DETAILED ACTION
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 03/10/2026 has been entered.
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 .
This action is in response to amendments and remarks filed on 03/10/2026. In the current amendments, claims 1, 3-4, 10, 14, 16, and 18 are amended, claims 2, 13, and 19 are cancelled, and claims 21-23 are newly presented. Claims 1, 3-12, 14-18, and 20-23 are pending and have been examined.
In response to amendments and remarks filed on 03/10/2026, the 35 U.S.C. 112(b) rejections made in the previous office action are withdrawn.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 10, 18, and 21-23 are rejected under 35 U.S.C. 103 as being unpatentable over Chakrabarty et al. (US 2021/0402980 A1) in view of Cherian et al. (US 2022/0129666 A1).
Regarding Claim 1,
Chakrabarty et al. teaches a method for processing uncertainty in one or more variables of a robotic system, the one or more variables being provided as an input to a neural network ([0001]: "The invention relates generally to a system and method for motion control of a machine, and more particularly to a system and a method for data-driven reference generation for constraint satisfaction in machines with closed-loop control systems and uncertain dynamics or parameters" teaches a method for motion control of a machine (robotic system) with uncertain dynamics or parameters (uncertainty in variables). Fig. 1; [0041]: "FIG. 1 shows an example of a controlled machine, such as a machine, controlled by a previously designed legacy controller capable of tracking desired references 109, according to some embodiments. Together, we refer to the machine and the legacy controller as the legacy control system 101. The legacy system 101 can be a mechanical system, chemical system, or electrical system such as a motor drive, robot, automobile, or the like" teaches that the machine (shown as legacy control system 101) can be a robot (robotic system). Fig. 2; [0046]-[0047]: "The parameter estimator 231 estimates the parametric uncertainty in the machine using the output data 209 or some transformation of this data … The CAIS learning module 221 takes outputs of the parameter estimator 231, along with features and labels stored in the memory 223 to generate constraint-admissible sets" teaches that the uncertain parameters (uncertainty variables) are input to CAIS learner module (neural network). Fig. 5C; [0070]: "FIG. 5C shows various examples of bi-classification machine learning algorithms that can be used in the CAIS learner module" teaches that the CAIS learner module comprises a machine learning algorithm (neural network)), the method comprising:
executing a plurality of simulator instances to generate a plurality of input samples that collectively represent an input distribution of the one or more variables of the robotic system ([0063]: "Some embodiments simulate trajectories of the legacy system off-line, from different initial states sampled from R, reference inputs sampled from V, and parameters within Θ. At the end of each off-line simulation, if an initial condition xi ∈ R tracks a desired reference input vi ∈ V without violating the constraint (3) at any time in the simulation, for a parameter θi sampled within Θ, then the feature (xi, vi) is labeled ‘+1’ to indicate it resides within the parameter-robust constraint admissible set O(θi). Contrarily, if the constraint is violated at any time point in the simulation, the feature is labeled ‘−1’ to indicate it resides outside O(θi). This sets up a binary classification problem which can be solved via supervised machine learning" teaches simulating trajectories (executing simulator instances) of the legacy system (robotic system) by sampling estimated parameters θ within Θ (e.g. estimating parameters (variables) at each time step in the simulation will generate an input distribution of the estimated parameters). [0049]: "The uncertain parameters of the model are denoted by θ∈ Θ ⊂ Rnθ" teaches that Θ denotes the estimated set of uncertain parameters (input distribution of uncertain variables). Fig. 2; [0046]: "Estimated parameters could include a mass of a vehicle, an inertia of a vehicle, a tire friction coefficient of at least one tire mounted on a vehicle, or viscous damping coefficients in servomotors" teaches that the estimated parameter (input sample) provides a mass of the vehicle (variable of the robotic system). [0091]: "As used herein, a vehicle can mean any wheeled vehicle, including a passenger car, a bus, or a mobile robot" teaches that the vehicle is a robot (robotic system)).
Chakrabarty et al. does not appear to explicitly teach computing a Reproducing Kernel Hilbert Space (RKHS) embedding vector representing the input distribution; providing, as at least one component of the input to the neural network, the RKHS embedding vector; and processing, by the neural network, the input to the neural network to generate an output of the neural network that accounts for the input distribution represented by the plurality of input samples.
However, Cherian et al. teaches computing a Reproducing Kernel Hilbert Space (RKHS) embedding vector representing the input distribution (Fig. 5; [0076]-[0077]: "At step 504, the sequence of poses 102 is accepted as an input data, via the input interface 202 of the anomaly detector 106. The input data is indicative of a distribution of sequence of poses 102 … At step 506, the input data is embedded into a kernel space, e.g., the reproducing kernel Hilbert space (RKHS). The input data is embedded into the RKHS for learning the pair of complementary classifiers" teaches that the input data representing a distribution is embedded into the Reproducing Kernel Hilbert Space (RKHS) (e.g. computing a RKHS embedding vector representing the input distribution));
providing, as at least one component of the input to the neural network, the RKHS embedding vector (Fig. 5; [0078]-[0079]: "At step 508, embedded data is obtained based on the kernel embedding of the input data into the RKHS … At step 510, the embedded data is classified using the trained discriminative one-class classifier 206" teaches that the embedded data based on the kernel embedding of the input data into the RKHS (RKHS embedding vector) is obtained as input to the trained classifier (neural network). [0101]: "In some embodiments, the discriminative one-class classifier 206 is trained based on a classification model. The classification model may correspond to at least a bag-of-words model, a convolutional neural network (CNN) model, or a deep-learning model" teaches that the trained classifier is a neural network model); and
processing, by the neural network, the input to the neural network to generate an output of the neural network that accounts for the input distribution represented by the plurality of input samples (Fig. 5; [0078]-[0079]: "At step 508, embedded data is obtained based on the kernel embedding of the input data into the RKHS … At step 510, the embedded data is classified using the trained discriminative one-class classifier 206. In an example embodiment, the embedded data is classified based on the learned parameters of the pair of complementary classifiers" teaches that the input embedded data (RKHS embedding vector) is classified (processed) by the classifier (neural network) to generate an output based on the input distribution (distribution of the plurality of samples). [0101]: "In some embodiments, the discriminative one-class classifier 206 is trained based on a classification model. The classification model may correspond to at least a bag-of-words model, a convolutional neural network (CNN) model, or a deep-learning model" teaches that the trained classifier is a neural network model).
Chakrabarty et al. and Cherian et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate computing a Reproducing Kernel Hilbert Space (RKHS) embedding vector representing the input distribution; providing, as at least one component of the input to the neural network, the RKHS embedding vector; and processing, by the neural network, the input to the neural network to generate an output of the neural network that accounts for the input distribution represented by the plurality of input samples as taught by Cherian et al. to the disclosed invention of Chakrabarty et al.
One of ordinary skill in the art would have been motivated to make this modification because "the RKHS includes an infinite dimensional linear space that enables projection of the non-linear input data into a linear space. The infinite dimensional linear space of the RKHS allows better learning of the pair of complementary classifiers when compared against other lower dimensional spaces" (Cherian et al. [0041]).
Regarding Claim 10,
Chakrabarty et al. teaches a system configured to process uncertainty in one or more variables of a robotic system, the one or more variables being provided as an input to a neural network, the system comprising: a memory; and at least one processor coupled to the memory ([0001]: "The invention relates generally to a system and method for motion control of a machine, and more particularly to a system and a method for data-driven reference generation for constraint satisfaction in machines with closed-loop control systems and uncertain dynamics or parameters" teaches a system for motion control of a machine (robotic system) with uncertain dynamics or parameters (uncertainty in variables). [0011]: "according to some embodiments of the present disclosure, a controller is provided for operating a system under admissible states. The controller may include … a memory storing measured system states, admissible reference inputs and admissible parameter sets and computer-executable programs including a parameter estimator and an adaptive reference governor (ARG); a processor, in connection with the memory, configured to perform the ARG and the parameter estimator " teaches the system comprising a memory and a processor coupled to the memory. Fig. 1; [0041]: "FIG. 1 shows an example of a controlled machine, such as a machine, controlled by a previously designed legacy controller capable of tracking desired references 109, according to some embodiments. Together, we refer to the machine and the legacy controller as the legacy control system 101. The legacy system 101 can be a mechanical system, chemical system, or electrical system such as a motor drive, robot, automobile, or the like" teaches that the machine (shown as legacy control system 101) can be a robot (robotic system). Fig. 2; [0046]-[0047]: "The parameter estimator 231 estimates the parametric uncertainty in the machine using the output data 209 or some transformation of this data … The CAIS learning module 221 takes outputs of the parameter estimator 231, along with features and labels stored in the memory 223 to generate constraint-admissible sets" teaches that the uncertain parameters (uncertainty variables) are input to CAIS learner module (neural network). Fig. 5C; [0070]: "FIG. 5C shows various examples of bi-classification machine learning algorithms that can be used in the CAIS learner module" teaches that the CAIS learner module comprises a machine learning algorithm (neural network)) and configured to:
execute a plurality of simulator instances to generate a plurality of input samples that collectively represent an input distribution of the one or more variables of the robotic system ([0063]: "Some embodiments simulate trajectories of the legacy system off-line, from different initial states sampled from R, reference inputs sampled from V, and parameters within Θ. At the end of each off-line simulation, if an initial condition xi ∈ R tracks a desired reference input vi ∈ V without violating the constraint (3) at any time in the simulation, for a parameter θi sampled within Θ, then the feature (xi, vi) is labeled ‘+1’ to indicate it resides within the parameter-robust constraint admissible set O(θi). Contrarily, if the constraint is violated at any time point in the simulation, the feature is labeled ‘−1’ to indicate it resides outside O(θi). This sets up a binary classification problem which can be solved via supervised machine learning" teaches simulating trajectories (executing simulator instances) of the legacy system (robotic system) by sampling estimated parameters θ within Θ (e.g. estimating parameters (variables) at each time step in the simulation will generate an input distribution of the estimated parameters). [0049]: "The uncertain parameters of the model are denoted by θ∈ Θ ⊂ Rnθ" teaches that Θ denotes the estimated set of uncertain parameters (input distribution of uncertain variables). Fig. 2; [0046]: "Estimated parameters could include a mass of a vehicle, an inertia of a vehicle, a tire friction coefficient of at least one tire mounted on a vehicle, or viscous damping coefficients in servomotors" teaches that the estimated parameter (input sample) provides a mass of the vehicle (variable of the robotic system). [0091]: "As used herein, a vehicle can mean any wheeled vehicle, including a passenger car, a bus, or a mobile robot" teaches that the vehicle is a robot (robotic system)).
Chakrabarty et al. does not appear to explicitly teach compute a Reproducing Kernel Hilbert Space (RKHS) embedding vector representing the input distribution; provide, as at least one component of the input to the neural network, the RKHS embedding vector; and process, by the neural network, the input to the neural network to generate an output of the neural network that accounts for the input distribution represented by the plurality of input samples.
However, Cherian et al. teaches compute a Reproducing Kernel Hilbert Space (RKHS) embedding vector representing the input distribution (Fig. 5; [0076]-[0077]: "At step 504, the sequence of poses 102 is accepted as an input data, via the input interface 202 of the anomaly detector 106. The input data is indicative of a distribution of sequence of poses 102 … At step 506, the input data is embedded into a kernel space, e.g., the reproducing kernel Hilbert space (RKHS). The input data is embedded into the RKHS for learning the pair of complementary classifiers" teaches that the input data representing a distribution is embedded into the Reproducing Kernel Hilbert Space (RKHS) (e.g. computing a RKHS embedding vector representing the input distribution));
provide, as at least one component of the input to the neural network, the RKHS embedding vector (Fig. 5; [0078]-[0079]: "At step 508, embedded data is obtained based on the kernel embedding of the input data into the RKHS … At step 510, the embedded data is classified using the trained discriminative one-class classifier 206" teaches that the embedded data based on the kernel embedding of the input data into the RKHS (RKHS embedding vector) is obtained as input to the trained classifier (neural network). [0101]: "In some embodiments, the discriminative one-class classifier 206 is trained based on a classification model. The classification model may correspond to at least a bag-of-words model, a convolutional neural network (CNN) model, or a deep-learning model" teaches that the trained classifier is a neural network model); and
process, by the neural network, the input to the neural network to generate an output of the neural network that accounts for the input distribution represented by the plurality of input samples (Fig. 5; [0078]-[0079]: "At step 508, embedded data is obtained based on the kernel embedding of the input data into the RKHS … At step 510, the embedded data is classified using the trained discriminative one-class classifier 206. In an example embodiment, the embedded data is classified based on the learned parameters of the pair of complementary classifiers" teaches that the input embedded data (RKHS embedding vector) is classified (processed) by the classifier (neural network) to generate an output based on the input distribution (distribution of the plurality of samples). [0101]: "In some embodiments, the discriminative one-class classifier 206 is trained based on a classification model. The classification model may correspond to at least a bag-of-words model, a convolutional neural network (CNN) model, or a deep-learning model" teaches that the trained classifier is a neural network model).
Chakrabarty et al. and Cherian et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate compute a Reproducing Kernel Hilbert Space (RKHS) embedding vector representing the input distribution; provide, as at least one component of the input to the neural network, the RKHS embedding vector; and process, by the neural network, the input to the neural network to generate an output of the neural network that accounts for the input distribution represented by the plurality of input samples as taught by Cherian et al. to the disclosed invention of Chakrabarty et al.
One of ordinary skill in the art would have been motivated to make this modification because "the RKHS includes an infinite dimensional linear space that enables projection of the non-linear input data into a linear space. The infinite dimensional linear space of the RKHS allows better learning of the pair of complementary classifiers when compared against other lower dimensional spaces" (Cherian et al. [0041]).
Regarding Claim 18,
Chakrabarty et al. teaches a non-transitory computer readable medium storing instructions that, when executed by a processor, cause the processor to perform a method for processing uncertainty in one or more variables of a robotic system, the one or more variables being provided as an input to a neural network ([0001]: "The invention relates generally to a system and method for motion control of a machine, and more particularly to a system and a method for data-driven reference generation for constraint satisfaction in machines with closed-loop control systems and uncertain dynamics or parameters" teaches a method for motion control of a machine (robotic system) with uncertain dynamics or parameters (uncertainty in variables). [0016]: "another embodiment discloses a non-transitory computer readable storage medium embodied thereon a program executable by a processor for performing a method" teaches a non-transitory computer readable storage medium storing program instructions for execution by a processor for performing the method. Fig. 1; [0041]: "FIG. 1 shows an example of a controlled machine, such as a machine, controlled by a previously designed legacy controller capable of tracking desired references 109, according to some embodiments. Together, we refer to the machine and the legacy controller as the legacy control system 101. The legacy system 101 can be a mechanical system, chemical system, or electrical system such as a motor drive, robot, automobile, or the like" teaches that the machine (shown as legacy control system 101) can be a robot (robotic system). Fig. 2; [0046]-[0047]: "The parameter estimator 231 estimates the parametric uncertainty in the machine using the output data 209 or some transformation of this data … The CAIS learning module 221 takes outputs of the parameter estimator 231, along with features and labels stored in the memory 223 to generate constraint-admissible sets" teaches that the uncertain parameters (uncertainty variables) are input to CAIS learner module (neural network). Fig. 5C; [0070]: "FIG. 5C shows various examples of bi-classification machine learning algorithms that can be used in the CAIS learner module" teaches that the CAIS learner module comprises a machine learning algorithm (neural network)), the method comprising:
executing a plurality of simulator instances to generate a plurality of input samples that collectively represent an input distribution of the one or more variables of the robotic system ([0063]: "Some embodiments simulate trajectories of the legacy system off-line, from different initial states sampled from R, reference inputs sampled from V, and parameters within Θ. At the end of each off-line simulation, if an initial condition xi ∈ R tracks a desired reference input vi ∈ V without violating the constraint (3) at any time in the simulation, for a parameter θi sampled within Θ, then the feature (xi, vi) is labeled ‘+1’ to indicate it resides within the parameter-robust constraint admissible set O(θi). Contrarily, if the constraint is violated at any time point in the simulation, the feature is labeled ‘−1’ to indicate it resides outside O(θi). This sets up a binary classification problem which can be solved via supervised machine learning" teaches simulating trajectories (executing simulator instances) of the legacy system (robotic system) by sampling estimated parameters θ within Θ (e.g. estimating parameters (variables) at each time step in the simulation will generate an input distribution of the estimated parameters). [0049]: "The uncertain parameters of the model are denoted by θ∈ Θ ⊂ Rnθ" teaches that Θ denotes the estimated set of uncertain parameters (input distribution of uncertain variables). Fig. 2; [0046]: "Estimated parameters could include a mass of a vehicle, an inertia of a vehicle, a tire friction coefficient of at least one tire mounted on a vehicle, or viscous damping coefficients in servomotors" teaches that the estimated parameter (input sample) provides a mass of the vehicle (variable of the robotic system). [0091]: "As used herein, a vehicle can mean any wheeled vehicle, including a passenger car, a bus, or a mobile robot" teaches that the vehicle is a robot (robotic system)).
Chakrabarty et al. does not appear to explicitly teach computing a Reproducing Kernel Hilbert Space (RKHS) embedding vector representing the input distribution; providing, as at least one component of the input to the neural network, the RKHS embedding vector; and processing, by the neural network, the input to the neural network to generate an output of the neural network that accounts for the input distribution represented by the plurality of input samples.
However, Cherian et al. teaches computing a Reproducing Kernel Hilbert Space (RKHS) embedding vector representing the input distribution (Fig. 5; [0076]-[0077]: "At step 504, the sequence of poses 102 is accepted as an input data, via the input interface 202 of the anomaly detector 106. The input data is indicative of a distribution of sequence of poses 102 … At step 506, the input data is embedded into a kernel space, e.g., the reproducing kernel Hilbert space (RKHS). The input data is embedded into the RKHS for learning the pair of complementary classifiers" teaches that the input data representing a distribution is embedded into the Reproducing Kernel Hilbert Space (RKHS) (e.g. computing a RKHS embedding vector representing the input distribution));
providing, as at least one component of the input to the neural network, the RKHS embedding vector (Fig. 5; [0078]-[0079]: "At step 508, embedded data is obtained based on the kernel embedding of the input data into the RKHS … At step 510, the embedded data is classified using the trained discriminative one-class classifier 206" teaches that the embedded data based on the kernel embedding of the input data into the RKHS (RKHS embedding vector) is obtained as input to the trained classifier (neural network). [0101]: "In some embodiments, the discriminative one-class classifier 206 is trained based on a classification model. The classification model may correspond to at least a bag-of-words model, a convolutional neural network (CNN) model, or a deep-learning model" teaches that the trained classifier is a neural network model); and
processing, by the neural network, the input to the neural network to generate an output of the neural network that accounts for the input distribution represented by the plurality of input samples (Fig. 5; [0078]-[0079]: "At step 508, embedded data is obtained based on the kernel embedding of the input data into the RKHS … At step 510, the embedded data is classified using the trained discriminative one-class classifier 206. In an example embodiment, the embedded data is classified based on the learned parameters of the pair of complementary classifiers" teaches that the input embedded data (RKHS embedding vector) is classified (processed) by the classifier (neural network) to generate an output based on the input distribution (distribution of the plurality of samples). [0101]: "In some embodiments, the discriminative one-class classifier 206 is trained based on a classification model. The classification model may correspond to at least a bag-of-words model, a convolutional neural network (CNN) model, or a deep-learning model" teaches that the trained classifier is a neural network model).
Chakrabarty et al. and Cherian et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate computing a Reproducing Kernel Hilbert Space (RKHS) embedding vector representing the input distribution; providing, as at least one component of the input to the neural network, the RKHS embedding vector; and processing, by the neural network, the input to the neural network to generate an output of the neural network that accounts for the input distribution represented by the plurality of input samples as taught by Cherian et al. to the disclosed invention of Chakrabarty et al.
One of ordinary skill in the art would have been motivated to make this modification because "the RKHS includes an infinite dimensional linear space that enables projection of the non-linear input data into a linear space. The infinite dimensional linear space of the RKHS allows better learning of the pair of complementary classifiers when compared against other lower dimensional spaces" (Cherian et al. [0041]).
Regarding Claim 21,
Chakrabarty et al. in view of Cherian et al. teaches the method of claim 1.
In addition, Chakrabarty et al. further teaches wherein each of the plurality of input samples provides a discrete value for one of the one or more variables of the robotic system (Fig. 2; [0046]: "Estimated parameters could include a mass of a vehicle, an inertia of a vehicle, a tire friction coefficient of at least one tire mounted on a vehicle, or viscous damping coefficients in servomotors" teaches that the estimated parameter (input sample) provides a mass (discrete value) of the vehicle (robotic system). [0091]: "As used herein, a vehicle can mean any wheeled vehicle, including a passenger car, a bus, or a mobile robot" teaches that the vehicle is a robot (robotic system)).
Regarding Claim 22,
Chakrabarty et al. in view of Cherian et al. teaches the method of claim 21.
In addition, Chakrabarty et al. further teaches wherein the discrete value represents a quantity of a physical feature corresponding to the robotic system (Fig. 2; [0046]: "Estimated parameters could include a mass of a vehicle, an inertia of a vehicle, a tire friction coefficient of at least one tire mounted on a vehicle, or viscous damping coefficients in servomotors" teaches that the estimated parameter (input sample) provides a mass (discrete value representing a quantity of a physical feature) of the vehicle (robotic system). [0091]: "As used herein, a vehicle can mean any wheeled vehicle, including a passenger car, a bus, or a mobile robot" teaches that the vehicle is a robot (robotic system)).
Regarding Claim 23,
Chakrabarty et al. in view of Cherian et al. teaches the method of claim 22.
In addition, Chakrabarty et al. further teaches wherein the physical feature is a mass of an object corresponding to the robotic system (Fig. 2; [0046]: "Estimated parameters could include a mass of a vehicle, an inertia of a vehicle, a tire friction coefficient of at least one tire mounted on a vehicle, or viscous damping coefficients in servomotors" teaches that the estimated parameter (input sample) provides a mass (physical feature discrete value) of the vehicle (object corresponding to the robotic system). [0091]: "As used herein, a vehicle can mean any wheeled vehicle, including a passenger car, a bus, or a mobile robot" teaches that the vehicle is a robot (robotic system)).
Claims 3-4 and 14-15 are rejected under 35 U.S.C. 103 as being unpatentable over Chakrabarty et al. (US 2021/0402980 A1) in view of Cherian et al. (US 2022/0129666 A1) and further in view of Whitbrock et al. (US 2018/0330201 A1).
Regarding Claim 3,
Chakrabarty et al. in view of Cherian et al. teaches the method of claim 1.
Chakrabarty et al. in view of Cherian et al. does not appear to explicitly teach wherein the plurality of simulator instances are executed in parallel on a parallel processing unit.
However, Whitbrock et al. teaches wherein the plurality of simulator instances are executed in parallel on a parallel processing unit (Fig. 4; [0073]: "in some embodiments of the invention, the machine learning component 410 performs parallel computing associated with two or more processors that process one or more portions of the time-series data in parallel" teaches parallel processors (parallel processing unit) for processing portions of time-series data in parallel).
Chakrabarty et al., Cherian et al., and Whitbrock et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the plurality of simulator instances are executed in parallel on a parallel processing unit as taught by Whitbrock et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "reduction in both computation and memory storage [that] allows for more efficient training and testing when combining with empirical risk minimization (ERM) classifiers such as SVM" (Whitbrock et al. [0063]).
Regarding Claim 4,
Chakrabarty et al. in view of Cherian et al. teaches the method of claim 1.
Chakrabarty et al. in view of Cherian et al. does not appear to explicitly teach wherein the plurality of simulator instances generate the plurality of input samples for a discrete time step of a simulation.
However, Whitbrock et al. teaches wherein the plurality of simulator instances generate the plurality of input samples for a discrete time step of a simulation ([0066]: "In general, time-series data component 402 is configured to receive time-series data 416 and output a machine learning output 418. Time-series data 416 includes streams or sequences of time-series data. … In some embodiments of the invention, time-series data 416 include two or more streams of time-series data with equal time spans. As used herein, “time-series data” can be one or more sequences of data that is repeatedly generated and/or captured at a plurality of time values during a certain time interval. In some embodiments of the invention, time-series data 416 is raw time-series data (e.g., unprocessed time-series data)" teaches that the time-series data is input data for generating an output using the machine learning model (i.e. input of neural network) and includes two or more streams of data (plurality of samples) over a time span. Fig. 4; [0067]-[0068]: "The distribution generation component 404 is configured to generate one or more probability distributions for use by system 400. In some embodiments of the invention, distribution generation component 404 generates metadata pertaining to the time-series data 416. For example, in some embodiments of the invention distribution generation component 404 generates a probability distribution of the time-series data 416. In some embodiments of the invention, rather than generating a probability distribution from the time-series data 416, the distribution generation component generates or selects a random distribution such as a Gaussian distribution … In particular, reference time-series generation component 406 is configured to generate a set of reference time-series, in which each length is uniformly sampled from a predetermined minimum length to a predetermined maximum length to capture the optimal alignment of time-series data 416. The corresponding values of each reference time-series in the set are drawn from the distribution provided by distribution generation component 404" teaches obtaining a distribution representing data from the time-series data, wherein sets of reference time-series data are generated by sampling data (simulator instances) over a uniform time period (discrete time steps) (i.e. the two or more streams of raw time-series data are sampled to generate reference time series data)).
Chakrabarty et al., Cherian et al., and Whitbrock et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the plurality of simulator instances generate the plurality of input samples for a discrete time step of a simulation as taught by Whitbrock et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "reduction in both computation and memory storage [that] allows for more efficient training and testing when combining with empirical risk minimization (ERM) classifiers such as SVM" (Whitbrock et al. [0063]).
Regarding Claim 14,
Chakrabarty et al. in view of Cherian et al. teaches the system of claim 10.
Chakrabarty et al. in view of Cherian et al. does not appear to explicitly teach wherein the at least one processor comprises: a host processor; and a parallel processing unit configured to execute the plurality of simulator instances in a plurality of threads executing in parallel.
However, Whitbrock et al. teaches wherein the at least one processor comprises: a host processor (Fig. 4; [0065]: "Furthermore, time-series data component 402 in some embodiments of the invention includes a processor 414 to facilitate execution of the instructions (e.g., computer executable components and corresponding instructions) by time-series data component 402" teaches a processor 414 (host processor) for executing instructions); and
a parallel processing unit configured to execute the plurality of simulator instances in a plurality of threads executing in parallel (Fig. 4; [0073]: "in some embodiments of the invention, the machine learning component 410 performs parallel computing associated with two or more processors that process one or more portions of the time-series data in parallel" teaches parallel processors (parallel processing unit) for processing portions of time-series data (threads) in parallel).
Chakrabarty et al., Cherian et al., and Whitbrock et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the at least one processor comprises: a host processor; and a parallel processing unit configured to execute the plurality of simulator instances in a plurality of threads executing in parallel as taught by Whitbrock et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "reduction in both computation and memory storage [that] allows for more efficient training and testing when combining with empirical risk minimization (ERM) classifiers such as SVM" (Whitbrock et al. [0063]).
Regarding Claim 15,
Chakrabarty et al. in view of Cherian et al. and further in view of Whitbrock et al. teaches the system of claim 10.
In addition, Whitbrock et al. further teaches wherein the parallel processing unit is also configured to implement the neural network (Fig. 4; [0073]: "In some embodiments of the invention, the feature matrix generated by feature matrix generation component 408 is provided as an input for a machine learning model executed by the machine learning component 410. In certain embodiments of the invention, the machine learning component 410 employs parallel computing to process portions of the feature matrix with portions of the time-series data 416" teaches parallel processors (parallel processing unit) implement the machine learning model (neural network)).
Chakrabarty et al., Cherian et al., and Whitbrock et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the parallel processing unit is also configured to implement the neural network as taught by Whitbrock et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification to provide "reduction in both computation and memory storage [that] allows for more efficient training and testing when combining with empirical risk minimization (ERM) classifiers such as SVM" (Whitbrock et al. [0063]).
Claims 5-9, 11-12, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Chakrabarty et al. (US 2021/0402980 A1) in view of Cherian et al. (US 2022/0129666 A1) and further in view of Bouboulis et al. ("Efficient KLMS and KRLS algorithms: A random Fourier feature perspective")
Regarding Claim 5,
Chakrabarty et al. in view of Cherian et al. teaches the method of claim 1.
Chakrabarty et al. in view of Cherian et al. does not appear to explicitly teach wherein the computing the RKHS embedding vector is performed using a plurality of Random Fourier Features (RFFs), a dimension K of the RKHS embedding vector being equal to a number of the plurality of RFFs.
However, Bouboulis et al. teaches wherein the computing the RKHS embedding vector is performed using a plurality of Random Fourier Features (RFFs), a dimension K of the RKHS embedding vector being equal to a number of the plurality of RFFs (Section 1, third paragraph: "Instead of mapping the input data to an infinite dimensional Reproducing Kernel Hilbert Space, induced by the selected kernel, and subsequently sparsifying the solution, we map the input data to a finite (although larger than the input one) dimensional Euclidian space RD. However, this mapping is done in a sensible way that cares for a good approximation of the kernel evaluations. The mapping to RD is carried out using random features of the kernel’s Fourier transform. Following this approach, the resulting algorithm, which we call Random Fourier Features Kernel LMS or RFFKLMS for short, leads naturally to a standard linear LMS, with a fixed-size solution (i.e., a vector in RD)" teaches that mapping to the Reproducing Kernel Hilbert Space (RKHS embedding vector) is performed using random Fourier Features, with the dimension D (K) of the embedding being a fixed size based on the number of RFFs).
Chakrabarty et al., Cherian et al., and Bouboulis et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the computing the RKHS embedding vector is performed using a plurality of Random Fourier Features (RFFs), a dimension K of the RKHS embedding vector being equal to a number of the plurality of RFFs as taught by Bouboulis et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification because "the obtained algorithms are computationally significantly more efficient compared to previously derived variants, while, at the same time, they converge at similar speeds and to similar error floors" (Bouboulis et al., Abstract first paragraph).
Regarding Claim 6,
Chakrabarty et al. in view of Cherian et al. and further in view of Bouboulis et al. teaches the method of claim 5.
In addition, Bouboulis et al. further teaches wherein each RFF of the plurality of RFFs is implemented as a sum of cosine terms normalized by a number N of the plurality of input samples, wherein each cosine term takes the form: cos(ωixj + bi), where xj is a sample value of the plurality of input samples, j=1, ..., N is an index associated with the number N of the plurality of input samples, ωi is a randomly generated frequency component, bi is a randomly generated bias component, and i=1, ..., K is an index associated with the dimension K of the RKHS embedding vector (Abstract, first paragraph: "Instead of implicitly mapping the data to a RKHS (e.g., kernel trick), we map the data to a finite dimensional Euclidean space, using random features of the kernel’s Fourier transform" teaches that RFFs are used for RKHS. Section 3, third paragraph: "defining zω,b(x) = √2 cos(ωTx + b), it turns out that ... Following Theorem 1, we choose to approximate κ(xn − xm) using D random Fourier features, ω1, ω2, ..., ωD, (drawn from p) and D random numbers, b1, b2, ...,bD (drawn uniformly from [0, 2π]) that define a sample average (a similar rationale as the one used in Monte Carlo Methods; for Gaussian kernels such sampling is trivial):
PNG
media_image1.png
50
334
media_image1.png
Greyscale
"
teaches that each RFF is generated based on a random frequency ω and random bias b, and summing the result of a cosine function (cos(ωix + bi)) applied to D (N) samples, with i being an index).
Chakrabarty et al., Cherian et al., and Bouboulis et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein each RFF of the plurality of RFFs is implemented as a sum of cosine terms normalized by a number N of the plurality of input samples, wherein each cosine term takes the form: cos(ωixj + bi), where xj is a sample value of the plurality of input samples, j=1, ..., N is an index associated with the number N of the plurality of input samples, ωi is a randomly generated frequency component, bi is a randomly generated bias component, and i=1, ..., K is an index associated with the dimension K of the RKHS embedding vector as taught by Bouboulis et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification because "the obtained algorithms are computationally significantly more efficient compared to previously derived variants, while, at the same time, they converge at similar speeds and to similar error floors" (Bouboulis et al., Abstract first paragraph).
Regarding Claim 7,
Chakrabarty et al. in view of Cherian et al. and further in view of Bouboulis et al. teaches the method of claim 5.
In addition, Bouboulis et al. further teaches wherein the dimension K of the RKHS embedding vector is preset (Section 1, third paragraph: "Instead of mapping the input data to an infinite dimensional Reproducing Kernel Hilbert Space, induced by the selected kernel, and subsequently sparsifying the solution, we map the input data to a finite (although larger than the input one) dimensional Euclidian space RD. However, this mapping is done in a sensible way that cares for a good approximation of the kernel evaluations. The mapping to RD is carried out using random features of the kernel’s Fourier transform. Following this approach, the resulting algorithm, which we call Random Fourier Features Kernel LMS or RFFKLMS for short, leads naturally to a standard linear LMS, with a fixed-size solution (i.e., a vector in RD)" teaches that the dimension D (K) of the embedding is a finite fixed-sized preset).
Chakrabarty et al., Cherian et al., and Bouboulis et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the dimension K of the RKHS embedding vector is preset as taught by Bouboulis et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification because "the obtained algorithms are computationally significantly more efficient compared to previously derived variants, while, at the same time, they converge at similar speeds and to similar error floors" (Bouboulis et al., Abstract first paragraph).
Regarding Claim 8,
Chakrabarty et al. in view of Cherian et al. and further in view of Bouboulis et al. teaches the method of claim 5.
In addition, Bouboulis et al. further teaches wherein the dimension K of the RKHS embedding vector is adjusted dynamically in accordance with an optimization algorithm (Section 5, first paragraph: "In all experimental setups, we tuned (using multiple trials) so that it takes a value close to this “optimal”, so that to take the best possible MSE at the smallest time. On the other hand, the performance of RFFKLMS depends largely on D, which controls the quality of the kernel approximation. Similar to the case of QKLMS, there is a value for D so that RFFKLMS attains its lowest steady-state MSE" teaches that the dimension D (K) associated with the RFFs of the Reproducing Kernel Hilbert Space is adjusted based on a mean square error MSE algorithm (optimization algorithm)).
Chakrabarty et al., Cherian et al., and Bouboulis et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the dimension K of the RKHS embedding vector is adjusted dynamically in accordance with an optimization algorithm as taught by Bouboulis et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification because "the obtained algorithms are computationally significantly more efficient compared to previously derived variants, while, at the same time, they converge at similar speeds and to similar error floors" (Bouboulis et al., Abstract first paragraph).
Regarding Claim 9,
Chakrabarty et al. in view of Cherian et al. and further in view of Bouboulis et al. teaches the method of claim 8.
In addition, Bouboulis et al. further teaches wherein the optimization algorithm maximizes the dimension K in accordance with a time constraint associated with processing the RKHS embedding vector by the neural network, wherein maximizing the dimension K in accordance with a time constraint comprises, during one or more of a plurality of time steps: incrementing K based on a time to complete processing of the RKHS embedding vector by the neural network being less than or equal to a target, or decrementing K based on a time to complete processing of the RKHS embedding vector by the neural network being greater than the target (Section 5, first paragraph: "In all experimental setups, we tuned (using multiple trials) so that it takes a value close to this “optimal”, so that to take the best possible MSE at the smallest time. On the other hand, the performance of RFFKLMS depends largely on D, which controls the quality of the kernel approximation. Similar to the case of QKLMS, there is a value for D so that RFFKLMS attains its lowest steady-state MSE" teaches that the dimension D (K) is optimized (maximized) by using the best possible MSE (optimization algorithm output) at the smallest time (time constraint), meaning that the dimension D (K) is increased when less than the time constraint and decreased when greater than the time constraint).
Chakrabarty et al., Cherian et al., and Bouboulis et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the optimization algorithm maximizes the dimension K in accordance with a time constraint associated with processing the RKHS embedding vector by the neural network, wherein maximizing the dimension K in accordance with a time constraint comprises, during one or more of a plurality of time steps: incrementing K based on a time to complete processing of the RKHS embedding vector by the neural network being less than or equal to a target, or decrementing K based on a time to complete processing of the RKHS embedding vector by the neural network being greater than the target as taught by Bouboulis et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification because "the obtained algorithms are computationally significantly more efficient compared to previously derived variants, while, at the same time, they converge at similar speeds and to similar error floors" (Bouboulis et al., Abstract first paragraph).
Regarding Claim 11,
Chakrabarty et al. in view of Cherian et al. teaches the system of claim 10.
Chakrabarty et al. in view of Cherian et al. does not appear to explicitly teach wherein the memory is configured to store parameters for a reproducing kernel Hilbert space (RKHS) module, the parameters including a plurality of frequency components ωi and a plurality of corresponding bias components bi for a plurality of Random Fourier Features (RFFs).
However, Bouboulis et al. teaches wherein the memory is configured to store parameters for a reproducing kernel Hilbert space (RKHS) module, the parameters including a plurality of frequency components ωi and a plurality of corresponding bias components bi for a plurality of Random Fourier Features (RFFs) (Abstract, first paragraph: "Instead of implicitly mapping the data to a RKHS (e.g., kernel trick), we map the data to a finite dimensional Euclidean space, using random features of the kernel’s Fourier transform" teaches that RFFs are used for RKHS. Section 3, third paragraph: "defining zω,b(x) = √2 cos(ωTx + b), it turns out that ... Following Theorem 1, we choose to approximate κ(xn − xm) using D random Fourier features, ω1, ω2, ..., ωD, (drawn from p) and D random numbers, b1, b2, ...,bD (drawn uniformly from [0, 2π]) that define a sample average (a similar rationale as the one used in Monte Carlo Methods; for Gaussian kernels such sampling is trivial):
PNG
media_image1.png
50
334
media_image1.png
Greyscale
"
teaches that each RFF is generated based on random frequency ω and random bias b parameters).
Chakrabarty et al., Cherian et al., and Bouboulis et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the memory is configured to store parameters for a reproducing kernel Hilbert space (RKHS) module, the parameters including a plurality of frequency components ωi and a plurality of corresponding bias components bi for a plurality of Random Fourier Features (RFFs) as taught by Bouboulis et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification because "the obtained algorithms are computationally significantly more efficient compared to previously derived variants, while, at the same time, they converge at similar speeds and to similar error floors" (Bouboulis et al., Abstract first paragraph).
Regarding Claim 12,
Chakrabarty et al. in view of Cherian et al. and further in view of Bouboulis et al. teaches the system of claim 11.
In addition, Bouboulis et al. further teaches wherein the RKHS embedding vector is generated by applying the plurality of RFFs to each of the plurality of input samples (Section 3, third paragraph: " defining zω,b(x) = √2 cos(ωTx + b), it turns out that ... Following Theorem 1, we choose to approximate κ(xn − xm) using D random Fourier features, ω1, ω2, ..., ωD, (drawn from p) and D random numbers, b1, b2, ...,bD (drawn uniformly from [0, 2π]) that define a sample average (a similar rationale as the one used in Monte Carlo Methods; for Gaussian kernels such sampling is trivial):
PNG
media_image1.png
50
334
media_image1.png
Greyscale
"
teaches that the embedding is generated based on D (N) samples applied to the RFFs).
Chakrabarty et al., Cherian et al., and Bouboulis et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the RKHS embedding vector is generated by applying the plurality of RFFs to each of the plurality of input samples as taught by Bouboulis et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification because "the obtained algorithms are computationally significantly more efficient compared to previously derived variants, while, at the same time, they converge at similar speeds and to similar error floors" (Bouboulis et al., Abstract first paragraph).
Regarding Claim 20,
Chakrabarty et al. in view of Cherian et al. teaches the non-transitory computer readable medium of claim 18.
Chakrabarty et al. in view of Cherian et al. does not appear to explicitly teach wherein the computing the RKHS embedding vector is performed using a plurality of Random Fourier Features (RFFs), each RFF of the plurality of RFFs being implemented as a sum of cosine terms normalized by a number N of the plurality of input samples, wherein each cosine term takes the form: cos(ωixj + bi), where xj is a sample value of the plurality of input samples, j=1, ..., N is an index associated with the number N of the plurality of input samples, ωi is a randomly generated frequency component, bi is a randomly generated bias component, and i=1, ..., K is an index associated with the dimension K of the RKHS embedding vector.
However, Bouboulis et al. teaches wherein the computing the RKHS embedding vector is performed using a plurality of Random Fourier Features (RFFs), each RFF of the plurality of RFFs being implemented as a sum of cosine terms normalized by a number N of the plurality of input samples, wherein each cosine term takes the form: cos(ωixj + bi), where xj is a sample value of the plurality of input samples, j=1, ..., N is an index associated with the number N of the plurality of input samples, ωi is a randomly generated frequency component, bi is a randomly generated bias component, and i=1, ..., K is an index associated with the dimension K of the RKHS embedding vector (Abstract, first paragraph: "Instead of implicitly mapping the data to a RKHS (e.g., kernel trick), we map the data to a finite dimensional Euclidean space, using random features of the kernel’s Fourier transform" teaches that RFFs are used for RKHS. Section 3, third paragraph: "defining zω,b(x) = √2 cos(ωTx + b), it turns out that ... Following Theorem 1, we choose to approximate κ(xn − xm) using D random Fourier features, ω1, ω2, ..., ωD, (drawn from p) and D random numbers, b1, b2, ...,bD (drawn uniformly from [0, 2π]) that define a sample average (a similar rationale as the one used in Monte Carlo Methods; for Gaussian kernels such sampling is trivial):
PNG
media_image1.png
50
334
media_image1.png
Greyscale
"
teaches that each RFF is generated based on a random frequency ω and random bias b, and summing the result of a cosine function (cos(ωix + bi)) applied to D (N) samples, with i being an index).
Chakrabarty et al., Cherian et al., and Bouboulis et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate wherein the computing the RKHS embedding vector is performed using a plurality of Random Fourier Features (RFFs), each RFF of the plurality of RFFs being implemented as a sum of cosine terms normalized by a number N of the plurality of input samples, wherein each cosine term takes the form: cos(ωixj + bi), where xj is a sample value of the plurality of input samples, j=1, ..., N is an index associated with the number N of the plurality of input samples, ωi is a randomly generated frequency component, bi is a randomly generated bias component, and i=1, ..., K is an index associated with the dimension K of the RKHS embedding vector as taught by Bouboulis et al. to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification because "the obtained algorithms are computationally significantly more efficient compared to previously derived variants, while, at the same time, they converge at similar speeds and to similar error floors" (Bouboulis et al., Abstract first paragraph).
Claim 16 is rejected under 35 U.S.C. 103 as being unpatentable over Chakrabarty et al. (US 2021/0402980 A1) in view of Cherian et al. (US 2022/0129666 A1) and further in view of Sindhwani (US 20200189099 A1).
Regarding Claim 16,
Chakrabarty et al. in view of Cherian et al. teaches the system of claim 10.
Chakrabarty et al. in view of Cherian et al. does not appear to teach further comprising the robotic system, wherein at least one control signal for the robotic system is generated based on the output of the neural network.
However, Sindhwani teaches further comprising the robotic system, wherein at least one control signal for the robotic system is generated based on the output of the neural network (Fig. 1; [0079]: "The control policy system 120 includes a data engine 122 and a learning engine 124. In some implementations, more or fewer engines may be provided. In some implementations, the data engine 122 samples a distributed group of data points and provides them to learning engine 124 for use in generating a control policy, e.g., by learning a contracting vector field based on a plurality of kernels as described herein. In some implementations, the data engine 122 additionally or alternatively automatically generates a potential gradient for a group of data points, assigns the potential gradient to the data points of the group, and provides the assigned potential gradient to learning engine 124 for use in generating a control policy" teaches that the control policy (control signal) for the robotic system is generated based on the output of the learning engine (neural network)).
Chakrabarty et al., Cherian et al., and Sindhwani are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate further comprising the robotic system, wherein at least one control signal for the robotic system is generated based on the output of the neural network as taught by Sindhwani to the disclosed invention of Chakrabarty et al. in view of Cherian et al.
One of ordinary skill in the art would have been motivated to make this modification to "significantly improve training time required to generate robot control policies, which also conserves computing resources" (Sindhwani [0008]).
Claim 17 is rejected under 35 U.S.C. 103 as being unpatentable over Chakrabarty et al. (US 2021/0402980 A1) in view of Cherian et al. (US 2022/0129666 A1) and further in view of Dutordoir et al. (US 2021/0300390 A1)
Regarding Claim 17,
Chakrabarty et al. in view of Cherian et al. teaches the system of claim 10.
Chakrabarty et al. in view of Cherian et al. does not appear to teach further comprising a vehicle including one or more sensors, and wherein the plurality of input samples collectively representing the input distribution are generated by the one or more sensors.
However, Dutordoir et al. teaches further comprising a vehicle including one or more sensors, and wherein the plurality of input samples collectively representing the input distribution are generated by the one or more sensors (Fig. 1; [0016]: "a system including: one or more sensors configured to measure one or more characteristics of a physical system having a plurality of enforceable parameters; processing circuitry; and memory circuitry. The memory circuitry stores machine-readable instructions which, when executed by the processing circuitry, cause the system to determine a plurality of candidate sets of values for the enforceable parameters, and for each of the candidate sets of values for the plurality of enforceable parameters: obtain, from the one or more sensors, a measurement of each of the one or more characteristics of the physical system; determine a performance measurement for the physical system based on the obtained measurements of the one or more characteristics of the physical system; and generate a data point having an input portion indicative of the candidate set of values and an output portion indicative of the determined performance measurement, the input portion having a first number of dimensions. The instructions further cause the system to augment each data point to: include an additional dimension comprising a bias value; project each augmented data point onto a surface of a unit hypersphere of the first number of dimensions; determine, using the projected augmented data points, a set of parameter values for a sparse variational GP on said unit hypersphere, the sparse variational GP having a zonal kernel and depending on a set of inducing variables randomly distributed according to a multi-dimensional Gaussian distribution and each corresponding to a reproducing kernel Hilbert space inner product between the GP and a spherical harmonic of the first number of dimensions; and determine, using the sparse variational GP with the determined set of parameter values, a further set of values for the plurality of enforceable parameters" teaches that sensors generate characteristics measurements (input feature distributions) for the physical system. [0029]: "In the present example, the physical system 102 includes components of a vehicle. Specifically, the system includes aerodynamic components of the vehicle for which various geometrical parameters can be modified. In other examples, a physical system could be a specific part of a vehicle such as an engine or a braking system, or could be an entire vehicle" teaches that the physical system is a vehicle).
Chakrabarty et al., Cherian et al., and Dutordoir et al. are analogous to the claimed invention because they are directed to the use of Reproducing Kernel Hilbert Space (RKHS) for implementing a neural network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate further comprising a vehicle including one or more sensors, and wherein the plurality of input samples collectively representing the input distribution are generated by the one or more sensors as taught by Dutordoir et al. to the disclosed invention of Chakrabarty et al. in view of Song et al.
One of ordinary skill in the art would have been motivated to make this modification so that "accuracy and efficiency can be retained even for data with a large number of input dimensions" (Dutordoir et al. [0019]).
Response to Arguments
Applicant’s arguments, filed 03/10/2026, with respect to the prior art rejections of claims 1, 3-12, 14-18, and 20-23 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BRIAN J HALES whose telephone number is (571)272-0878. The examiner can normally be reached M-F 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, Kamran Afshar can be reached at (571) 272-7796. 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.
/BRIAN J HALES/Examiner, Art Unit 2125
/KAMRAN AFSHAR/Supervisory Patent Examiner, Art Unit 2125
Prosecution Insights