Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 101
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Regarding claims 1-15:
Step 1: With respect to claims 1-12 and 15, the preamble of claims 1-12 and 15 recite a computer-implemented method, which falls within the statutory category of a process. With respect to claim 13, the preamble of claim 13 recites an apparatus, which falls within the statutory category of an apparatus. With respect to claim 14, the preamble of claim 14 recites a non-transitory computer readable medium, which falls within the statutory category of a manufacture.
Regarding claim 1,
Step 2A — Prong One: Claim 1 recites an abstract idea. The limitation of “classifying, during the at least one discrimination phase when training the generative machine learning model, the input to the discriminator model as either a data series from the data space or a synthetic data series, using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;” is an abstract idea directed to mental processes, for example, a human could use evaluation and judgement to classify, during at least one discrimination phase when training the generative machine learning model, the input to the discriminator as either a data series from the data space or a synthetic data series, using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series. A human could evaluate the input to the discriminator, during a phase in which the input will be discriminated (or classified/evaluated by the human), write the data using a physical aid, determine the respective inputs to the first and second functions, and, using judgement, classify the observed data as either a synthetic data series or as data from the data space.
Step 2A — Prong Two: Claim 1 does not integrate the judicial exception into practical application. The additional element of claim 1, “obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot; generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution; inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series;” is insignificant extra-solution activity (See MPEP 2106.05(g)) that amounts to no more than mere data gathering. The additional element of “and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model” is an attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Step 2B: Claim 1 does not contain any additional elements that would amount to significantly more than the judicial exception. The additional element of claim 1, “obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot; generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution; inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series;” is insignificant extra-solution activity (See MPEP 2106.05(g)) that amounts to no more than mere data gathering. Data gathering is a well-understood, routine conventional activity as recognized by the courts (See MPEP 2106.05(d)(II)). Simply gathering a plurality of data series from a data space, generating synthetic data by sampling from a distribution, and inputting either data into a discriminator does not impose any meaningful limitations or improvements to a generative adversarial network that models operation profiles of a vehicle. The additional element of “and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model” is an attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Regarding claim 12,
Step 2A — Prong One: Claim 12 recites an abstract idea. The limitation of “classifying, during the at least one discrimination phase when training the generative machine learning model, the input to the discriminator model as either a data series from the data space or a synthetic data series, using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;” is an abstract idea directed to mental processes, for example, a human could use evaluation and judgement to classify, during at least one discrimination phase when training the generative machine learning model, the input to the discriminator as either a data series from the data space or a synthetic data series, using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series. A human could evaluate the input to the discriminator, during a phase in which the input will be discriminated (or classified/evaluated by the human), write the data using a physical aid, determine the respective inputs to the first and second functions, and, using judgement, classify the observed data as either a synthetic data series or as data from the data space.
Step 2A — Prong Two: Claim 12 does not integrate the judicial exception into practical application. The additional element of claim 12, “obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot; generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution; inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series;” is insignificant extra-solution activity (See MPEP 2106.05(g)) that amounts to no more than mere data gathering. The additional element of “obtaining a plurality of samples by sampling from a distribution configured to provide pseudo random vectors; decoding the plurality of the samples of the distribution to obtain a further plurality of synthetic data samples in the data space; and outputting the further plurality of synthetic data samples representing synthetic operation profiles of a vehicle or robot” is also insignificant extra-solution activity (See MPEP 2106.05(g)) that amounts to no more than mere data gathering. The additional element of “and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model” is an attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Step 2B: Claim 12 does not contain any additional elements that would amount to significantly more than the judicial exception. The additional element of claim 12, “obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot; generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution; inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series;” is insignificant extra-solution activity (See MPEP 2106.05(g)) that amounts to no more than mere data gathering. The additional element of “obtaining a plurality of samples by sampling from a distribution configured to provide pseudo random vectors; decoding the plurality of the samples of the distribution to obtain a further plurality of synthetic data samples in the data space; and outputting the further plurality of synthetic data samples representing synthetic operation profiles of a vehicle or robot” is also insignificant extra-solution activity (See MPEP 2106.05(g)) that amounts to no more than mere data gathering. Data gathering is a well-understood, routine conventional activity as recognized by the courts (See MPEP 2106.05(d)(II)). Simply gathering a plurality of data series from a data space, generating synthetic data by sampling from a distribution, and inputting either data into a discriminator does not impose any meaningful limitations or improvements to a generative adversarial network that models operation profiles of a vehicle. Similarly, obtaining a plurality of samples by sampling from a distribution configured to provide pseudo-random vectors, decoding the plurality of samples of the distribution to obtain a further plurality of synthetic data samples, and outputting these synthetic data samples imposes no meaningful limits on modeling the operation profiles of a vehicle through a generative adversarial network. The additional element of “and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model” is an attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Regarding claim 13,
Step 2A — Prong One: Claim 13 recites an abstract idea. The limitation of “during the at least one discrimination phase when training the generative machine learning model, to classify the input to the discriminator model as either a data series from the data space, or a synthetic data series using at least a first and a second function of the discriminator model;” is an abstract idea directed to mental processes, for example, a human could use evaluation and judgement to classify the input to the discriminator model as either a data series from the data space, a synthetic data series using at least a first and a second function of the discriminator model, during at least one discrimination phase when training the generative machine learning model. A human could evaluate the input to the discriminator, during a phase in which the input will be discriminated (or classified/evaluated by the human), write the data using a physical aid, determine the respective inputs to the first and second functions, and, using judgement, classify the observed data as either a synthetic data series or as data from the data space.
Step 2A — Prong Two: Claim 13 does not integrate the judicial exception into practical application. The additional elements of “an input interface configured to obtain a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot; a processor configured to, during at least one generation phase when training the generative machine learning model, generate at least one synthetic data series by sampling from a distribution; wherein the processor is configured to, during at least one discrimination phase when training the generative machine learning model, input into the discriminator model either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the at least one synthetic data series;”, “wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;”, and “and wherein the apparatus further comprises: an output interface configured to output the trained machine learning model” are all insignificant extra-solution activity (See MPEP 2106.05(g)) that amounts to no more than mere data gathering. The additional elements of “the processor is configured” and “wherein the processor is configured to iteratively train the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model;” are mere attempts to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Step 2B: Claim 13 does not contain any additional elements that would amount to significantly more than the judicial exception. The additional elements of “an input interface configured to obtain a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot; a processor configured to, during at least one generation phase when training the generative machine learning model, generate at least one synthetic data series by sampling from a distribution; wherein the processor is configured to, during at least one discrimination phase when training the generative machine learning model, input into the discriminator model either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the at least one synthetic data series;”, “wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;”, and “and wherein the apparatus further comprises: an output interface configured to output the trained machine learning model” are all insignificant extra-solution activity (See MPEP 2106.05(g)) that amounts to no more than mere data gathering. Data gathering is a well-understood, routine conventional activity as recognized by the courts (See MPEP 2106.05(d)(II)). The additional elements of “the processor is configured” and “wherein the processor is configured to iteratively train the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model;” are mere attempts to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Regarding claim 14,
Step 2A — Prong One: Claim 14 recites an abstract idea. The limitation of “classifying, during the at least one discrimination phase when training the generative machine learning model, the input to the discriminator model as either a data series from the data space or a synthetic data series, using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;” is an abstract idea directed to mental processes, for example, a human could use evaluation and judgement to classify, during at least one discrimination phase when training the generative machine learning model, the input to the discriminator as either a data series from the data space or a synthetic data series, using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series. A human could evaluate the input to the discriminator, during a phase in which the input will be discriminated (or classified/evaluated by the human), write the data using a physical aid, determine the respective inputs to the first and second functions, and, using judgement, classify the observed data as either a synthetic data series or as data from the data space.
Step 2A — Prong Two: Claim 14 does not integrate the judicial exception into practical application. The additional element of claim 14, “obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot; generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution; inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series;” is insignificant extra-solution activity (See MPEP 2106.05(g)) that amounts to no more than mere data gathering. The additional element of “and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model” is an attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Step 2B: Claim 14 does not contain any additional elements that would amount to significantly more than the judicial exception. The additional element of claim 14, “obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot; generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution; inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series;” is insignificant extra-solution activity (See MPEP 2106.05(g)) that amounts to no more than mere data gathering. Data gathering is a well-understood, routine conventional activity as recognized by the courts (See MPEP 2106.05(d)(II)). Simply gathering a plurality of data series from a data space, generating synthetic data by sampling from a distribution, and inputting either data into a discriminator does not impose any meaningful limitations or improvements to a generative adversarial network that models operation profiles of a vehicle. The additional element of “and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model” is an attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Regarding claim 15,
Step 2A — Prong One: Claim 15 recites an abstract idea. The limitation of “classifying, during the at least one discrimination phase when training the generative machine learning model, the input to the discriminator model as either a data series from the data space or a synthetic data series, using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;” is an abstract idea directed to mental processes, for example, a human could use evaluation and judgement to classify, during at least one discrimination phase when training the generative machine learning model, the input to the discriminator as either a data series from the data space or a synthetic data series, using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series. A human could evaluate the input to the discriminator, during a phase in which the input will be discriminated (or classified/evaluated by the human), write the data using a physical aid, determine the respective inputs to the first and second functions, and, using judgement, classify the observed data as either a synthetic data series or as data from the data space.
Step 2A — Prong Two: Claim 15 does not integrate the judicial exception into practical application. The additional elements of “obtaining a further plurality of synthetic data series representing synthetic operation profiles of a vehicle as generated by: configuring a generative machine learning model according to a plurality of model parameters obtained by: obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot; generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution; inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the » > plurality of synthetic data series;”, “obtaining a plurality of samples by sampling from a distribution configured to provide pseudo random vectors; decoding the plurality of the samples of the distribution to obtain a further plurality of synthetic data samples in the data space; and outputting the further plurality of synthetic data samples representing synthetic operation profiles of a vehicle or robot; inputting the further plurality of synthetic data series representing synthetic operation profiles of a vehicle into a further machine learning model;”, and “and outputting a further plurality of model parameters of the further machine learning model for modelling operation profiles of a vehicle” are all insignificant extra-solution activities that amount to no more than mere data gathering. The additional elements of “and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model;” and “iteratively training the further machine learning model” are mere attempts to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Step 2B: Claim 15 does not contain any additional elements that would amount to significantly more than the judicial exception. The additional elements of “obtaining a further plurality of synthetic data series representing synthetic operation profiles of a vehicle as generated by: configuring a generative machine learning model according to a plurality of model parameters obtained by: obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot; generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution; inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series;”, “obtaining a plurality of samples by sampling from a distribution configured to provide pseudo random vectors; decoding the plurality of the samples of the distribution to obtain a further plurality of synthetic data samples in the data space; and outputting the further plurality of synthetic data samples representing synthetic operation profiles of a vehicle or robot; inputting the further plurality of synthetic data series representing synthetic operation profiles of a vehicle into a further machine learning model;”, and “and outputting a further plurality of model parameters of the further machine learning model for modelling operation profiles of a vehicle” are all insignificant extra-solution activities that amount to no more than mere data gathering. Data gathering is a well-understood, routine conventional activity as recognized by the courts (See MPEP 2106.05(d)(II)). The additional elements of “and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model,” and “iteratively training the further machine learning model” are mere attempts to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Regarding claim 2,
Step 2A — Prong Two: Claim 2 fails to integrate the judicial exception into practical application. The element of claim 2, “wherein an input of the first function is a time indexed function of the input to the discriminator model, and an input of the second function is a differentiable function of the input to the discriminator model” is insignificant extra-solution activity that amounts to no more than mere data gathering (See MPEP 2106.05(g)).
Step 2B: Claim 2 does not contain any additional elements that would amount to significantly more than the judicial exception. The element of claim 2, “wherein an input of the first function is a time indexed function of the input to the discriminator model, and an input of the second function is a differentiable function of the input to the discriminator model” is insignificant extra-solution activity that amounts to no more than mere data gathering (See MPEP 2106.05(g)). Data gathering is a well-understood, routine conventional activity as recognized by the courts (See MPEP 2106.05(d)(II)). The input to the first function being a time indexed function and the input to the second function being differentiable does not impose any meaningful limits on modeling the operation profiles of a vehicle with a generative adversarial network.
Regarding claim 3,
Step 2A — Prong One: Claim 3 recites an abstract idea. The limitation of “combining outputs of the first and second functions using a discriminator neural network” is an abstract idea directed to mental processes, for example, a human could use evaluation and judgement to combine the outputs of the first and second functions using a discriminator neural network.
Step 2A — Prong Two: Claim 3 fails to integrate the judicial exception into practical application. The additional element of claim 3, “wherein: the first function is a recurrent neural network, optionally a Long Short Term Memory Network, and/or the second function is obtained by projecting at least one of the synthetic data series and obtaining a result of the second function of the discriminator model based on a combination of the projected at least one synthetic data series and at least one data series from the plurality of data series” is a mere attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Step 2B: Claim 3 does not contain any additional elements that would amount to significantly more than the judicial exception. The additional element of claim 3, “wherein: the first function is a recurrent neural network, optionally a Long Short Term Memory Network, and/or the second function is obtained by projecting at least one of the synthetic data series and obtaining a result of the second function of the discriminator model based on a combination of the projected at least one synthetic data series and at least one data series from the plurality of data series” is a mere attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Regarding claim 4,
Step 2A — Prong One: Claim 4 recites an abstract idea. The limitation of “calculates a linear combination of the first and second functions of the discriminator model” is an abstract idea directed to mental processes, for example, a human could use evaluation to calculate a linear combination of the first and second functions of the discriminator model.
Step 2A — Prong Two: Claim 4 fails to integrate the judicial exception into practical application. The additional element of claim 4, “wherein the discriminator neural network” is a mere attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Step 2B: Claim 4 does not contain any additional elements that would amount to significantly more than the judicial exception. The additional element of claim 4, “wherein the discriminator neural network” is a mere attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Regarding claim 5,
Step 2A — Prong Two: Claim 5 fails to integrate the judicial exception into practical application. The limitation of “wherein each data series of the plurality of data series is either a time series or a distance series, and wherein each synthetic data series of the plurality of synthetic data series is either a synthetic time series or a synthetic distance data series” is insignificant extra-solution activity that amounts to no more than mere data gathering (See MPEP 2106.05(g)).
Step 2B: Claim 5 does not contain any additional elements that would amount to significantly more than the judicial exception. The limitation of “wherein each data series of the plurality of data series is either a time series or a distance series, and wherein each synthetic data series of the plurality of synthetic data series is either a synthetic time series or a synthetic distance data series” is insignificant extra-solution activity that amounts to no more than mere data gathering (See MPEP 2106.05(g)). Data gathering is a well-understood, routine conventional activity as recognized by the courts (See MPEP 2106.05(d)(II)).
Regarding claim 6,
Step 2A — Prong Two: Claim 6 fails to integrate the judicial exception into practical application. The limitation of “wherein, in the at least one generation phase, a latent space is sampled to provide the plurality of synthetic data series” is insignificant extra-solution activity that amounts to no more than mere data gathering (See MPEP 2106.05(g)).
Step 2B: Claim 6 does not contain any additional elements that would amount to significantly more than the judicial exception. The limitation of “wherein, in the at least one generation phase, a latent space is sampled to provide the plurality of synthetic data series” is insignificant extra-solution activity that amounts to no more than mere data gathering (See MPEP 2106.05(g)). Data gathering is a well-understood, routine conventional activity as recognized by the courts (See MPEP 2106.05(d)(II)).
Regarding claim 7,
Step 2A — Prong One: Claim 7 recites an abstract idea. The limitation of “stopping the iterative training of the generator model using at least one stopping criterion, wherein the at least one stopping criterion is obtained by evaluating at least one synthetic data series generated by an iteration of a generator model using the second function of the discriminator model, wherein the second function outputs the stopping criterion” is an abstract idea directed to mental processes, for example, a human could use observation, evaluation, and judgement to stop the iterative training of the generator model using at least one stopping criterion, wherein the stopping criterion is obtained by evaluating at least one synthetic data series generated by an iteration of a generator model using the second function of the discriminator model, wherein the second function outputs the stopping criterion.
Step 2A — Prong Two: Claim 7 does not contain any additional elements that would integrate judicial exception into practical application.
Step 2B: Claim 7 does not contain any additional elements that would amount to significantly more than the judicial exception.
Regarding claim 8,
Step 2A — Prong One: Claim 8 recites an abstract idea. The limitation of “(i) the second function of the discriminator model is evaluated on a series of values sampled from a distribution resulting from real or synthetic data series based on at least one data series representing vehicle velocity, and at least one synthetic data series representing synthetic vehicle velocity;” is an abstract idea directed to mental processes, for example, a human could use evaluation to evaluate the second function on a series of values sampled from a distribution resulting from real or synthetic data series based on at least one data series representing vehicle velocity, and at least one synthetic data series representing synthetic vehicle velocity. The limitation of “or (ii) the second function of the discriminator model is configured to evaluate a time spent at zero velocity based on at least one synthetic data series representing vehicle velocity;” is an abstract idea directed to mental processes, for example, a human could use evaluation to evaluate a time spent at zero velocity based on at least one synthetic data series representing vehicle velocity.
Step 2A — Prong Two: Claim 8 fails to integrate the abstract idea into practical application. The elements of “or (iii) the plurality of data series describe a plurality of operation profiles of a vehicle including at least one of a velocity series, or an engine temperature, or engine speed, or pedal positions, or steering angle, or a gear change function of a vehicle driving along a route, and the input to the second function of the discriminator model includes a plurality of velocity-acceleration pairs sampled from a velocity-acceleration histogram; or (iv) the plurality of data series describe a plurality of operation profiles of an autonomous or semiautonomous robot including at least one of a displacement series, or a velocity series, or a series representing the position of an actuator, of the autonomous or semi-autonomous robot” are insignificant extra-solution activity that amounts to no more than mere data gathering (See MPEP 2106.05(g)).
Step 2B: Claim 8 does not contain any additional elements that would amount to significantly more than the judicial exception. The elements of “or (iii) the plurality of data series describe a plurality of operation profiles of a vehicle including at least one of a velocity series, or an engine temperature, or engine speed, or pedal positions, or steering angle, or a gear change function of a vehicle driving along a route, and the input to the second function of the discriminator model includes a plurality of velocity-acceleration pairs sampled from a velocity-acceleration histogram; or (iv) the plurality of data series describe a plurality of operation profiles of an autonomous or semi-autonomous robot including at least one of a displacement series, or a velocity series, or a series representing the position of an actuator, of the autonomous or semi-autonomous robot” are insignificant extra-solution activity that amounts to no more than mere data gathering (See MPEP 2106.05(g)). Data gathering is a well-understood, routine conventional activity as recognized by the courts (See MPEP 2106.05(d)(II)). Stating that the plurality of data series describes operation profiles of a vehicle or robot does not impose meaningful limitations on a generative adversarial network.
Regarding claim 9,
Step 2A — Prong One: Claim 9 recites an abstract idea. The limitation of “wherein an output of the discriminator model is defined by r(frnnx1:T,fexp(x)) =λϕfrnnx1:T+1-λψfexpx1:T with λ∈[0,1]; wherein frnnx1:T is the first function, fexpx1:T is the second function, and ϕ(.) and ψ(.) are projections” is an abstract idea. The recited formula is clearly a mathematical formula or equation, and the determination is a mathematical calculation. Thus, the claim recites a mathematical formula or equation as well as a mathematical calculation, both of which fall within the mathematical concepts grouping of abstract ideas.
Step 2A — Prong Two: Claim 9 fails to integrate the judicial exception into practical application. The element of “implemented as multi-layer perceptrons” is a mere attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Step 2B: Claim 9 does not contain any additional elements that would amount to significantly more than the judicial exception. The element of “implemented as multi-layer perceptrons” is a mere attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Regarding claim 10,
Step 2A — Prong Two: Claim 10 fails to integrate the judicial exception into practical application. The element of “wherein the data series input to the second function of the discriminator model defines one or a plurality, of the following: integer number of vehicle stops per unit time, duration of vehicle acceleration to a predefined velocity, duration of vehicle deceleration to a predefined velocity, duration of vehicle deceleration to zero velocity, vehicle velocity versus route curvature, vehicle velocity versus route inclination, vehicle velocity versus total trip duration, integer number of gear changes per unit time, gear change function to reach a predefined velocity, gear change function to reach a predefined velocity, gear change function to reach zero velocity, gear change function versus route curvature, gear change function versus route inclination, gear change function versus total trip duration, robot actuator position” is insignificant extra-solution activity that amounts to no more than mere data gathering.
Step 2B: Claim 10 does not contain any additional elements that would amount to significantly more than the judicial exception. The element of “wherein the data series input to the second function of the discriminator model defines one or a plurality, of the following: integer number of vehicle stops per unit time, duration of vehicle acceleration to a predefined velocity, duration of vehicle deceleration to a predefined velocity, duration of vehicle deceleration to zero velocity, vehicle velocity versus route curvature, vehicle velocity versus route inclination, vehicle velocity versus total trip duration, integer number of gear changes per unit time, gear change function to reach a predefined velocity, gear change function to reach a predefined velocity, gear change function to reach zero velocity, gear change function versus route curvature, gear change function versus route inclination, gear change function versus total trip duration, robot actuator position” is insignificant extra-solution activity that amounts to no more than mere data gathering. Data gathering is a well-understood, routine conventional activity as recognized by the courts (See MPEP 2106.05(d)(II)). It is necessary to acquire the data in order to use the recited judicial exception to perform the calculation, and the “defines one or a plurality of the following” does not impose any other meaningful limits on the claim.
Regarding claim 11,
Step 2A — Prong One: Claim 11 recites an abstract idea. The limitation of “to generate a plurality of synthetic data series having a statistical distribution that is more similar to a statistical distribution of the plurality of data series, as compared to a statistical distribution of a plurality of synthetic data series that would be generated without use of the second function in the discriminator model” is an abstract idea directed to mental processes, for example, a human could use observation and evaluation to generate a plurality of synthetic data series having a statistical distribution that is more similar to a statistical distribution of the plurality of data series.
Step 2A — Prong Two: Claim 11 fails to integrate the judicial exception into practical application. The element of “wherein the trained generative machine learning model is configured” is a mere attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Step 2B: Claim 11 does not contain any additional elements that would amount to significantly more than the judicial exception. The element of “wherein the trained generative machine learning model is configured” is a mere attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Claim Rejections - 35 USC § 103
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claim(s) 1, 5-6, 10, and 12-14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Arnelid et al. (NPL: Recurrent Conditional Generative Adversarial Networks for Autonomous Driving Sensor Modelling, hereinafter "Arnelid") in view of Hyland et al. (NPL: Real-Valued (Medical) Time Series Generation with Recurrent Conditional GANS, hereinafter "Hyland").
Regarding claim 1, Arnelid teaches a computer-implemented method for training a generative machine learning model for modelling operation profiles of a vehicle or a robot, including adversarially training a generator model and a discriminator model, the method comprises the following steps:
obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot (Arnelid, Section 2A Paragraph 3 — “The data set that we use consists of sensor outputs collected from four days of driving on European highways and trunk roads, spanning over 12,000 multivariate time series, describing variables such as position, speed, heading etc.” — here, Arnelid teaches obtaining a plurality of data series wherein each series of the plurality of data described operation profiles of a vehicle);
generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution (Arnelid, Section 3C Paragraph 1 — “The generator learns a distribution pg over the data x, whereas the goal of the discriminator is to discriminate between the synthetic data G(z) generated by G and the real data x.” — here, Arnelid teaches during at least one generation phase generating a plurality of synthetic data series by sampling from a distribution pg. In addition to the previously cited passages, Arnelid further teaches in Section 3C Paragraph 1 – “In practice, this is a minimax game problem described with the value function V (D, G) as Eq. (8)” and in Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” & Fig. 2 – teaches the synthetic data G(z) generated by the generator G by sampling from a distribution (sample zt as in Fig. 2) during at least one generation phase (each pass of zt in G is a generation phase) when training (minimax game V(G, D)) the generative machine learning model);
inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series (Arnelid, Section 3D Paragraph 2 — “the discriminator takes either a real or synthetic sequence as input” — Arnelid teaches during at least one discrimination phase inputting either a data series from the plurality of obtained data series or synthetic data series from the plurality of synthetic data series);
classifying, during the at least one discrimination phase when training the generative machine learning model, the input to the discriminator model as either a data series from the data space or a synthetic data series (Arnelid, Section 3D Paragraph 2 — “the discriminator takes either a real or synthetic sequence as input where the network outputs a softmax probability at each time step classifying if the sample is real or synthetic” — Arnelid teaches the discriminator classifying the input as either a data series from the data space or a synthetic data series. In addition to the previously cited passages, Arnelid further teaches in Section 3C Paragraph 1 – “In practice, this is a minimax game problem described with the value function V (D, G) as Eq. (8)” and in Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” & Fig. 2 – teaches classifying (D(x|y) in Eq. (9) and in Fig. 2, shows sample xt, real or fake, being input to the discriminator to be classified), during at least one discrimination phase (each pass of xt in D(x|y) is a discrimination phase) when training (minimax game, V(G, D) the generative machine learning model), using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series (Arnelid, Section 3D Paragraph 4 — “The discriminator has a simple structure where the data point(s) from either a real or synthetic time series xt is being concatenated with the corresponding condition data yt at each time frame and is fed to a deep RNN.” — Arnelid teaches a first function based on a recurrent neural network in which synthetic data (or real data) is fed into for classification), and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;
Arnelid fails to explicitly teach at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model.
However, analogous to the field of generative adversarial networks, Hyland teaches:
at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series (Hyland, Section 3 Paragraph 1-2 — “we show how the discriminator RNN takes the generated sequence, together with an additional input if it is a RCGAN, and produces a classification as synthetic or real for each time step of the input sequence. Specifically, the discriminator is trained to minimise the average negative cross-entropy between its predictions per time-step and the labels of the sequence” — here, Hyland teaches a function that takes at least two data samples of synthetic data series (“takes the generated sequence”) taken from different steps (“per time step” of the synthetic data series); and
iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model (Hyland, Section 5.1 Paragraph 3 — “we train the RCGAN for 1000 epochs, saving one version of the dataset every 50 epochs” — here, Hyland describes iteratively training the generative adversarial network, which would comprise both the generator and discriminator, to yield a trained machine learning model).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the second function and iterative training of Hyland to the first function and data reception/generating of Arnelid to create a generative adversarial network to model the operation profiles of a vehicle or robot. Doing so would enable modeling the raw detections of sensors in autonomous vehicles (Arnelid, Section 1).
Regarding claim 12, Arnelid teaches a computer-implemented method for generating synthetic data samples representing synthetic operation profiles of a vehicle or robot using a generative machine learning model, comprising:
configuring a generative machine learning model according to a plurality of model parameters obtained by: obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot (Arnelid, Section 2A Paragraph 3 — “The data set that we use consists of sensor outputs collected from four days of driving on European highways and trunk roads, spanning over 12,000 multivariate time series, describing variables such as position, speed, heading etc.” — here, Arnelid teaches obtaining a plurality of data series wherein each series of the plurality of data described operation profiles of a vehicle);
generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution (Arnelid, Section 3C Paragraph 1 — “The generator learns a distribution pg over the data x, whereas the goal of the discriminator is to discriminate between the synthetic data G(z) generated by G and the real data x.” — here, Arnelid teaches during at least one generation phase generating a plurality of synthetic data series by sampling from a distribution pg. In addition to the previously cited passages, Arnelid further teaches in Section 3C Paragraph 1 – “In practice, this is a minimax game problem described with the value function V (D, G) as Eq. (8)” and in Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” & Fig. 2 – teaches the synthetic data G(z) generated by the generator G by sampling from a distribution (sample zt as in Fig. 2) during at least one generation phase (each pass of zt in G is a generation phase) when training (minimax game V(G, D)) the generative machine learning model);
inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series (Arnelid, Section 3D Paragraph 2 — “the discriminator takes either a real or synthetic sequence as input” — Arnelid teaches during at least one discrimination phase inputting either a data series from the plurality of obtained data series or synthetic data series from the plurality of synthetic data series);
classifying, during the at least one discrimination phase when training the generative machine learning model, the input to the discriminator model as either a data series from the data space or a synthetic data series, using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series (Arnelid, Section 3D Paragraph 2 — “the discriminator takes either a real or synthetic sequence as input where the network outputs a softmax probability at each time step classifying if the sample is real or synthetic” — Arnelid teaches the discriminator classifying the input as either a data series from the data space or a synthetic data series and in Section 3D Paragraph 4 — “The discriminator has a simple structure where the data point(s) from either a real or synthetic time series xt is being concatenated with the corresponding condition data yt at each time frame and is fed to a deep RNN.” — Arnelid teaches a first function based on a recurrent neural network in which synthetic data (or real data) is fed into for classification. In addition to the previously cited passages, Arnelid further teaches in Section 3C Paragraph 1 – “In practice, this is a minimax game problem described with the value function V (D, G) as Eq. (8)” and in Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” & Fig. 2 – teaches classifying (D(x|y) in Eq. (9) and in Fig. 2, shows sample xt, real or synthetic, being input to the discriminator to be classified), during at least one discrimination phase (each pass of xt in D(x|y) is a discrimination phase) when training (minimax game, V(G, D) the generative machine learning model), and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;
and outputting the further plurality of synthetic data samples representing synthetic operation profiles of a vehicle or robot (Arnelid, Section 2A Paragraph 1 — “The output from sensors used in ADAS and AD considered in this paper is in the form of dynamic state vectors over time, describing variables such as object position, velocity and acceleration relative to the host vehicle collecting the sensor data. These outputs from production sensors inherently exhibit noise and inaccuracies. The main contribution in this paper is to create a model for generating synthetic but realistic production sensor outputs.” — Arnelid teaches outputting the further plurality of synthetic data samples representing operation profiles of a vehicle).
Arnelid fails to explicitly teach at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series; and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model; obtaining a plurality of samples by sampling from a distribution configured to provide pseudo random vectors; decoding the plurality of the samples of the distribution to obtain a further plurality of synthetic data samples in the data space.
However, analogous to the field of generative adversarial networks, Hyland teaches:
and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series (Hyland, Section 3 Paragraph 1-2 — “we show how the discriminator RNN takes the generated sequence, together with an additional input if it is a RCGAN, and produces a classification as synthetic or real for each time step of the input sequence. Specifically, the discriminator is trained to minimise the average negative cross-entropy between its predictions per time-step and the labels of the sequence” — here, Hyland teaches a function that takes at least two data samples of synthetic data series (“takes the generated sequence”) taken from different steps (“per time step” of the synthetic data series);
and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model (Hyland, Section 5.1 Paragraph 3 — “we train the RCGAN for 1000 epochs, saving one version of the dataset every 50 epochs” — here, Hyland describes iteratively training the generative adversarial network, which would comprise both the generator and discriminator, to yield a trained machine learning model);
obtaining a plurality of samples by sampling from a distribution configured to provide pseudo random vectors (Hyland, Section 3 Paragraph 1 — “denote by RNN(X) the vector or matrix comprising the T outputs from a RNN receiving a sequence of T vectors” and to further supply the vectors are sampled from a distribution, in Section 3.1.1 Paragraph 1 — “We consider a GAN successful if it implicitly learns the distribution of the true data” — here, Hyland teaches obtain samples as vectors from a distribution, as the generative adversarial network is learning a distribution of the true data);
decoding the plurality of the samples of the distribution to obtain a further plurality of synthetic data samples in the data space (Hyland, Section 4.1 Paragraph 3 — “It consists of replacing the encoder and decoder of a VAE with RNNs, and then using the last hidden state of the encoder RNN as the encoded representation of the input sequence. After performing the reparameterization trick, the resulting encoded representation is used to initialize the hidden state of the decoder RNN.” — here, Hyland teaches using a hidden state of the decoder RNN to generate a further plurality of synthetic data samples);
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the second function and iterative training of Hyland to the first function and data reception/generating of Arnelid to create a generative adversarial network to model the operation profiles of a vehicle or robot. Doing so would introduce virtual verification to the field of autonomous vehicles, where models of the environment and sensors are made to match reality (Arnelid, Section 1).
Regarding claim 13, Arnelid teaches an apparatus for training a generative machine learning model for modelling operation profiles of a vehicle or robot comprising adversarially training a generator model and a discriminator model stored in a memory, the apparatus comprising:
an input interface configured to obtain a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot (Arnelid, Section 2A Paragraph 3 — “The data set that we use consists of sensor outputs collected from four days of driving on European highways and trunk roads, spanning over 12,000 multivariate time series, describing variables such as position, speed, heading etc.” — here, Arnelid teaches obtaining a plurality of data series wherein each series of the plurality of data describes operation profiles of a vehicle);
a processor configured to, during at least one generation phase when training the generative machine learning model, generate at least one synthetic data series by sampling from a distribution (Arnelid, Section 3C Paragraph 1 — “The generator learns a distribution pg over the data x, whereas the goal of the discriminator is to discriminate between the synthetic data G(z) generated by G and the real data x.” — here, Arnelid teaches during at least one generation phase generating a plurality of synthetic data series by sampling from a distribution pg. In addition to the previously cited passages, Arnelid further teaches in Section 3C Paragraph 1 – “In practice, this is a minimax game problem described with the value function V (D, G) as Eq. (8)” and in Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” & Fig. 2 – teaches the synthetic data G(z) generated by the generator G by sampling from a distribution (sample zt as in Fig. 2) during at least one generation phase (each pass of zt in G is a generation phase) when training (minimax game V(G, D)) the generative machine learning model);
wherein the processor is configured to, during at least one discrimination phase when training the generative machine learning model, input into the discriminator model either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the at least one synthetic data series (Arnelid, Section 3D Paragraph 2 — “the discriminator takes either a real or synthetic sequence as input” — Arnelid teaches during at least one discrimination phase inputting either a data series from the plurality of obtained data series or synthetic data series from the plurality of synthetic data series);
during the at least one discrimination phase when training the generative machine learning model, the processor is configured to classify the input to the discriminator model as either a data series from the data space, or a synthetic data series using at least a first and a second function of the discriminator model (Arnelid, Section 3D Paragraph 2 — “the discriminator takes either a real or synthetic sequence as input where the network outputs a softmax probability at each time step classifying if the sample is real or synthetic” — Arnelid teaches the discriminator classifying the input as either a data series from the data space or a synthetic data series. In addition to the previously cited passages, Arnelid further teaches in Section 3C Paragraph 1 – “In practice, this is a minimax game problem described with the value function V (D, G) as Eq. (8)” and in Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” & Fig. 2 – teaches classifying (D(x|y) in Eq. (9) and in Fig. 2, shows sample xt, real or fake, being input to the discriminator to be classified), during at least one discrimination phase (each pass of xt in D(x|y) is a discrimination phase) when training (minimax game, V(G, D) the generative machine learning model);
wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series (Arnelid, Section 3D Paragraph 4 — “The discriminator has a simple structure where the data point(s) from either a real or synthetic time series xt is being concatenated with the corresponding condition data yt at each time frame and is fed to a deep RNN.” — Arnelid teaches a first function based on a recurrent neural network in which synthetic data (or real data) is fed into for classification), and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;
Arnelid fails to explicitly teach at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series; wherein the processor is configured to iteratively train the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model; and wherein the apparatus further comprises: an output interface configured to output the trained machine learning model.
However, analogous to the field of generative adversarial networks, Hyland teaches:
and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series (Hyland, Section 3 Paragraph 1-2 — “we show how the discriminator RNN takes the generated sequence, together with an additional input if it is a RCGAN, and produces a classification as synthetic or real for each time step of the input sequence. Specifically, the discriminator is trained to minimise the average negative cross-entropy between its predictions per time-step and the labels of the sequence” — here, Hyland teaches a function that takes at least two data samples of synthetic data series (“takes the generated sequence”) taken from different steps (“per time step” of the synthetic data series);
wherein the processor is configured to iteratively train the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model (Hyland, Section 5.1 Paragraph 3 — “we train the RCGAN for 1000 epochs, saving one version of the dataset every 50 epochs” — here, Hyland describes iteratively training the generative adversarial network, which would comprise both the generator and discriminator, to yield a trained machine learning model);
and wherein the apparatus further comprises: an output interface configured to output the trained machine learning model (Hyland — Section 3.1 Paragraph 2 — “Therefore, in this work, we start by demonstrating our model with a number of toy datasets that can be visually evaluated.” — Hyland teaches outputting the trained machine learning model for visual evaluation on a multitude of datasets).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the second function and iterative training of Hyland to the first function and data reception/generating of Arnelid to create a generative adversarial network to model the operation profiles of a vehicle or robot. Doing so would introduce virtual verification to the field of autonomous vehicles, where models of the environment and sensors are made to match reality (Arnelid, Section 1).
Regarding claim 14, Arnelid teaches a non-transitory computer readable medium on which is stored a computer program for training a generative machine learning model for modelling operation profiles of a vehicle or a robot, including adversarially training a generator model and a discriminator model, the computer program, when executed by a computer, causing the computer to perform the following steps:
obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot (Arnelid, Section 2A Paragraph 3 — “The data set that we use consists of sensor outputs collected from four days of driving on European highways and trunk roads, spanning over 12,000 multivariate time series, describing variables such as position, speed, heading etc.” — here, Arnelid teaches obtaining a plurality of data series wherein each series of the plurality of data describes operation profiles of a vehicle);
generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution (Arnelid, Section 3C Paragraph 1 — “The generator learns a distribution pg over the data x, whereas the goal of the discriminator is to discriminate between the synthetic data G(z) generated by G and the real data x.” — here, Arnelid teaches during at least one generation phase generating a plurality of synthetic data series by sampling from a distribution pg. In addition to the previously cited passages, Arnelid further teaches in Section 3C Paragraph 1 – “In practice, this is a minimax game problem described with the value function V (D, G) as Eq. (8)” and in Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” & Fig. 2 – teaches the synthetic data G(z) generated by the generator G by sampling from a distribution (sample zt as in Fig. 2) during at least one generation phase (each pass of zt in G is a generation phase) when training (minimax game V(G, D)) the generative machine learning model);
inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series (Arnelid, Section 3D Paragraph 2 — “the discriminator takes either a real or synthetic sequence as input” — Arnelid teaches during at least one discrimination phase inputting either a data series from the plurality of obtained data series or synthetic data series from the plurality of synthetic data series);
classifying, during the at least one discrimination phase when training the generative machine learning model, the input to the discriminator model as either a data series from the data space or a synthetic data series (Arnelid, Section 3D Paragraph 2 — “the discriminator takes either a real or synthetic sequence as input where the network outputs a softmax probability at each time step classifying if the sample is real or synthetic” — Arnelid teaches the discriminator classifying the input as either a data series from the data space or a synthetic data series. In addition to the previously cited passages, Arnelid further teaches in Section 3C Paragraph 1 – “In practice, this is a minimax game problem described with the value function V (D, G) as Eq. (8)” and in Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” & Fig. 2 – teaches classifying (D(x|y) in Eq. (9) and in Fig. 2, shows sample xt, real or synthetic, being input to the discriminator to be classified), during at least one discrimination phase (each pass of xt in D(x|y) is a discrimination phase) when training (minimax game, V(G, D) the generative machine learning model), using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series (Arnelid, Section 3D Paragraph 4 — “The discriminator has a simple structure where the data point(s) from either a real or synthetic time series xt is being concatenated with the corresponding condition data yt at each time frame and is fed to a deep RNN.” — Arnelid teaches a first function based on a recurrent neural network in which synthetic data (or real data) is fed into for classification), and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;
Arnelid fails to explicitly teach at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series; and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model.
However, analogous to the field of generative adversarial networks, Hyland teaches:
and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series (Hyland, Section 3 Paragraph 1-2 — “we show how the discriminator RNN takes the generated sequence, together with an additional input if it is a RCGAN, and produces a classification as synthetic or real for each time step of the input sequence. Specifically, the discriminator is trained to minimise the average negative cross-entropy between its predictions per time-step and the labels of the sequence” — here, Hyland teaches a function that takes at least two data samples of synthetic data series (“takes the generated sequence”) taken from different steps (“per time step” of the synthetic data series);
and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model (Hyland, Section 5.1 Paragraph 3 — “we train the RCGAN for 1000 epochs, saving one version of the dataset every 50 epochs” — here, Hyland describes iteratively training the generative adversarial network, which would comprise both the generator and discriminator, to yield a trained machine learning model).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the second function and iterative training of Hyland to the first function and data reception/generating of Arnelid to create a generative adversarial network to model the operation profiles of a vehicle or robot. Doing so would introduce virtual verification to the field of autonomous vehicles, where models of the environment and sensors are made to match reality (Arnelid, Section 1).
Regarding claim 5, the combination of Arnelid and Hyland teach the computer-implemented method according to claim 1, wherein each data series of the plurality of data series is either a time series or a distance series, and wherein each synthetic data series of the plurality of synthetic data series is either a synthetic time series or a synthetic distance data series (Arnelid, Section 2D Paragraph 4 — “The discriminator has a simple structure where the data point(s) from either a real or synthetic time series xt is being concatenated with the corresponding condition data yt at each time frame and is fed to a deep RNN.” — here, Arnelid teaches wherein each of the data series and synthetic data series are both time series and synthetic time series, respectively).
Regarding claim 6, the combination of Arnelid and Hyland teach the computer implemented method according to claim 1, wherein, in the at least one generation phase, a latent space is sampled to provide the plurality of synthetic data series (Arnelid, Section 2D Paragraph 2 — “In the same manner as with the original GAN, the generator takes a latent vector z sampled from a known distribution as input” — here, Arnelid teaches sampling from a latent space for the generator to generate a plurality of synthetic data series).
Regarding claim 10, the combination of Arnelid and Hyland teach the computer-implemented method according to claim 1, wherein the data series input to the second function of the discriminator model defines one or a plurality, of the following: integer number of vehicle stops per unit time, duration of vehicle acceleration to a predefined velocity, duration of vehicle deceleration to a predefined velocity, duration of vehicle deceleration to zero velocity, vehicle velocity versus route curvature, vehicle velocity versus route inclination, vehicle velocity versus total trip duration, integer number of gear changes per unit time, gear change function to reach a predefined velocity, gear change function to reach a predefined velocity, gear change function to reach zero velocity, gear change function versus route curvature, gear change function versus route inclination, gear change function versus total trip duration, robot actuator position (Arnelid, Section 2A Paragraph 3 — “The data set that we use consists of sensor outputs collected from four days of driving on European highways and trunk roads, spanning over 12,000 multivariate time series, describing variables such as position, speed, heading etc.” — here, Arnelid teaches the data series input to the discriminator defining one or a plurality of position, speed, heading, etc. which would include vehicle stops, duration of acceleration to predefined velocity, deceleration to zero velocity, velocity versus route curvature, and velocity versus route inclination).
Claim(s) 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Arnelid and Hyland, and further in view of Meadows et al. (US Pub. No. 2019/0122409, hereinafter "Meadows").
Regarding claim 15, Arnelid teaches a computer-implemented method for training a further machine learning model for modelling operation profiles of a vehicle, comprising:
obtaining a further plurality of synthetic data series representing synthetic operation profiles of a vehicle as generated by:
configuring a generative machine learning model according to a plurality of model parameters obtained by: obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot (Arnelid, Section 2A Paragraph 3 — “The data set that we use consists of sensor outputs collected from four days of driving on European highways and trunk roads, spanning over 12,000 multivariate time series, describing variables such as position, speed, heading etc.” — here, Arnelid teaches obtaining a plurality of data series wherein each series of the plurality of data describes operation profiles of a vehicle);
generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution (Arnelid, Section 3C Paragraph 1 — “The generator learns a distribution pg over the data x, whereas the goal of the discriminator is to discriminate between the synthetic data G(z) generated by G and the real data x.” — here, Arnelid teaches during at least one generation phase generating a plurality of synthetic data series by sampling from a distribution pg. In addition to the previously cited passages, Arnelid further teaches in Section 3C Paragraph 1 – “In practice, this is a minimax game problem described with the value function V (D, G) as Eq. (8)” and in Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” & Fig. 2 – teaches the synthetic data G(z) generated by the generator G by sampling from a distribution (sample zt as in Fig. 2) during at least one generation phase (each pass of zt in G is a generation phase) when training (minimax game V(G, D)) the generative machine learning model);
inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series (Arnelid, Section 3D Paragraph 2 — “the discriminator takes either a real or synthetic sequence as input” — Arnelid teaches during at least one discrimination phase inputting either a data series from the plurality of obtained data series or synthetic data series from the plurality of synthetic data series);
classifying, during the at least one discrimination phase when training the generative machine learning model, the input to the discriminator model as either a data series from the data space or a synthetic data series (Arnelid, Section 3D Paragraph 2 — “the discriminator takes either a real or synthetic sequence as input where the network outputs a softmax probability at each time step classifying if the sample is real or synthetic” — Arnelid teaches the discriminator classifying the input as either a data series from the data space or a synthetic data series. In addition to the previously cited passages, Arnelid further teaches in Section 3C Paragraph 1 – “In practice, this is a minimax game problem described with the value function V (D, G) as Eq. (8)” and in Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” & Fig. 2 – teaches classifying (D(x|y) in Eq. (9) and in Fig. 2, shows sample xt, real or synthetic, being input to the discriminator to be classified), during at least one discrimination phase (each pass of xt in D(x|y) is a discrimination phase) when training (minimax game, V(G, D) the generative machine learning model), using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series (Arnelid, Section 3D Paragraph 4 — “The discriminator has a simple structure where the data point(s) from either a real or synthetic time series xt is being concatenated with the corresponding condition data yt at each time frame and is fed to a deep RNN.” — Arnelid teaches a first function based on a recurrent neural network in which synthetic data (or real data) is fed into for classification), and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;
outputting the further plurality of synthetic data samples representing synthetic operation profiles of a vehicle or robot (Arnelid, Section 2A Paragraph 1 — “The output from sensors used in ADAS and AD considered in this paper is in the form of dynamic state vectors over time, describing variables such as object position, velocity and acceleration relative to the host vehicle collecting the sensor data. These outputs from production sensors inherently exhibit noise and inaccuracies. The main contribution in this paper is to create a model for generating synthetic but realistic production sensor outputs.” — Arnelid teaches outputting the further plurality of synthetic data samples representing operation profiles of a vehicle);
Arnelid fails to explicitly teach and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series; and iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model; obtaining a plurality of samples by sampling from a distribution configured to provide pseudo random vectors; decoding the plurality of the samples of the distribution to obtain a further plurality of synthetic data samples in the data space; inputting the further plurality of synthetic data series representing synthetic operation profiles of a vehicle into a further machine learning model; iteratively training the further machine learning model; and outputting a further plurality of model parameters of the further machine learning model for modelling operation profiles of a vehicle.
However, analogous to the field of the claimed invention, Hyland teaches:
and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series (Hyland, Section 3 Paragraph 1-2 — “we show how the discriminator RNN takes the generated sequence, together with an additional input if it is a RCGAN, and produces a classification as synthetic or real for each time step of the input sequence. Specifically, the discriminator is trained to minimise the average negative cross-entropy between its predictions per time-step and the labels of the sequence” — here, Hyland teaches a function that takes at least two data samples of synthetic data series (“takes the generated sequence”) taken from different steps (“per time step” of the synthetic data series);
iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model (Hyland, Section 5.1 Paragraph 3 — “we train the RCGAN for 1000 epochs, saving one version of the dataset every 50 epochs” — here, Hyland describes iteratively training the generative adversarial network, which would comprise both the generator and discriminator, to yield a trained machine learning model);
obtaining a plurality of samples by sampling from a distribution configured to provide pseudo random vectors (Hyland, Section 3 Paragraph 1 — “denote by RNN(X) the vector or matrix comprising the T outputs from a RNN receiving a sequence of T vectors” and to further supply the vectors are sampled from a distribution, in Section 3.1.1 Paragraph 1 — “We consider a GAN successful if it implicitly learns the distribution of the true data” — here, Hyland teaches obtain samples as vectors from a distribution, as the generative adversarial network is learning a distribution of the true data);
decoding the plurality of the samples of the distribution to obtain a further plurality of synthetic data samples in the data space (Hyland, Section 4.1 Paragraph 3 — “It consists of replacing the encoder and decoder of a VAE with RNNs, and then using the last hidden state of the encoder RNN as the encoded representation of the input sequence. After performing the reparameterization trick, the resulting encoded representation is used to initialize the hidden state of the decoder RNN.” — here, Hyland teaches using a hidden state of the decoder RNN to generate a further plurality of synthetic data samples);
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the second function and iterative training of Hyland to the first function and data reception/generating of Arnelid to create a generative adversarial network to model the operation profiles of a vehicle or robot. Doing so would introduce virtual verification to the field of autonomous vehicles, where models of the environment and sensors are made to match reality (Arnelid, Section 1).
The combination of Arnelid and Hyland fails to explicitly teach inputting the further plurality of synthetic data series representing synthetic operation profiles of a vehicle into a further machine learning model; iteratively training the further machine learning model; and outputting a further plurality of model parameters of the further machine learning model for modelling operation profiles of a vehicle.
However, analogous to the field of generative adversarial networks, Meadows teaches:
inputting the further plurality of synthetic data series representing synthetic operation profiles of a vehicle into a further machine learning model (Meadows, [0108] — “the given behavioral agent may receive one or more inputs (which may include one or more outputs from one or more of the other behavioral agents) and may provide an output corresponding to one or more features” — here, Meadows teaches inputting a further plurality of synthetic data representing operation profiles of an agent, to further supply the agent is given generated data, in [0137] — “a generative adversarial network is used to train a group of behavioral agents”);
iteratively training the further machine learning model (Meadows, [0150] — “in some embodiments of method 600, at least some of the operations are performed iteratively until a convergence criterion is achieved” — here, Meadows teaches iteratively training the generative adversarial network until convergence is achieved); and
outputting a further plurality of model parameters of the further machine learning model for modelling operation profiles of a vehicle (Meadows, [0108] — “the parameters of and/or the outputs from the given behavioral agent may be recorded, which may be used as part of a future training dataset for the group of behavioral agents” — here, Meadows teaches outputting a further plurality of parameters and other outputs for further use in modeling operation profiles of agents, wherein the agents are trained by a generative adversarial network in [0137] — “a generative adversarial network is used to train a group of behavioral agents”).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the further machine learning model of Meadows to the method of Arnelid and Hyland in order to further supply the model parameters to the further machine learning model to model operation profiles. Doing so would enable the computer system to continue to develop a training dataset that facilitates improved mimicking of the one or more attributes of the individual by the group of behavioral agents (Meadows, [0107]).
Claim(s) 2 is/are rejected under 35 U.S.C. 103 as being unpatentable over Arnelid and Hyland as applied to claim 1 above, and further in view of Goodfellow et al. (NPL: Generative Adversarial Nets, hereinafter "Goodfellow").
Regarding claim 2, the combination of Arnelid and Hyland teach the computer-implemented method according to claim 1, wherein an input of the first function is a time indexed function of the input to the discriminator model (Arnelid, Section 3B — “Then the output gate is computed in Equation (6), and lastly the final output ht is computed in Equation (7).” — Arnelid teaches an output h, where t represents a time index, and the output is based on time indexed inputs in equations 3-5),
The combination of Arnelid and Hyland fails to explicitly teach and an input of the second function is a differentiable function of the input to the discriminator model.
However, analogous to the field of generative adversarial networks, Goodfellow teaches:
an input of the second function is a differentiable function of the input to the discriminator model (Goodfellow, Section 3 Paragraph 1 — “where G is a differentiable function represented by a multilayer perceptron with parameters θg. We also define a second multilayer perceptron D(x; θd) that outputs a single scalar” — here, Goodfellow teaches a multilayer perceptron discriminator and in Equation 1 — “D(G(z))” represents input G(z), which is a differentiable function, into the discriminator).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the differentiable input of Goodfellow to the method of Arnelid and Hyland in order to establish unique inputs into the functions of the discriminator. Doing so would maximize the probability of the discriminator assigning the correct label to both training examples and samples from the generator (Goodfellow, Section 3).
Claim(s) 3 is/are rejected under 35 U.S.C. 103 as being unpatentable over Arnelid and Hyland as applied to claim 1 above, and further in view of Miyato et al. (NPL: cGANs with Projection Discriminator, hereinafter "Miyato").
Regarding claim 3, the combination of Arnelid and Hyland teach the computer-implemented method according to claim 1, further comprising: combining outputs of the first and second functions using a discriminator neural network, wherein: the first function is a recurrent neural network, optionally a Long Short Term Memory Network (Arnelid, Section 3B Paragraph 1 — “A drawback of using vanilla RNNs that are trained with gradient descent methods is the vanishing gradient problem where the gradients corresponding to past observations become vanishingly small and weights do not get updated properly. One solution to this problem is the Long Short-Term Memory network which incorporates a memory cell ¢ together with an input gate i, an output gate o and a forget gate f [14]. The memory cell enables the network to remember its state over time, and by doing so it is possible for the full network to capture long-term temporal dependencies present in the training data.” — here, Arnelid teaches a first function being a recurrent neural network, specifically a Long Short Term Memory Network),
The combination of Arnelid and Hyland fail to explicitly teach and/or the second function is obtained by projecting at least one of the synthetic data series and obtaining a result of the second function of the discriminator model based on a combination of the projected at least one synthetic data series and at least one data series from the plurality of data series.
However, analogous to the field of generative adversarial networks, Miyato teaches:
and/or the second function is obtained by projecting at least one of the synthetic data series and obtaining a result of the second function of the discriminator model based on a combination of the projected at least one synthetic data series and at least one data series from the plurality of data series (Miyato, Section 5.1 Paragraph 1 — “Our proposed projection model discriminator is equipped with a ‘projection layer’ that takes inner product between the embedded one-hot vector y and the intermediate output” — here, Miyato teaches a function of the discriminator that is obtained by projecting at least one of the synthetic data series, in Miyato represented by variable x, and obtaining a result of the second function based on a combination of the projected synthetic data series and at least one data series from the plurality of data series, in Miyato represented by variable y).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the projection discriminator of Miyato to the method of Arnelid and Hyland in order to create a discriminator based on two functions. Doing so would introduce the label information via an inner product, as opposed to concatenation (Miyato, Section 3 Paragraph 5).
Claim(s) 4 and 9 is/are rejected under 35 U.S.C. 103 as being unpatentable over Arnelid, Hyland, and Miyato as applied to claim 3 above, and further in view of Zheng et al. (NPL from IDS: Data synthesis using dual discriminator conditional generative adversarial networks for imbalanced unit fault diagnosis of rolling bearings, hereinafter “Zheng”).
Regarding claim 4, the combination of Arnelid, Hyland, and Miyato teach the computer-implemented method according to claim 3.
The combination of Arnelid, Hyland, and Miyato fails to explicitly teach wherein the discriminator neural network calculates a linear combination of the first and second functions of the discriminator model.
However, analogous to the field of generative adversarial networks, Zheng teaches:
wherein the discriminator neural network calculates a linear combination of the first and second functions of the discriminator model (Zheng, Section 2.3 Paragraph 3 — “In training process, the loss of two discriminators is combined, and generator training in D2CGANs is consistent with GAN.” — here, Zheng teaches combining the values of the dual discriminators in a linear combination, represented by the sum in Eq. 5).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the linear combination of Zheng to the functions of Arnelid, Hyland, and Miyato in order to combine the functions of the discriminator. Doing so would address issues of mode collapse in generative adversarial networks (Zheng, Section 1).
Regarding claim 9, the combination of Arnelid and Hyland teach the computer-implemented method according to claim 1, wherein an output of the discriminator model is defined by r(frnnx1:T,fexp(x)) =λϕfrnnx1:T+1-λψfexpx1:T with λ∈[0,1]; wherein frnnx1:T is the first function, (Arnelid, Section 3D Paragraph 4 — “The discriminator has a simple structure where the data point(s) from either a real or synthetic time series xt is being concatenated with the corresponding condition data yt at each time frame and is fed to a deep RNN.” — Arnelid teaches a first function based on a recurrent neural network in which synthetic data (or real data) is fed into for classification),
The combination of Arnelid and Hyland fails to explicitly teach fexp(x1:T) is the second function , and ¢(.) and y(.) are projections, implemented as multi-layer perceptrons.
However, analogous to the field of generative adversarial networks, Miyato teaches:
fexpx1:T is the second function, and ϕ(.) and ψ(.) are projections, (Miyato, Section 5.1 Paragraph 1 — “Our proposed projection model discriminator is equipped with a ‘projection layer’ that takes inner product between the embedded one-hot vector y and the intermediate output” — here, Miyato teaches a function of the discriminator that is obtained by projecting at least one of the synthetic data series, creating an expert function for the second function of the discriminator).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the second function and projections of Miyato to the method of Arnelid and Hyland in order to further enhance the discriminator with a second function capable of projecting expert features. Doing so would incorporate additional information via inner product as opposed to concatenation (Miyato, Section 3 Paragraph 5).
The combination of Arnelid, Hyland, and Miyato fails to explicitly teach implemented as multi-layer perceptrons.
However, analogous to the field of generative adversarial networks, Zheng teaches:
implemented as multi-layer perceptrons (Zheng, Table 6 and 7 describe the time consumption of the dual discriminator model implemented with multi-layer perceptrons).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention to incorporate the multi-layer perceptrons of Zheng to the method of Arnelid, Hyland, and Miyato in order to implement the discriminator functions as a multi-layer perceptron. Doing so would assist in faster fault classifier training and ensure timely outputs for online applications (Zheng, Page 10 Paragraph 2).
Claim(s) 7 is/are rejected under 35 U.S.C. 103 as being unpatentable over Arnelid and Hyland as applied to claim 1 above, and further in view of Zhang et al. (FOR: CN 112634429B, hereinafter “Zhang”).
Regarding claim 7, the combination of Arnelid and Hyland teach the computer implemented method according to claim 1.
The combination of Arnelid and Hyland fails to explicitly teach further comprising: stopping the iterative training of the generator model using at least one stopping criterion, wherein the at least one stopping criterion is obtained by evaluating at least one synthetic data series generated by an iteration of a generator model using the second function of the discriminator model, wherein the second function outputs the stopping criterion.
However, analogous to the field of generative adversarial networks, Zhang teaches:
further comprising: stopping the iterative training of the generator model using at least one stopping criterion, wherein the at least one stopping criterion is obtained by evaluating at least one synthetic data series generated by an iteration of a generator model using the second function of the discriminator model, wherein the second function outputs the stopping criterion (Zhang, Page 5 Paragraph 7 — “the invention claims a Early-stop training monitoring mechanism, namely through the loss of the arbiter and a discriminator for the combined evaluation of the evaluation value S (tassel) of the generated sample, determining whether to start the Early-stop termination current training” — here, Zhang teaches an early stopping mechanism based on a function of the discriminator with an input of a generated (synthetic) data sample).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the early stopping mechanism of Zhang to the method of Arnelid and Hyland in order to establish an early stopping mechanism based on a function of the discriminator. Doing so would improve training efficiency and ensuring the generated sample is more similar to the real sample (Zhang, Page 6 Paragraph 4).
Claim(s) 8 is/are rejected under 35 U.S.C. 103 as being unpatentable over Arnelid and Hyland as applied to claim 1 above, and further in view of Li et al. (NPL: His-GAN: A histogram-based GAN to improve data generation quality, hereinafter "Li").
Regarding claim 8, the combination of Arnelid and Hyland teach the computer-implemented method according to claim 1, wherein: (i) the second function of the discriminator model is evaluated on a series of values sampled from a distribution resulting from real or synthetic data series based on at least one data series representing vehicle velocity, and at least one synthetic data series representing synthetic vehicle velocity (Arnelid, Section 2A Paragraph 3 — “The data set that we use consists of sensor outputs collected from four days of driving on European highways and trunk roads, spanning over 12,000 multivariate time series, describing variables such as position, speed, heading etc.” — here, Arnelid teaches the data consists of sensors that capture variable such as position, speed, heading, etc., which would comprise data series and synthetic data series representing vehicle velocity to be input to the second function of the discriminator of Hyland); or (ii) the second function of the discriminator model is configured to evaluate a time spent at zero velocity based on at least one synthetic data series representing vehicle velocity (Arnelid, Section 2A Paragraph 3 — “The data set that we use consists of sensor outputs collected from four days of driving on European highways and trunk roads, spanning over 12,000 multivariate time series, describing variables such as position, speed, heading etc.” — here, Arnelid teaches the data consisting speed, position, heading, etc., which would also include a time spent at zero velocity based on at least one synthetic data series representing vehicle velocity to be evaluated by the second function of Hyland); or (iii) the plurality of data series describe a plurality of operation profiles of a vehicle including at least one of a velocity series, or an engine temperature, or engine speed, or pedal positions, or steering angle, or a gear change function of a vehicle driving along a route (Arnelid, Section 2A Paragraph 3 — “The data set that we use consists of sensor outputs collected from four days of driving on European highways and trunk roads, spanning over 12,000 multivariate time series, describing variables such as position, speed, heading etc.” — here, Arnelid teaches the data consisting speed, position, heading, etc., which would also include at least one synthetic data series representing vehicle velocity or engine speed to be evaluated by the second function of Hyland), and the input to the second function of the discriminator model includes a plurality of velocity-acceleration pairs sampled from a velocity-acceleration histogram; or (iv) the plurality of data series describe a plurality of operation profiles of an autonomous or semi-autonomous robot including at least one of a displacement series, or a velocity series, or a series representing the position of an actuator, of the autonomous or semi-autonomous robot (Arnelid, Section 2A Paragraph 3 — “The data set that we use consists of sensor outputs collected from four days of driving on European highways and trunk roads, spanning over 12,000 multivariate time series, describing variables such as position, speed, heading etc.” — here, Arnelid teaches the data consisting speed, position, heading, etc., for autonomous vehicles, thus also provides a plurality of data describing at least one of a displacement series or velocity series of an autonomous robot).
The combination of Arnelid and Hyland fails to explicitly teach and the input to the second function of the discriminator model includes a plurality of velocity-acceleration pairs sampled from a velocity-acceleration histogram.
However, analogous to the field of generative adversarial networks, Li teaches:
and the input to the second function of the discriminator model includes a plurality of velocity-acceleration pairs sampled from a velocity-acceleration histogram (Li, Section 4.2 Paragraph 3 — “For numeric datasets, we can use an array with arbitrary size to capture each number. For example, there is a dataset with two features ranging [[35, 60], [38, 65]]. Here we can use an array with size of 2 to save those values, array [0] indicates the range from 30 to 40 and its value is 2, while array [1] indicates the range from 60 to 70 and its value is also 2. Also, we can use an array with size of 10 to save those values, array[3] indicates the range from 30 to 40 and array[6] indicates the range from 60 to 70. From a statistical perspective, we do not care where the value is from, we just count this value in histogram.” — here, Li teaches mapping a histogram of numeric datasets to a generative adversarial network, which includes counting values of velocity-accelerations pairs for input to the second function of the discriminator).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the histogram of Li to the method of Arnelid and Hyland in order to sample data from the histogram for use in the generative adversarial network. Doing so would improve the quality of the generated data (Li, Section 1 Paragraph 7).
Claim(s) 11 is/are rejected under 35 U.S.C. 103 as being unpatentable over Arnelid and Hyland as applied to claim 1 above, and further in view of Zheng et al. (NPL From IDS: Data synthesis using dual discriminator conditional generative adversarial networks for imbalanced unit fault diagnosis of rolling bearings, hereinafter "Zheng").
Regarding claim 11, the combination of Arnelid and Hyland teaches the computer-implemented method of claim 1.
The combination of Arnelid and Hyland fails to explicitly teach wherein the trained generative machine learning model is configured to generate a plurality of synthetic data series having a statistical distribution that is more similar to a statistical distribution of the plurality of data series, as compared to a statistical distribution of a plurality of synthetic data series that would be generated without use of the second function in the discriminator model.
However, analogous to the field of generative adversarial networks, Zheng teaches:
wherein the trained generative machine learning model is configured to generate a plurality of synthetic data series having a statistical distribution that is more similar to a statistical distribution of the plurality of data series, as compared to a statistical distribution of a plurality of synthetic data series that would be generated without use of the second function in the discriminator model (Zheng, Section 5 Paragraph 1 - “The method combines the advantage of CGAN which can learn the multimodal data distribution under help of conditional information and D2GAN which can avoid mode collapse in data synthesis process using dual discriminator structure.” — here, Zheng teaches generating a plurality of synthetic data series having a distribution more similar to a distribution of the plurality of the data series, as a result of generating the synthetic data series using a two function discriminator (dual discriminators).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the linear combination of Zheng to the functions of Arnelid and Hyland in order to combine the functions of the discriminator. Doing so would to help accurate model training of data synthesis, and dual discriminator structure aims to overcome mode collapse problem of multimodal generation task and improve quality and diversity of the synthesized data for augmenting imbalanced dataset (Zheng, Section 1).
Response to Arguments
Applicant's arguments on pages 1-8 of Remarks regarding the 35 U.S.C. 101 rejection of claims 1-15, filed 23 July 2025, have been fully considered but they are not persuasive.
Applicant argues on page 3, first paragraph that, “Similarly, the claims at issue here are directed to ‘training a generative machine learning model for modelling operation profiles of a vehicle or robot.’ Since the specification addressed at [0055] explains that ‘[a] generative model is a neural network architecture…,’ the claim assessed in Example 39 is relevant to the claims rejected under Section 101 here.” Examiner respectfully disagrees. The claim assessed in Example 39 does not recite an abstract idea, while the claims at issue in the present application do recite an abstract idea in “classifying, during at least one discrimination phase when training the generative machine learning model, the input to the discriminator model as either a data series from the data space or a synthetic data series, using at least a first and a second function of the discriminator model, wherein a plurality of inputs of the first function include a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series;” which is a mental process.
Applicant states on page 5, first complete paragraph, “For instance, the Patent Office does not explain how the cognitive faculties of a human being, with no assistance except that which is provided by pen and paper, can time the performance of the claimed classifying to occur ‘during the at least one discrimination phase when training the generative machine learning model.’ In addition, the Patent Office fails to explain why the mental powers of a human being can practically perform actually supply to first and second functions of a discriminator model the particular synthetic data series in the manner recited in the claim.” Examiner respectfully disagrees. As stated in the previous office action, a human could use evaluation and judgement to classify, during a discrimination phase when training the generative machine learning model, the input to the discriminator model as either a data series from the data space or a synthetic data series, using at least a first and second function of the discriminator model. A human could evaluate the data being input to the discriminator during a discrimination phase, and use judgement to classify whether the data is from the data space or a synthetic data series. A human could evaluate the classification of the data using a first and second function of the discriminator model, wherein the first function of the discriminator model includes a plurality of sequential samples of the synthetic data series, and at least one input of the second function includes at least two data samples of the synthetic data series taken from different steps of the synthetic data series. A human could evaluate the input to the discriminator, during a phase in which the input will be discriminated (or classified/evaluated by the human), write the data using a physical aid, determine the respective inputs to the first and second functions, and, using judgement, classify the observed data as either a synthetic data series or as data from the data space. The classifying limitation, as drafted, is a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic training. That is, other than reciting “when training the generative machine learning model”, nothing in the claim element precludes the step from practically being performed in the human mind. The mere nominal recitation of the generic training does not take the classifying limitation out of the mental process grouping.
Applicant argues, on page 8, second paragraph that “Thus, the iterative training limitation represents a technological advance in the field of generative machine learning model training. Thus, the specification, which is to be consulted under Prong Two, establishes that the technical field of generative machine learning model training has been improved for the reasons given above, and the claims reflect this improvement by reciting the claim limitation quoted above that the specification links to this improvement in better quality simulations. Applicant notes, however, that the Patent Office does not evaluate Prong Two in light of the specification, as it ought to have.” Examiner respectfully disagrees. MPEP 2106.05(a) states that, “After the examiner has consulted the specification and determined that the disclosed invention improves technology, the claim must be evaluated to ensure the claim itself reflects the disclosed improvement in technology.” Upon reference the specification, in [0012], “An effect is that the synthetic data generated by a machine learning model learning model trained according to the method match, or more closely approximate, a real distribution with respect to specific characteristics of a real vehicle application scenario. This results in better quality simulations when a generator trained using the discriminator is used to generate synthetic data samples.” The specification provides a purported improvement of better quality simulations when a generator trained using the discriminator is used to generate synthetic data samples. Upon reference to claim 1, as drafted, the limitation of “iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a generator” does not reflect the improvement in [0012], and thus does not integrate the judicial exception into practical application and does not amount to significantly more than the judicial exception. Simulations are not recited in the claim or as to how the generator trained by the discriminator improves simulations. As stated in the previous office action, the additional elements of claim 1, “obtaining a plurality of data series from a data space, wherein each data series of the plurality of data series describes at least one operation profile of a vehicle or robot; generating, during at least one generation phase when training the generative machine learning model, a plurality of synthetic data series by sampling from a distribution; inputting into the discriminator model, during at least one discrimination phase when training the generative machine learning model, either (i) a data series from the plurality of obtained data series, or (ii) a synthetic data series from the plurality of synthetic data series;” are insignificant extra-solution activities (See MPEP 2106.05(g)) that amount to no more than mere data gathering. The additional element of claim 1, “iteratively training the generator model and the discriminator model of the generative machine learning model to yield a trained machine learning model comprising a trained generator model” is an attempt to generally link the use of the judicial exception to the technological environment of a computer (See MPEP 2106.05(h)).
Applicant's arguments on pages 9-12 of Remarks regarding the 35 U.S.C. 103 rejection of independent claims 1 and 12-14, filed 23 July 2025, have been fully considered but they are not persuasive. Applicant argues on page 9, last paragraph that, “no synthetic data series generation is disclosed in this blurb from Arnelid as occurring ‘during at least one generation phase when training the generative machine learning model.’ That is, the blurb discloses ‘synthetic data G(z) generated by G,’ but does not disclose that this generation is performed ‘during at least one generation phase when training the generative machine learning model.’ Specifically, even if the ‘generator learning a distribution pg’ in this blurb refers to a training of some kind, the use of the past participle ‘generated’ in relation to the synthetic data G(z) indicates that the synthetic data of Arnelid has already been generated by the time distribution learning occurs.” Examiner respectfully disagrees, and points to Page 10, Lines 17-22 of the specification of the claimed invention, which states “In other words, the training process of the cGAN involves optimizing a loss function for the generative model 10a and loss function for the discriminator model 12. More specifically, the training process is a minimax process involving minimizing the loss function of the discriminator model 12 and minimizing the loss function of the generative 10a error (which implies maximizing the loss of the discriminator for the synthetic data)”, which clearly states that the training of the cGAN is a minimax process. In addition to the previously cited passages, Arnelid further teaches in Section 2C Paragraph 1 –“ In practice, this is a minimax game problem described with the value function V (D, G) as Eq. 8” and further in Fig. 2 & Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” – which teaches training of the generative machine learning model (by a minimax game, the same as the training method recited in Page 10, Lines 17-22 of the specification of the claimed invention), and the synthetic data G(z) generated by the generator G by sampling from a distribution (sample zt as in Fig. 2) during at least one generation phase (each pass of zt in G is a generation phase) when training (minimax game V(G, D)) the generative machine learning model.
Applicant argues on page 10, first paragraph that, “Second, it is not the case that Arnelid discloses ‘classifying, during the at least one discrimination phase when training the generative machine learning model, the input to the discriminator model as either a data series from the data space or a synthetic data series…’” and on page 10, second paragraph, “To meet this claim limitation, Arnelid would have to disclose that the classifying discloses in this blurb is performed, ‘during the at least one discrimination phase when training the generative machine learning model.’ Nothing in this blurb establishes this to be the case. Indeed, the blurb fails to mention any training whatsoever. Thus, the absence of any explanation by the Patent Office as to how the ‘during’ condition applies to the classification performed by the Arnelid discriminator amounts to a failure to comply with the clear explanation mandate of 37 C.F.R. 1.104(c)(2)…”. Examiner respectfully disagrees, and points to Page 10, Lines 17-22 of the specification of the claimed invention, which states “In other words, the training process of the cGAN involves optimizing a loss function for the generative model 10a and loss function for the discriminator model 12. More specifically, the training process is a minimax process involving minimizing the loss function of the discriminator model 12 and minimizing the loss function of the generative 10a error (which implies maximizing the loss of the discriminator for the synthetic data”, which clearly states that the training of the cGAN is a minimax process. In addition to the previously cited passages, Arnelid further teaches in Section 3C Paragraph 1 – “In practice, this is a minimax game problem described with the value function V (D, G) as Eq. (8)” and in Section 3D Paragraph 1 – “The second change is that the output from both the generator and the discriminator is conditioned on an input vector y, which makes it possible for the GAN to learn the conditional probability distribution p(x|y) as described in [2]. The value function is then expressed as Eq. (9)” & Fig. 2 – which teaches training of the generative machine learning model (by a minimax game, the same as the training method recited in Page 10, Lines 17-22 of the specification of the claimed invention), and teaches classifying the input to the discriminator model (D(x|y) in Eq. (9) and in Fig. 2, shows sample xt, real or synthetic, being input to the discriminator to be classified), during at least one discrimination phase (each pass of xt in D(x|y) is a discrimination phase) when training (minimax game, V(G, D) the generative machine learning model.
Applicant argues on page 11 last paragraph – page 12 first paragraph, that “In short, the Patent Office combines these references on the basis of impermissible hindsight.” In response to applicant's argument that the examiner's conclusion of obviousness is based upon improper hindsight reasoning, it must be recognized that any judgment on obviousness is in a sense necessarily a reconstruction based upon hindsight reasoning. But so long as it takes into account only knowledge which was within the level of ordinary skill at the time the claimed invention was made, and does not include knowledge gleaned only from the applicant's disclosure, such a reconstruction is proper. See In re McLaughlin, 443 F.2d 1392, 170 USPQ 209 (CCPA 1971).
Regarding claim 1, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the teachings of Hyland to the method of Arnelid, as doing so would enable modeling the raw detections of a sensor, specifically advance driver assistance system and autonomous driving sensors (Arnelid, Section 1), in which the modeling of these autonomous driving sensors would capture the stochastic behaviors of sensor and thus capture an operation profile of a vehicle. In the previous office action with regard to independent claims 12-14, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to combine Arnelid and Hyland to introduce virtual verification to the field of autonomous vehicles, such that models of the environment and sensors are made to match reality (Arnelid, Section 1).
Applicant's arguments, regarding claims 5-6 and 10, filed 23 July 2025, have been fully considered but they are not persuasive. Applicant argued claims 5-6 and 10 based on the virtue of their dependence, directly or indirectly, from independent claim 1. As set forth in the current Office Action, claim 1 stands rejected under Arnelid and Hyland. No separate arguments were presented for claims 5-6 and 10. The rejection of this dependent claim is maintained.
Applicant's arguments, regarding claim 15, filed 23 July 2025 have been fully considered but they are not persuasive. Applicant argues claim 15 on the same arguments presented for claim 1. Examiner respectfully disagrees, and points to the above response to arguments with regard to 35 U.S.C. 103 rejection of claim 1. Claim 15 stands rejected under Arnelid, in view of Hyland, and further in view of Meadows.
Applicant's arguments, regarding claim 2, filed 23 July 2025, have been fully considered but they are not persuasive. Applicant argued claim 2 based on the virtue of its dependence, directly or indirectly, from independent claim 1. As set forth in the current Office Action, claim 1 stands rejected under Arnelid and Hyland. No separate arguments were presented for claim 2. The rejection of this dependent claim is maintained.
Applicant's arguments, regarding claim 3, filed 23 July 2025, have been fully considered but they are not persuasive. Applicant argued claim 3 based on the virtue of its dependence, directly or indirectly, from independent claim 1. As set forth in the current Office Action, claim 1 stands rejected under Arnelid and Hyland. No separate arguments were presented for claim 3. The rejection of this dependent claim is maintained.
Applicant's arguments, regarding claims 4 and 9, filed 23 July 2025, have been fully considered but they are not persuasive. Applicant argued claims 4 and 9 based on the virtue of their dependence, directly or indirectly, from independent claim 1. As set forth in the current Office Action, claim 1 stands rejected under Arnelid and Hyland. No separate arguments were presented for claims 4 and 9. The rejection of these dependent claims is maintained.
Applicant's arguments, regarding claim 7, filed 23 July 2025, have been fully considered but they are not persuasive. Applicant argued claim 7 based on the virtue of its dependence, directly or indirectly, from independent claim 1. As set forth in the current Office Action, claim 1 stands rejected under Arnelid and Hyland. No separate arguments were presented for claim 7. The rejection of this dependent claim is maintained.
Applicant's arguments, regarding claim 8, filed 23 July 2025, have been fully considered but they are not persuasive. Applicant argued claim 8 based on the virtue of its dependence, directly or indirectly, from independent claim 1. As set forth in the current Office Action, claim 1 stands rejected under Arnelid and Hyland. No separate arguments were presented for claim 8. The rejection of this dependent claim is maintained.
Applicant's arguments, regarding claim 11, filed 23 July 2025, have been fully considered but they are not persuasive. Applicant argued claim 11 based on the virtue of its dependence, directly or indirectly, from independent claim 1. As set forth in the current Office Action, claim 1 stands rejected under Arnelid and Hyland. No separate arguments were presented for claim 11. The rejection of this dependent claim is maintained.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LOUIS C NYE whose telephone number is 571-272-0636. The examiner can normally be reached Monday - Friday 9:00AM - 5:00PM.
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, KIEU VU can be reached at 571-272-4057. 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.
/LOUIS CHRISTOPHER NYE/Examiner, Art Unit 2141
/KIEU D VU/Supervisory Patent Examiner, Art Unit 2141
Prosecution Insights