivatives, wherein the machine learning is applied to adapt an anomaly detection model to a given set of anomaly detection tasks of audio recordings from a set of similar microphones, wherein the anomaly detection model is an autoencoder, wherein the method of machine learning is used to adapt a learning rule used to train the autoencoder such that autoencoder learns to detect anomalies in recordings from the similar microphones.
In the same field of endeavor, Xu teaches wherein the determining of the second parameter or third parameter includes determining a solution of an initial value problem that depends on a derivative of the measure with respect to the first parameter (In the initialization of the parameter, the problem is transformed into an initial value problem (IVP) of an ordinary differential equation (ODP). An example of a derivative of a measure with respect to a first parameter may be represented by Lyapunov’s direct method, which based on constructing a positive definite Lyapunov function ε. An example of a solution of a IVP is y(t) with initial condition y(0) [page 2, 1. Introduction; page 3, 2. Preliminaries and page 17, Theorem B.2.]);
wherein the determining of the solution of the initial value problem includes
determining a first part of the solution of the initial value problem depending on an initial value (y(t) is the solution of an IVP with an initial condition y(0) for any t>=0, or y(t) equation is the first part of the solution of the IVP under the form y’ which is a simple first-order separable ODE [page 17, Theorem B.2.]),
determining a second part of the solution of the initial value problem depending on the first part (then y(t) converges to γ/ ζ whenever ζ > 0. The sufficient conditions may be derived to get the final result y(t) of IVP [page 17, Theorem B.2.]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated the concept of initial value problem of the ordinary differential equation as suggested by Xu which combined into the concept of determining model parameter, updated parameter and return parameter in MAML algorithm as suggested by Finn because all of these systems addressing the process of applying Reinforcement Learning system in meta-learning to improve the deep neural network training. Doing so would be desirable because the system of Finn should combine the ODE into the MAML to establish a linear convergence rate to the global minimum of the MAML loss function for strongly convex task losses (Xu, [page 11, 5. Conclusions]).
However, the combination of Finn and Xu does not particularly disclose determining for the first part a partial derivative, determining for the second part a partial derivative, and determining the second parameter or the third parameter depending on at least one of the partial derivatives, and wherein the machine learning is applied to adapt an anomaly detection model to a given set of anomaly detection tasks of audio recordings from a set of similar microphones, wherein the anomaly detection model is an autoencoder, wherein the method of machine learning is used to adapt a learning rule used to train the autoencoder such that autoencoder learns to detect anomalies in recordings from the similar microphones.
In the same field of endeavor, Xu Z teaches determining for the first part a partial derivative, determining for the second part a partial derivative (the partial derivative may be used in the differentiated meta-objective function for updating the policy parameters [par. 0031-0033]), and
determining the second parameter or the third parameter depending on at least one of the partial derivatives (by using at least one of the partial derivatives, the updated parameter or the return parameter may be determined after updating the function by adding a differential [par. 0031-0033, 0036-0039]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated the concept of using partial derivatives in differential equations as suggested by Xu Z into the concept of initial value problem of the ordinary differential equation as suggested by Xu which combined into the concept of determining model parameter, updated parameter and return parameter in MAML algorithm as suggested by Finn because all of these systems addressing the process of applying Reinforcement Learning system in meta-learning to improve the deep neural network training. Doing so would be desirable because the system of Finn should combine the ODE into the MAML to establish a linear convergence rate to the global minimum of the MAML loss function for strongly convex task losses (Xu, [page 11, 5. Conclusions]), wherein the system of Xu should make use of partial derivatives in meta-learning of MAML to improve the return parameters and to therefore improve the effectiveness of the training (Xu Z, [par. 0015]).
However, the combination of Finn, Xu and Xu Z does not disclose wherein the machine learning is applied to adapt an anomaly detection model to a given set of anomaly detection tasks of audio recordings from a set of similar microphones, wherein the anomaly detection model is an autoencoder, wherein the method of machine learning is used to adapt a learning rule used to train the autoencoder such that autoencoder learns to detect anomalies in recordings from the similar microphones.
In the same field of endeavor, Yamaguchi teaches wherein the machine learning is applied to adapt an anomaly detection model to a given set of anomaly detection tasks of audio recordings from a set of similar microphones, wherein the anomaly detection model is an autoencoder, wherein the method of machine learning is used to adapt a learning rule used to train the autoencoder such that autoencoder learns to detect anomalies in recordings from the similar microphones (anomaly sound is detected in an equipment by a microphone, which is utilized by an anomalous sound detection technique. The parameter updating unit 440 updates the parameter of the first autoencoder to optimize the value of the loss function, such that it uses a gradient descent method for updating the parameter. The autoencoder adaptive learning apparatus includes a recording unit, wherein the recording unit is a component which records information necessary for processing of the autoencoder as appropriate. The recording unit records information necessary for processing of the anomaly detection apparatus, and the second autoencoder learned using the autoencoder adaptive learning apparatus [par. 0002, 0003, 0169-0177, 0183 and FIG. 4]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated the concept of estimating an anomaly degree indicating a degree of anomaly of anomaly detection target equipment from sound emitted from the equipment as suggested by Yamaguchi into the combination of Finn, Xu and Xu Z because all of these systems addressing the process of applying a neural network technique for adjusting parameters in performing a machine learning model. Doing so would be desirable because the combination of Finn, Xu and Xu Z would be more efficient by using an anomaly detection technique to estimate an anomaly degree of an equipment based on one or more autoencoders which restore normal sound emitted from the equipment (Yamaguchi, [par. 0022-0026]).
With respect to claim 2, the combination of Finn, Xu, Xu Z and Yamaguchi teaches sampling the task from a distribution (Finn, a distribution over the tasks is used in the meta-learning scenario, for instance, a task Ti is sampled from p(T), the model is trained with K samples and feedback from the corresponding loss L(Ti) from Ti [page 2, 2.1. Meta-Learning Problem Set-Up]).
With respect to claim 3, the combination of Finn, Xu, Xu Z and Yamaguchi teaches sampling a batch of tasks from the distribution (Finn, a distribution over the tasks is used in the meta-learning scenario, for instance, a task Ti is sampled from p(T), the model is trained with K samples and feedback from the corresponding loss L(Ti) from Ti. At the end of meta-training, new tasks are sampled from p(T) [page 2, 2.1. Meta-Learning Problem Set-Up]);
determining a plurality of partial derivatives for the batch of tasks (Xu Z, a plurality of tasks may be performed by an agent in an environment in order to retrieve a plurality of experiences from a reinforcement learning neural network, and the use of partial derivatives may be essential in determining the results of a plurality of tasks [par. 0009, 0015, 0032-0039]); and
determining the second parameter or the third parameter depending on the plurality of partial derivatives (Xu Z, by combining the use of partial derivatives in differential equation of the meta-objective function, the return policy parameters are determined and updated [par. 0015, 0032-0039]).
With respect to claim 4, the combination of Finn, Xu, Xu Z and Yamaguchi teaches determining the partial derivatives with respect to the first parameter (Xu Z, the initial parameters are the policy parameters those may be determined to be updated using partial derivatives [par. 0009, 0015]), and determining a change of the second parameter depending on a function, the function being a sum of the partial derivatives (Xu Z, ƒ(τ, θ, η) is used as an update function to return the updated policy parameters, wherein the function is the sum of the current partial derivatives applied on the respective policy parameters [par. 0015, 0032-0039]); or
determining the partial derivative with respect to the third parameter and determining a change of the third parameter depending on a function, the function being a sum of the partial derivatives (Xu Z, α is a learning rate that is applied to the function when updating the policy parameters, and α is multiplied by a partial derivative of the respective policy parameter. The function is the sum of the current partial derivatives applied on the respective policy parameters [par. 0032-0040]).
With respect to claim 5, the combination of Finn, Xu, Xu Z and Yamaguchi teaches the determining of the second parameter includes randomly initializing the initial value in a first step of a plurality of steps of solving the initial value problem (Xu, the problem that has been transformed into the IVP of an ODE that has one or more parameter values initialized in the first step of solving the IVP [page 2, 1. Introduction]).
With respect to claim 6, the combination of Finn, Xu, Xu Z and Yamaguchi teaches the third parameter is initialized in the first step of the plurality of steps as a positive scalar or a vector or matrix of positive scalars (Xu Z, the α as learning rate or the c as a coefficient may act as the hyperparameter and may be initialized at the first stage of updating the policy parameters [par. 0040-0043]).
With respect to claim 8, the combination of Finn, Xu, Xu Z and Yamaguchi teaches determining for the task a plurality of parts of the solution of the initial value problem including the first part and the second part (Xu, a solution y(t) is provided for an IVP that includes the first part of computing y(t) using initial conditions of y(0) and the second part of deriving sufficient conditions that converges to γ/ ζ [page 17, Theorem B.2.]), and storing at least a part of the plurality of parts in memory (Xu Z, the initial parameters and the return parameters after the calculation by using partial derivatives may be stored in the local memory for the next use [par. 0093, 0094, 0101]).
With respect to claim 9, the combination of Finn, Xu, Xu Z and Yamaguchi teaches the determining of the at least one of the plurality of partial derivatives for the task depending on the plurality of parts of the solution of the initial value problem includes reading a first subset of the plurality of parts of the solution of the initial value problem from memory (Xu Z, the return function may adjust the return parameters stored in the local memory after reading the values of the return parameters [par. 0094, 0101]) and determining a second subset of the plurality of parts of the solution of the initial value problem depending on at least one part of the solution of the initial value problem of the first subset (Xu Z, the updated return parameters may be accessed by the policy training module to iteratively update the policy and the return function [par. 0094, 0101]).
With respect to claim 10, the combination of Finn, Xu, Xu Z and Yamaguchi teaches the rate is defined depending on an ordinary differential equation including a derivative of a temporal course of the first parameter with respect to time (Xu, Lyapunov’s direct method is commonly used for analyzing the convergence of ODE over time, which is an idea of measuring the “energy” in a system [page 3, 2. Preliminaries]) and a partial derivative of a temporal course of the measure with respect to the temporal course of the first parameter (Xu Z, β is the learning rate for updating the meta-parameters, where the meta-parameter is used as a reference value in calculation of the differential equation using partial derivative, and the method of meta-gradient starts with the initial parameters [par. 0110-0117]).
With respect to claim 11, the combination of Finn, Xu, Xu Z and Yamaguchi teaches determining in iterations different second parameter or third parameter for different batches of tasks sampled from the distribution (Finn, MAML is evaluated based on a several sets of tasks based off of simulated continuous control environments. For meta-learning updates, the standard linear feature baseline is used to fit separately at each iteration for each sampled task in the batch. One of the baseline models is training a policy from randomly initialized weights [page 7 & 8, 5.3. Reinforcement Learning]);
wherein the method further comprises changing the first parameter after at least one of the iterations according to the second parameter or third parameter of the at least one of the iterations (Finn, the model of meta-learning should be able to adapt to a distribution of tasks. A task is trained with K samples using the model, and then the model is tested on the new samples. The model is then improved by considering how the test error on new data changes with respect to the parameters [page 2, 2.1. Meta-Learning Problem Set-Up]).
With respect to claim 12, the combination of Finn, Xu, Xu Z and Yamaguchi teaches assigning the second parameter to the first parameter in a first iteration (Finn, when the model parameter θ is adapted to a new task Ti, θ becomes an updated parameter θi which is computed using one or more gradient descent updates on task Ti [page 3, 2.2. A Model-Agnostic Meta-Learning Algorithm]), determining the rate for changing the first parameter in the first iteration depending on the third parameter (Finn, a hyperparameter or a MAML parameter is included with the function along with the model parameter to act as a step size, for instance, α is represented to determine a rate of changing of the model parameter θ [par 3, 2.2. A Model-Agnostic Meta-Learning Algorithm]), and changing the first parameter in the first iteration or a second iteration after the first iteration with the rate (Finn, the meta-optimization across tasks is performed using stochastic gradient descent, such that the model parameter θ is updated continuously with β is the meta step size [page 3, 2.2. A Model-Agnostic Meta-Learning Algorithm]).
With respect to claim 13, it is a device for machine learning model claim that corresponding to the computer-implemented method of claim 1. Therefore, it is rejected for the same reason as claimed in claim 1 above.
With respect to claim 14, it is a non-transitory computer-readable claim that corresponding to the computer-implemented method of claim 1. Therefore, it is rejected for the same reason as claimed in claim 1 above.
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Finn et al (“Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks”) hereafter Finn, in view of Xu et al (“Meta Learning in the Continuous Time Limit”) hereafter Xu, further in view of Xu et al (US 20190354859 A1) hereafter Xu Z, and further in view of Yamaguchi et al (US 20210327456 A1) hereafter Yamaguchi, as claimed in claim 1 above, and further in view of Im et al (“Model-Agnostic Meta-Learning using Runge-Kutta Methods”) hereafter Im.
Finn was cited in the IDS filed on 08/19/2021.
Xu was cited in the IDS filed on 08/19/2021.
Im was cited in the IDS filed on 08/19/2021.
With respect to claim 7, the combination of Finn, Xu, Xu Z and Yamaguchi teaches all the limitations as claimed in claim 1 above.
However, the combination of Finn, Xu, Xu Z and Yamaguchi does not teach the determining of the solution of the initial value problem includes solving the initial value problem with an ordinary differential equation solver according to an explicit Runge-Kutta method other than the Euler method.
In the same field of endeavor, Im teaches the determining of the solution of the initial value problem includes solving the initial value problem with an ordinary differential equation solver according to an explicit Runge-Kutta method other than the Euler method (Runge-Kutta method is applied into the MAML optimization that can take advantage of computing the gradients multiple-steps ahead of updating the meta-model, this also generalize MAML to using higher-order gradients [page 2, 1. Introduction]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated the concept of using Runge-Kutta method in the MAML as suggested by Im into the combination of Finn, Xu, Xu Z and Yamaguchi because all of these systems addressing the process of applying Reinforcement Learning system in MAML to improve the deep neural network training. Doing so would be desirable because the combination of Finn, Xu, Xu Z and Yamaguchi should combine with the Runge-Kutta method to control various important aspects of the meta-learning process to improve the performance on regression, classification and reinforcement learning tasks (Im, [page 9, 6. Discussion and Future Work]).
Response to Arguments
The examiner respectfully acknowledges the applicant’s amendments to claims 1, 13 and 14.
Applicant’s arguments filed on 10/29/2025 regarding the rejections to claims 1-14 under 35 U.S.C. 101 have been fully considered but are not persuasive.
Applicant argued that “Step 2A, Prong One
Applicant submits that the machine learning audio anomaly detection amendment made to the independent claims does not recite a judicial exception, in alignment with the conclusion reached by Desjardins with respect to the machine learning model invention it analyzed. In Desjardins, …
Thus, the logic of Desjardins when applied here leads to the conclusion that the limitation “wherein the machine learning … from the similar microphones” does not recite a mental and/or mathematical judicial exception. Therefore, it qualifies for analysis under Prong Two.”
Examiner respectfully disagrees.
Based on what is recited in claim 1, the broadest reasonable interpretation (BRI) in view of Specification of the claim limitations “wherein the machine learning is applied to adapt an anomaly detection model to a given set of anomaly detection tasks of audio recordings from a set of similar microphones, wherein the anomaly detection model is an autoencoder, wherein the method of machine learning is used to adapt a learning rule used to train the autoencoder such that autoencoder learns to detect anomalies in recordings from the similar microphones” encompass a mentally performable process, and these limitations recite a specific type of neural network which is “autoencoder” that involves a mathematical algorithm. The claim as a whole is still directed to an abstract idea that includes mathematical algorithm and mental process for data analysis.
Applicant argued that “Step 2A, Prong Two
As Desjardins acknowledged, under the analysis called for in Prong Two, …
Similarly, the specification here establishes a link between the claim amendment made here and the technological improvement in the operation of an AI-related machine learning model in the area of audio anomaly detection. [0072] of the specification provides describes the audio recording anomaly detection that is recited in the claim, as well as its benefits:
[0072] …
This improvement in greater sample efficiency and faster audio recording anomaly
detection is linked to the machine learning method recited in the present claims, particularly the claim amendment "wherein the machine learning is applied to adapt an anomaly detection model to a given set of anomaly detection tasks of audio recordings from a set of similar microphones, wherein the anomaly detection model is an autoencoder, wherein the method of machine learning is used to adapt a learning rule used to train the autoencoder such that autoencoder learns to detect anomalies in recordings from the similar microphones." Since this represents a concrete improvement in the area of audio recording anomaly detection, the claimed invention can be properly characterized as an improved way of performing anomaly detection through machine learning.”
Examiner respectfully disagrees.
The claim recites at a high level of generality, wherein (1) generally linking the use of judicial exception to a field of use or a particular technological environment, e.g. applied to adapt an anomaly detection model to a given set of anomaly detection tasks of audio recordings from a set of similar microphones, also wherein the anomaly detection model is an autoencoder (see MPEP 2106.05(h)), and (2) reciting outcome of the method of machine learning being used to adapt a learning rule used to train the autoencoder such that autoencoder learns to detect anomalies in recordings from the similar microphones but failing to provide details as to how a learning rule is used to train the autoencoder to learn to detect anomalies in recordings from the similar microphones, e.g. such as an architecture of the autoencoder, how does the recording from similar microphones get abstract by decoder of autoencoder/encoded such that anomalies can be detected.
Merely performing a known abstract idea faster or more efficiently on a generic computer is not sufficient to integrate the abstract idea into a patent-eligible application. The claim and specification do not disclose a specific technical improvement to the functioning of the computer itself, or an unconventional improvement to the machine learning model architecture or training method that provides a technological solution to a technical problem beyond a generic "apply it" instruction. The use of an autoencoder for anomaly detection, while applied to a specific data type (microphone audio recordings), uses known machine learning models and techniques in a conventional manner. The specification's assertion of increased speed or efficiency is an expected result of implementing an algorithm on a computer, rather than a novel technical solution that is "significantly more" than the abstract idea itself. Thus, these additional elements merely represent equivalent of stating "apply it" to the judicial exception and cannot integrate the judicial exception into a practical application (see MPEP 2106.05(f)).
Applicant argued that “Step 2B
Even if the Patent Office disagrees that the claims are eligible under Prong Two of Step 2A, Applicant submits that the Patent Office has improperly applied Step 2B to all the limitations that qualify as additional elements …
This is improper (and would be improper if applied to the limitations added by the claim amendment made here) because the analysis fails to comply with any of requirements (A)-(D) above. Since the Patent Office takes the same improper approach for every additional limitation identified in the rejection of the dependent claims, Applicant traverses the Step 2B analysis for each of these claims for the same reasons given above against the Step 2B analysis of the pending claims.”
Examiner respectfully disagrees.
As the amended limitation is considered ineligibility under step 2A-Prong Two, that would combine with the additional elements in step 2B to conclude that the amended claim as a whole do not provide an inventive concept. The steps described are all conventional, routine, and well-understood activities in the field of machine learning and computer programming (determining parameters, using derivatives, solving initial value problems). Applying these known techniques to a new data set or field of use does not render the claim non-conventional or provide the necessary "significantly more" to transform the abstract idea into patent-eligible subject matter.
As set forth in MPEP 2106.05(a)(II), to show that the involvement of a computer assists in improving the technology, the claims must recite the details regarding how a computer aids the method, the extent to which the computer aids the method, or the significance of a computer to the performance of the method. The Courts have provided several examples those may not be sufficient to show an improvement to technology: ii. Using well-known standard laboratory techniques to detect enzyme levels in a bodily sample such as blood or plasma, Cleveland Clinic Foundation v. True Health Diagnostics, LLC, 859 F.3d 1352, 1355, 1362, 123 USPQ2d 1081, 1082-83, 1088 (Fed. Cir. 2017); and iii. Gathering and analyzing information using conventional techniques and displaying the result, TLI Communications, 823 F.3d at 612-13, 118 USPQ2d at 1747-48. If the Specification explicitly sets forth an improvement but in a conclusory manner (i.e., a bare assertion of an improvement without the detail necessary to be apparent to a person of ordinary skill in the art), the claim as a whole may not be proved as an improved technology.
Therefore, the amended claim 1 and its corresponding claims 13 and 14 remain ineligible under 35 USC 101 for at least the reasons above. Dependent claims 2-12, those depended on independent claim 1, are not patent-eligible for the same reasons.
Applicant’s arguments filed on 10/29/2025 regarding the rejections to claims 1-14 under 35 U.S.C. 103 have been fully considered and moot in view of new ground of rejection (see rejection above).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Pardeshi et al (US 20210334645 A1) disclosed apparatuses, systems, and techniques to determine actions to be taken for data anomalies. In at least one embodiment, audio and video data captured for an environment of a user can be analyzed to detect one or more data anomalies and determine whether to notify this user depending on whether the anomalies are applicable to this user.
Kain et al (US 20200342329 A1) disclosed methods, systems, and computer-readable storage media for defining an autoencoder architecture including a neural network, during training of the autoencoder, recording a loss value at each iteration to provide a plurality of loss values, the autoencoder being trained using a data set that is associated with a domain, and a learning rate to provide a trained autoencoder, calculating a penalty score using at least a portion of the plurality of loss values, the penalty score being based on a loss span penalty P.sub.LS, a convergence penalty P.sub.C, and a fluctuation penalty P.sub.F, comparing the penalty score P to a threshold penalty score to affect a comparison, and selectively employing the trained autoencoder for anomaly detection within the domain based on the comparison.
Vanga et al (US 20200334680 A1) disclosed techniques relating to detecting anomalous transactions using machine learning. For example, in various embodiments, an anomaly detection computer system may access an input dataset that includes data indicative of transactions submitted to a transaction network for both a first entity and by a plurality of other entities.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Quoc Phung whose telephone number is (703) 756 1330. The examiner can normally be reached on Monday through Friday from 9am to 5pm PT.
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) athttp://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Jennifer Welch can be reached on 571-272-7212. 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.
/Q.L.P./Examiner, Art Unit 2143
/JENNIFER N WELCH/Supervisory Patent Examiner, Art Unit 2143