DETAILED ACTION
This action is in response to the application filed 05/05/2026. Claims 1-4 and 6-19 are pending and have been examined.
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 .
Information Disclosure Statement
The information disclosure statement filed 05/19/2023 fails to comply with 37 CFR 1.98(a)(2), which requires a legible copy of each cited foreign patent document; each non-patent literature publication or that portion which caused it to be listed; and all other information or that portion which caused it to be listed. It has been placed in the application file, but the information referred to therein has not been considered.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-4 and 6-19 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1:
Subject Matter Eligibility Analysis Step 1:
Claim 1 recites an artificial intelligence (AI) system and is thus an apparatus, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 1 recites
encode each input data point of the time series input data from the input state space into a latent space to produce latent data points indexed in time according to time indices of corresponding input data points (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally encoding data.)
and propagate the latent data points backward in time with a neural Ordinary Differential Equation (ODE) approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space; (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally propagating data points backward in time with an equation.)
propagate the initial point of latent dynamics of the device forward in time until a time index of interest using the neural ODE to produce a state of latent dynamics of the device at the time index of interest; (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally propagating a data point forward in time with an equation.)
decode the state of latent dynamics of the device into the output state space different from the input state space to produce output data including the state of the device at the time index of interest (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally decoding the state of latent dynamics of the device.)
Therefore, claim 1 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 1 further recites additional elements of
An Artificial Intelligence (AI) system (This element does not integrate the abstract idea into a practical application because it recites generic computing components on which to perform the abstract idea (see MPEP 2106.05(f)).)
for sensing a state of a device with continuous-time dynamics, (This element does not integrate the abstract idea into a practical application because it recites a technological environment in which to apply a judicial exception (see MPEP 2106.05(h)).)
the AI system including a neural network having an autoencoder architecture adapted for dynamic transformation of time series input data from an input state space indicative of the state of the device into an output state space indicative of the state of the device, comprising: (This element does not integrate the abstract idea into a practical application because it recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g)).)
at least one processor; and a memory having instructions stored thereon that cause the at least one processor to execute the neural network, train the neural network, or both, the autoencoder architecture configured to: (This element does not integrate the abstract idea into a practical application because it recites generic computing components on which to perform the abstract idea (see MPEP 2106.05(f)).)
wherein the input state space is a signal space, and wherein the output state space is a location space (This element does not integrate the abstract idea into a practical application because it recites a technological environment in which to apply a judicial exception (see MPEP 2106.05(h)).)
Therefore, claim 1 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 1 do not provide significantly more than the abstract idea itself, taken alone and in combination because
An Artificial Intelligence (AI) system uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
for sensing a state of a device with continuous-time dynamics specifies a particular technological environment to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(h)).
the AI system including a neural network having an autoencoder architecture adapted for dynamic transformation of time series input data from an input state space indicative of the state of the device into an output state space indicative of the state of the device, comprising is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)).
at least one processor; and a memory having instructions stored thereon that cause the at least one processor to execute the neural network, train the neural network, or both, the autoencoder architecture configured to uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
wherein the input state space is a signal space, and wherein the output state space is a location space specifies a particular technological environment to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(h)).
Therefore, claim 1 is subject-matter ineligible.
Regarding Claim 2:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 2 recites
decode the state of latent dynamics of the device into the output state space same as the input state space to reconstruct the time series input data. (This limitation is a mental process as it encompasses a human mentally decoding the state of latent dynamics.)
Therefore, claim 2 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 2 further recites additional elements of
wherein the autoencoder architecture is further configured to decode (This element does not integrate the abstract idea into a practical application because it amounts to mere “apply it on a computer” (see MPEP 2106.05(f)).)
Therefore, claim 2 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 2 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein the autoencoder architecture is further configured to decode uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
Therefore, claim 2 is subject-matter ineligible.
Regarding Claim 3:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 3 recites
interpolate the trajectory of the device based on the state of latent dynamics of the device (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally interpolating the trajectory.)
Therefore, claim 3 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 3 further recites additional elements of
wherein the state of the device corresponds to a trajectory of the device, (This element does not integrate the abstract idea into a practical application because it recites a technological environment in which to apply a judicial exception (see MPEP 2106.05(h)).)
and wherein the autoencoder architecture is further configured to interpolate (This element does not integrate the abstract idea into a practical application because it amounts to mere “apply it on a computer” (see MPEP 2106.05(f)).)
Therefore, claim 3 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 3 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein the state of the device corresponds to a trajectory of the device, specifies a particular technological environment to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(h)).
and wherein the autoencoder architecture is further configured to interpolate uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
Therefore, claim 3 is subject-matter ineligible.
Regarding Claim 4:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 4 recites
extrapolate the trajectory of the device based on the state of latent dynamics of the device (This limitation is a mental process as it encompasses a human mentally extrapolating the trajectory.)
Therefore, claim 4 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 4 further recites additional elements of
wherein the autoencoder architecture is further configured to extrapolate (This element does not integrate the abstract idea into a practical application because it amounts to mere “apply it on a computer” (see MPEP 2106.05(f)).)
Therefore, claim 4 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 4 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein the autoencoder architecture is further configured to extrapolate uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
Therefore, claim 4 is subject-matter ineligible.
Regarding Claim 6:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 6 recites the same abstract ideas as claim 1. Therefore, claim 6 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 6 further recites additional elements of
wherein the device is a mobile robot including a Wi-Fi receiver, wherein the signal space is parameterized on Wi-Fi measurements of the Wi-Fi receiver, and wherein the location space is parametrized on coordinates of the mobile robot. (This element does not integrate the abstract idea into a practical application because it recites a technological environment in which to apply a judicial exception (see MPEP 2106.05(h)).)
Therefore, claim 6 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 6 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein the device is a mobile robot including a Wi-Fi receiver, wherein the signal space is parameterized on Wi-Fi measurements of the Wi-Fi receiver, and wherein the location space is parametrized on coordinates of the mobile robot specifies a particular technological environment to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(h)).
Therefore, claim 6 is subject-matter ineligible.
Regarding Claim 7:
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 7 recites the same abstract ideas as claim 1. Therefore, claim 7 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 7 further recites additional elements of
wherein the device is a vehicle, wherein the signal space is parametrized on acceleration measurements of the vehicle, and wherein the location space is parametrized on coordinates of the vehicle. (This element does not integrate the abstract idea into a practical application because it recites a technological environment in which to apply a judicial exception (see MPEP 2106.05(h)).)
Therefore, claim 7 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 7 do not provide significantly more than the abstract idea itself, taken alone and in combination because
wherein the device is a vehicle, wherein the signal space is parametrized on acceleration measurements of the vehicle, and wherein the location space is parametrized on coordinates of the vehicle specifies a particular technological environment to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(h)).
Therefore, claim 7 is subject-matter ineligible.
Regarding Claim 8:
Subject Matter Eligibility Analysis Step 1:
Claim 8 recites a method and is thus a process, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 8 recites
encoding each input data point of time series input data from an input state space into a latent space to produce latent data points indexed in time according to time indices of corresponding input data points, (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally encoding data.)
propagating the latent data points backward in time with a neural Ordinary Differential Equation (ODE) approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space; (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally propagating data points backward in time with an equation.)
propagating the initial point of latent dynamics of the device forward in time until a time index of interest using the neural ODE to produce a state of latent dynamics of the device at the time index of interest; (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally propagating a data point forward in time with an equation.)
decoding the state of latent dynamics of the device into the output state space different from the input state space to produce output data including the state of the device at the time index of interest (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally decoding the state of latent dynamics of the device.)
Therefore, claim 8 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 8 further recites additional elements of
for sensing a state of a device with continuous-time dynamics, (This element does not integrate the abstract idea into a practical application because it recites a technological environment in which to apply a judicial exception (see MPEP 2106.05(h)).)
wherein the input state space is a signal space, and wherein the output state space is a location space (This element does not integrate the abstract idea into a practical application because it recites a technological environment in which to apply a judicial exception (see MPEP 2106.05(h)).)
Therefore, claim 8 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 8 do not provide significantly more than the abstract idea itself, taken alone and in combination because
for sensing a state of a device with continuous-time dynamics, specifies a particular technological environment to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(h)).
wherein the input state space is a signal space, and wherein the output state space is a location space specifies a particular technological environment to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(h)).
Therefore, claim 8 is subject-matter ineligible.
Regarding claim 9, claim 9 recites substantially similar limitations to claim 2, and is therefore rejected under the same analysis.
Regarding claim 10, claim 10 recites substantially similar limitations to claim 3, and is therefore rejected under the same analysis.
Regarding claim 11, claim 11 recites substantially similar limitations to claim 4, and is therefore rejected under the same analysis.
Regarding claim 12, claim 12 recites substantially similar limitations to claim 6, and is therefore rejected under the same analysis.
Regarding claim 13, claim 13 recites substantially similar limitations to claim 7, and is therefore rejected under the same analysis.
Regarding Claim 14:
Subject Matter Eligibility Analysis Step 1:
Claim 14 recites a non-transitory computer readable storage medium and is thus an article of manufacture, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Claim 14 recites
encoding each input data point of time series input data from an input state space into a latent space to produce latent data points indexed in time according to time indices of corresponding input data points, (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally encoding data.)
propagating the latent data points backward in time with a neural Ordinary Differential Equation (ODE) approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space; (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally propagating data points backward in time with an equation.)
propagating the initial point of latent dynamics of the device forward in time until a time index of interest using the neural ODE to produce a state of latent dynamics of the device at the time index of interest; (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally propagating a data point forward in time with an equation.)
decoding the state of latent dynamics of the device into the output state space different from the input state space to produce output data including the state of the device at the time index of interest (This limitation is a mental process based on mathematical concepts as it encompasses a human mentally decoding the state of latent dynamics of the device.)
Therefore, claim 14 recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
Claim 14 further recites additional elements of
A non-transitory computer readable storage medium embodied thereon a program executable by a processor for performing a method (This element does not integrate the abstract idea into a practical application because it recites generic computing components on which to perform the abstract idea (see MPEP 2106.05(f)).)
for sensing a state of a device with continuous-time dynamics, (This element does not integrate the abstract idea into a practical application because it recites a technological environment in which to apply a judicial exception (see MPEP 2106.05(h)).)
wherein the input state space is a signal space, and wherein the output state space is a location space (This element does not integrate the abstract idea into a practical application because it recites a technological environment in which to apply a judicial exception (see MPEP 2106.05(h)).)
Therefore, claim 14 is not integrated into a practical application.
Subject Matter Eligibility Analysis Step 2B:
The additional elements of claim 14 do not provide significantly more than the abstract idea itself, taken alone and in combination because
A non-transitory computer readable storage medium embodied thereon a program executable by a processor for performing a method uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)).
for sensing a state of a device with continuous-time dynamics, specifies a particular technological environment to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(h)).
wherein the input state space is a signal space, and wherein the output state space is a location space specifies a particular technological environment to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(h)).
Therefore, claim 14 is subject-matter ineligible.
Regarding claim 15, claim 15 recites substantially similar limitations to claim 2, and is therefore rejected under the same analysis.
Regarding claim 16, claim 16 recites substantially similar limitations to claim 3, and is therefore rejected under the same analysis.
Regarding claim 17, claim 17 recites substantially similar limitations to claim 4, and is therefore rejected under the same analysis.
Regarding claim 18, claim 18 recites substantially similar limitations to claim 6, and is therefore rejected under the same analysis.
Regarding claim 19, claim 19 recites substantially similar limitations to claim 7, and is therefore rejected under the same analysis.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
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.
Claim(s) 1-4, 7-11, 13-17, and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Garsdal et al. (“Generative time series models using Neural ODE in Variational Autoencoders”) (hereafter referred to as Garsdal) in view of Kum et al. (US 2021/0316762 A1) (hereafter referred to as Kum).
Regarding claim 1, Garsdal teaches
An Artificial Intelligence (AI) system for sensing a state of a device with continuous-time dynamics (Garsdal, page 2, 2nd column, last paragraph, “We have trained the NODE model on three different data sets to test the robustness and performance of the method across multiple data sets and use cases” where “The solar power data set comes from real life solar power data where the power output of a solar cell is measured at a 30 minute interval throughout the day” (Garsdal, page 3, 1st column, last paragraph). Examiner notes that the solar cell is the device.),
the AI system including a neural network having an autoencoder architecture adapted for dynamic transformation of time series input data from an input state space indicative of the state of the device into an output state space indicative of the state of the device, comprising (Garsdal, page 3, 1st paragraph of Section C. Models, “The VAE [variational autoencoder] NODE [neural ordinary differential equation] models follow the structure seen in [1] and Figure 2 where time series data is encoded into a latent space using a time variant neural net such as a Recurrent Neural Network (RNN), or in our case a Long-Short Term Memory (LSTM) network. The latent state is represented from zt0 to ensure that decoding happens from the beginning of the time series” where “In this setting the latent space corresponds to the initial state of the latent trajectory z0, which an ODEsolver can take as input, in order to compute the entire latent trajectory z(t). The hidden trajectory is then evaluated at specific time steps, and these steps are then fed to a decoder that transforms it back into the temporal space” (Garsdal, page 2, 2nd column, 2nd paragraph) and “in general, the NODE VAE was able to capture the dynamics of the underlying data”. Examiner notes that the initial state is the input data from an input state space indicative of the state of the device. Examiner further notes that the transformed data from the decoder is the output state space indicative of the state of the device):
encode each input data point of the time series input data from the input state space into a latent space to produce latent data points indexed in time according to time indices of corresponding input data points (Garsdal, page 3, 1st paragraph of Section C. Models, “The VAE NODE models follow the structure seen in [1] and Figure 2 where time series data is encoded into a latent space using a time variant neural net such as a Recurrent Neural Network (RNN), or in our case a Long-Short Term Memory (LSTM) network. The latent state is represented from zt0 to ensure that decoding happens from the beginning of the time series” and Garsdal, page 2, Figure 2,
PNG
media_image1.png
542
1186
media_image1.png
Greyscale
Examiner notes that the RNN encoder takes the observed time and encodes it into the latent space according to the indices of the data points.)
and propagate the latent data points backward in time with a neural Ordinary Differential Equation (ODE) approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space (Garsdal, page 3, 1st paragraph of Section C. Models, “The VAE NODE models follow the structure seen in [1] and Figure 2 where time series data is encoded into a latent space using a time variant neural net such as a Recurrent Neural Network (RNN), or in our case a Long-Short Term Memory (LSTM) network. The latent state is represented from zt0 to ensure that decoding happens from the beginning of the time series. In order to enable this, we encode the time series in reverse thus ending up in zt0” where “This augmented ODE is solved backwards in time, starting from the final time step t1 and going to the initial time step t0” (Garsdal, page 2, 1st column, 1st paragraph) and Garsdal, page 2, Figure 2,
PNG
media_image2.png
542
1186
media_image2.png
Greyscale
Examiner notes that encoding in reverse is propagating the latent data points backward in time. Examiner further notes the boxed section in Figure 2 is the neural Ordinary Differential Equation approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space. Examiner notes that the estimated initial point of latent dynamics is the circle zt0 that points down to t0.);
propagate the initial point of latent dynamics of the device forward in time until a time index of interest using the neural ODE to produce a state of latent dynamics of the device at the time index of interest (Garsdal, page 2, Figure 2,
PNG
media_image3.png
542
1186
media_image3.png
Greyscale
Examiner notes that the ODE Solve propagates the initial point zt0 forward in time until the time index of interest, tm. Examiner further notes that the state of latent dynamics of the device at the time index of interest is ztM.);
decode the state of latent dynamics of the device into the output state space different from the input state space to produce output data including the state of the device at the time index of interest (Garsdal, page 2, Figure 2,
PNG
media_image2.png
542
1186
media_image2.png
Greyscale
and “Taking the NODE one step further, it can be utilized in continuous time extra- and interpolation for sequential data, by implementing the NODE as a decoder of a Variational Auto Encoder” (Garsdal, page 1, Introduction). Examiner notes that the boxed area in Figure 2 is decodes the state of latent dynamics and the output state space different from the input state space is the Extrapolation of tn+1 and tm. Examiner further notes that the state of the device at tn+1 and tm are the dots on the curve within the data space and the time index of interest is tm.).
Garsdal does not teach, but Kum does teach
at least one processor; and a memory having instructions stored thereon that cause the at least one processor to execute the neural network, train the neural network, or both (Kum, page 14, paragraph 0007, “According to various embodiments, an electronic device includes a memory and a processor connected to the memory and configured to execute at least one instruction stored in the memory. The processor may be configured to detect input data having a first time interval, detect first prediction data having a second time interval based on the input data using a preset recursive network, and detect second prediction data second prediction data having a third time interval based on the input data and the first prediction data using the recursive network” where “in this case, the recursive network 200 may include a plurality of encoders 410, a plurality of attention modules 420, and a plurality of decoders 430” (Kum, page 16, paragraph 0041) and “the first encoder 610 may include a plurality of recurrent neural networks (RNN)” (Kum, page 16, paragraph 0042). Examiner notes that the recursive network has the recurrent neural networks ),
wherein the input state space is a signal space (Kum, page 17, paragraph 0046, “Referring to FIG. 9, at operation 910, the electronic device 100 may detect input data X. In this case, the input data X may be a time-series data. The processor 180 may detect the input data X having a first time interval. For example, the electronic device 100 may be related to the vehicle 300. In such a case, the processor 180 may check moving trajectories of the surrounding vehicles 301. The processor 180 may collect information on a surrounding situation of the electronic device 100. In this case, the processor 180 may collect the information on a surrounding situation of the electronic device 100, based on at least one of image data obtained through the camera module 120 or sensing data obtained through the sensor module 130. Accordingly, the processor 180 may check moving the trajectories of the surrounding vehicles 301 based on the information on a surrounding situation of the electronic device 100. That is, the moving trajectories of the surrounding vehicles 301 may be detected as the input data X.” Examiner notes that the moving trajectories of the surrounding vehicles is the signal space.)
and wherein the output state space is a location space (Kum, page 17, paragraph 0044, “The first decoder 710 may detect a lateral movement of each surrounding vehicle 310. The second decoder 720 may detect a longitudinal movement of each surrounding vehicle 310. Accordingly the decoder may generate the first prediction data (Yinitial;
Y
i
^
) or the second prediction data (Yfinal;
Y
i
^
) by combining the lateral movement and longitudinal movement of each surrounding vehicle 310” where “according to various embodiments, the processor 180 may be configured to update the future trajectory based on the moving trajectory and future trajectory of the surrounding vehicle 301 using the recursive network 200 and to output the updated future trajectory as the second prediction data (Yfinal)” (Kum, page 19, paragraph 0078). Examiner notes that the output state space is the future trajectory and the location space is the future trajectory of the surrounding vehicles based on the latitudinal and longitudinal movements.)
Garsdal and Kum are analogous to the claimed invention because they both use encoders and decoders with time series data. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date to have implemented the autoencoder of Garsdal on the processor and memory of Kum. Thus, this would be applying a known technique (encoding and decoding time series) to a known device (processor and memory) ready for improvement to yield predictable results (encoded and decoded time series) (MPEP 2143 I. (C) Use of known technique to improve similar devices (methods, or products) in the same way). It would have been obvious to a person having ordinary skill in the art prior to the effective filing date to have implemented the autoencoder of Garsdal on the vehicle of Kum which uses a signal space and locations space as the input and output state spaces. Thus, this would be applying a known technique (encoding and decoding time series) to a known device (a vehicle) ready for improvement to yield predictable results (encoded and decoded time series) (MPEP 2143 I. (C) Use of known technique to improve similar devices (methods, or products) in the same way).
Regarding claim 2, Garsdal in view of Kum teaches the AI system of claim 1. Garsdal further teaches
wherein the autoencoder architecture is further configured to decode the state of latent dynamics of the device into the output state space same as the input state space to reconstruct the time series input data (Garsdal, page 2, Figure 2,
PNG
media_image2.png
542
1186
media_image2.png
Greyscale
and “Taking the NODE one step further, it can be utilized in continuous time extra- and interpolation for sequential data, by implementing the NODE as a decoder of a Variational Auto Encoder” (Garsdal, page 1, Introduction). Examiner notes that part of the autoencoder structure is the boxed area in Figure 2. Examiner further notes that the output state space same as the input state space is the dots on the curve corresponding to t0, t1, and tN. Additionally, Examiner notes that the Prediction is the reconstruction of the time series input data.).
Regarding claim 3, Garsdal in view of Kum teaches the AI system of claim 1. Garsdal further teaches
wherein the state of the device corresponds to a trajectory of the device (Garsdal, page 1, 1st column, 1st paragraph, “To solve the derivatives given in Equation (2) the NODE uses a black box ordinary differential equation solver, which takes as input an initial hidden state h(0) to solve an initial value problem up to some given time T. This way the ODE solver yields a representation of a continuous hidden state trajectory, instead of a discrete amount of hidden states. This also means that any specific hidden state along the hidden trajectory can be evaluated, even with uneven step sizes, which is one of the advantages with this approach”),
and wherein the autoencoder architecture is further configured to interpolate the trajectory of the device based on the state of latent dynamics of the device (Garsdal, page 2, Section C. Using Neural ODE’s as VAE, “NODE’s are intrinsically well suited for temporal data and have many advantages due to the continuous aspect of the underlying ODE. It can be used as a generative model utilizing learned representations of a latent space. In this setting it can be used to predict or extrapolate data, as well as interpolate and impute missing data within a time series” where “Our work revolves around the NODE framework as a VAE, where it can learn the latent space distribution of a time series. In this setting the latent space corresponds to the initial state of the latent trajectory z0, which ODEsolver can take as input, in order to compute the entire latent trajectory z(t). The hidden trajectory is then evaluated at specific time steps, and these steps are then fed to a decoder that transforms it back into the temporal space” (Garsdal, page 2, 2nd column, 2nd paragraph).).
Regarding claim 4, Garsdal in view of Kum teaches the AI system of claim 3. Garsdal further teaches
wherein the autoencoder architecture is further configured to extrapolate the trajectory of the device based on the state of latent dynamics of the device (Garsdal, page 2, Section C. Using Neural ODE’s as VAE, “NODE’s are intrinsically well suited for temporal data and have many advantages due to the continuous aspect of the underlying ODE. It can be used as a generative model utilizing learned representations of a latent space. In this setting it can be used to predict or extrapolate data, as well as interpolate and impute missing data within a time series” where “Our work revolves around the NODE framework as a VAE, where it can learn the latent space distribution of a time series. In this setting the latent space corresponds to the initial state of the latent trajectory z0, which ODEsolver can take as input, in order to compute the entire latent trajectory z(t). The hidden trajectory is then evaluated at specific time steps, and these steps are then fed to a decoder that transforms it back into the temporal space” (Garsdal, page 2, 2nd column, 2nd paragraph).).
Regarding claim 7, Garsdal in view of Kum teaches the AI system of claim 1. Garsdal in view of Kum further teach
wherein the device is a vehicle, wherein the signal space is parametrized on acceleration measurements of the vehicle (Kum, page 17, paragraph 0046, “Referring to FIG. 9, at operation 910, the electronic device 100 may detect input data X. In this case, the input data X may be a time-series data. The processor 180 may detect the input data X having a first time interval. For example, the electronic device 100 may be related to the vehicle 300. In such a case, the processor 180 may check moving trajectories of the surrounding vehicles 301. The processor 180 may collect information on a surrounding situation of the electronic device 100. In this case, the processor 180 may collect the information on a surrounding situation of the electronic device 100, based on at least one of image data obtained through the camera module 120 or sensing data obtained through the sensor module 130. Accordingly, the processor 180 may check moving the trajectories of the surrounding vehicles 301 based on the information on a surrounding situation of the electronic device 100. That is, the moving trajectories of the surrounding vehicles 301 may be detected as the input data X.” Examiner notes that the moving trajectories of the surrounding vehicles is the signal space parameterized on acceleration measurements.),
and wherein the location space is parametrized on coordinates of the vehicle (Kum, page 17, paragraph 0044, “The first decoder 710 may detect a lateral movement of each surrounding vehicle 310. The second decoder 720 may detect a longitudinal movement of each surrounding vehicle 310. Accordingly the decoder may generate the first prediction data (Yinitial;
Y
i
^
) or the second prediction data (Yfinal;
Y
i
^
) by combining the lateral movement and longitudinal movement of each surrounding vehicle 310” where “according to various embodiments, the processor 180 may be configured to update the future trajectory based on the moving trajectory and future trajectory of the surrounding vehicle 301 using the recursive network 200 and to output the updated future trajectory as the second prediction data (Yfinal)” (Kum, page 19, paragraph 0078). Examiner notes that the output state space is the future trajectory and the location space parametrized on coordinates of the vehicle is the future trajectory of the surrounding vehicles based on the latitudinal and longitudinal movements.).
Garsdal and Kum are analogous to the claimed invention because they both use encoders and decoders with time series data. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date to have implemented the autoencoder of Garsdal on the vehicle of Kum. Thus, this would be applying a known technique (encoding and decoding time series) to a known device (a vehicle) ready for improvement to yield predictable results (encoded and decoded time series) (MPEP 2143 I. (C) Use of known technique to improve similar devices (methods, or products) in the same way).
Regarding claim 8, Garsdal teaches
A method for sensing a state of a device with continuous-time dynamics, comprising (Garsdal, page 2, 2nd column, last paragraph, “We have trained the NODE model on three different data sets to test the robustness and performance of the method across multiple data sets and use cases” where “The solar power data set comes from real life solar power data where the power output of a solar cell is measured at a 30 minute interval throughout the day” (Garsdal, page 3, 1st column, last paragraph). Examiner notes that the solar cell is the device.):
encoding each input data point of time series input data from an input state space into a latent space to produce latent data points indexed in time according to time indices of corresponding input data points (Garsdal, page 3, 1st paragraph of Section C. Models, “The VAE NODE models follow the structure seen in [1] and Figure 2 where time series data is encoded into a latent space using a time variant neural net such as a Recurrent Neural Network (RNN), or in our case a Long-Short Term Memory (LSTM) network. The latent state is represented from zt0 to ensure that decoding happens from the beginning of the time series” and Garsdal, page 2, Figure 2,
PNG
media_image1.png
542
1186
media_image1.png
Greyscale
Examiner notes that the RNN encoder takes the observed time and encodes it into the latent space according to the indices of the data points.),
propagating the latent data points backward in time with a neural Ordinary Differential Equation (ODE) approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space (Garsdal, page 3, 1st paragraph of Section C. Models, “The VAE NODE models follow the structure seen in [1] and Figure 2 where time series data is encoded into a latent space using a time variant neural net such as a Recurrent Neural Network (RNN), or in our case a Long-Short Term Memory (LSTM) network. The latent state is represented from zt0 to ensure that decoding happens from the beginning of the time series. In order to enable this, we encode the time series in reverse thus ending up in zt0” where “This augmented ODE is solved backwards in time, starting from the final time step t1 and going to the initial time step t0” (Garsdal, page 2, 1st column, 1st paragraph) and Garsdal, page 2, Figure 2,
PNG
media_image2.png
542
1186
media_image2.png
Greyscale
Examiner notes that solving the ODE backwards in time is propagating the latent data points backward in time. Examiner further notes the boxed section in Figure 2 is the neural Ordinary Differential Equation approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space. Examiner notes that the estimated initial point of latent dynamics is the circle zt0 that points down to t0.);
propagating the initial point of latent dynamics of the device forward in time until a time index of interest using the neural ODE to produce a state of latent dynamics of the device at the time index of interest (Garsdal, page 2, Figure 2,
PNG
media_image3.png
542
1186
media_image3.png
Greyscale
Examiner notes that the ODE Solve is the latent subnetwork that propagates the initial point zt0 forward in time until the time index of interest, tM. Examiner further notes that the state of latent dynamics of the device at the time index of interest is ztM.);
and decoding the state of latent dynamics of the device into an output state space different from the input state space to produce output data including the state of the device at the time index of interest (Garsdal, page 2, Figure 2,
PNG
media_image2.png
542
1186
media_image2.png
Greyscale
and “Taking the NODE one step further, it can be utilized in continuous time extra- and interpolation for sequential data, by implementing the NODE as a decoder of a Variational Auto Encoder” (Garsdal, page 1, Introduction). Examiner notes that the boxed area in Figure 2 is the extended decoder and the output state space different from the input state space is the Extrapolation of tn+1 and tM. Examiner further notes that the state of the device at tn+1 and tm are the dots on the curve within the data space and the time index of interest is tM.).
Garsdal does not teach, but Kum does teach
wherein the input state space is a signal space (Kum, page 17, paragraph 0046, “Referring to FIG. 9, at operation 910, the electronic device 100 may detect input data X. In this case, the input data X may be a time-series data. The processor 180 may detect the input data X having a first time interval. For example, the electronic device 100 may be related to the vehicle 300. In such a case, the processor 180 may check moving trajectories of the surrounding vehicles 301. The processor 180 may collect information on a surrounding situation of the electronic device 100. In this case, the processor 180 may collect the information on a surrounding situation of the electronic device 100, based on at least one of image data obtained through the camera module 120 or sensing data obtained through the sensor module 130. Accordingly, the processor 180 may check moving the trajectories of the surrounding vehicles 301 based on the information on a surrounding situation of the electronic device 100. That is, the moving trajectories of the surrounding vehicles 301 may be detected as the input data X.” Examiner notes that the moving trajectories of the surrounding vehicles is the signal space.)
and wherein the output state space is a location space (Kum, page 17, paragraph 0044, “The first decoder 710 may detect a lateral movement of each surrounding vehicle 310. The second decoder 720 may detect a longitudinal movement of each surrounding vehicle 310. Accordingly the decoder may generate the first prediction data (Yinitial;
Y
i
^
) or the second prediction data (Yfinal;
Y
i
^
) by combining the lateral movement and longitudinal movement of each surrounding vehicle 310” where “according to various embodiments, the processor 180 may be configured to update the future trajectory based on the moving trajectory and future trajectory of the surrounding vehicle 301 using the recursive network 200 and to output the updated future trajectory as the second prediction data (Yfinal)” (Kum, page 19, paragraph 0078). Examiner notes that the output state space is the future trajectory and the location space is the future trajectory of the surrounding vehicles based on the latitudinal and longitudinal movements.)
Garsdal and Kum are analogous to the claimed invention because they both use encoders and decoders with time series data. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date to have implemented the autoencoder of Garsdal on the vehicle of Kum which uses a signal space and locations space as the input and output state spaces. Thus, this would be applying a known technique (encoding and decoding time series) to a known device (a vehicle) ready for improvement to yield predictable results (encoded and decoded time series) (MPEP 2143 I. (C) Use of known technique to improve similar devices (methods, or products) in the same way).
Regarding claim 9, claim 9 recites substantially similar limitations to claim 2, and is therefore rejected under the same analysis.
Regarding claim 10, claim 10 recites substantially similar limitations to claim 3, and is therefore rejected under the same analysis.
Regarding claim 11, claim 11 recites substantially similar limitations to claim 4, and is therefore rejected under the same analysis.
Regarding claim 13, claim 13 recites substantially similar limitations to claim 7, and is therefore rejected under the same analysis.
Regarding claim 14, Garsdal teaches
A method for sensing a state of a device with continuous-time dynamics (Garsdal, page 2, 2nd column, last paragraph, “We have trained the NODE model on three different data sets to test the robustness and performance of the method across multiple data sets and use cases” where “The solar power data set comes from real life solar power data where the power output of a solar cell is measured at a 30 minute interval throughout the day” (Garsdal, page 3, 1st column, last paragraph). Examiner notes that the solar cell is the device.),
encoding each input data point of the time series input data from the input state space into a latent space to produce latent data points indexed in time according to time indices of corresponding input data points (Garsdal, page 3, 1st paragraph of Section C. Models, “The VAE NODE models follow the structure seen in [1] and Figure 2 where time series data is encoded into a latent space using a time variant neural net such as a Recurrent Neural Network (RNN), or in our case a Long-Short Term Memory (LSTM) network. The latent state is represented from zt0 to ensure that decoding happens from the beginning of the time series” and Garsdal, page 2, Figure 2,
PNG
media_image1.png
542
1186
media_image1.png
Greyscale
Examiner notes that the RNN encoder takes the observed time and encodes it into the latent space according to the indices of the data points.)
propagating the latent data points backward in time with a neural Ordinary Differential Equation (ODE) approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space (Garsdal, page 3, 1st paragraph of Section C. Models, “The VAE NODE models follow the structure seen in [1] and Figure 2 where time series data is encoded into a latent space using a time variant neural net such as a Recurrent Neural Network (RNN), or in our case a Long-Short Term Memory (LSTM) network. The latent state is represented from zt0 to ensure that decoding happens from the beginning of the time series. In order to enable this, we encode the time series in reverse thus ending up in zt0” where “This augmented ODE is solved backwards in time, starting from the final time step t1 and going to the initial time step t0” (Garsdal, page 2, 1st column, 1st paragraph) and Garsdal, page 2, Figure 2,
PNG
media_image2.png
542
1186
media_image2.png
Greyscale
Examiner notes that encoding in reverse is propagating the latent data points backward in time. Examiner further notes the boxed section in Figure 2 is the neural Ordinary Differential Equation approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space. Examiner notes that the estimated initial point of latent dynamics is the circle zt0 that points down to t0.);
propagating the initial point of latent dynamics of the device forward in time till a time index of interest using the neural ODE to produce a state of latent dynamics of the device at the time index of interest (Garsdal, page 2, Figure 2,
PNG
media_image3.png
542
1186
media_image3.png
Greyscale
Examiner notes that the ODE Solve is the latent subnetwork that propagates the initial point zt0 forward in time until the time index of interest, tm. Examiner further notes that the state of latent dynamics of the device at the time index of interest is ztM.);
decoding the state of latent dynamics of the device into the output state space different from the input state space to produce output data including the state of the device at the time index of interest (Garsdal, page 2, Figure 2,
PNG
media_image2.png
542
1186
media_image2.png
Greyscale
and “Taking the NODE one step further, it can be utilized in continuous time extra- and interpolation for sequential data, by implementing the NODE as a decoder of a Variational Auto Encoder” (Garsdal, page 1, Introduction). Examiner notes that the boxed area in Figure 2 is the extended decoder and the output state space different from the input state space is the Extrapolation of tn+1 and tm. Examiner further notes that the state of the device at tn+1 and tm are the dots on the curve within the data space and the time index of interest is tm.).
Garsdal does not teach, but Kum does teach
A non-transitory computer readable storage medium embodied thereon a program executable by a processor for performing a method (Kum, page 19, paragraph 0079, “For example, a processor (e.g., the processor 180) of the computer device may invoke at least one of the one or more instructions stored in the storage medium, and may execute the instruction. This enables the computer device to operate to perform at least one function based on the invoked at least one instruction. The one or more instructions may include a code generated by a compiler or a code executable by an interpreter. The storage medium readable by the computer device may be provided in the form of a non-transitory storage medium. In this case, the term ‘non-transitory’ merely means that the storage medium is a tangible device and does not include a signal (e.g., electromagnetic wave).”):
Garsdal and Kum are analogous to the claimed invention because they both use encoders and decoders with time series data. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date to have implemented the autoencoder of Garsdal on the non-transitory computer readable storage medium of Kum. Thus, this would be applying a known technique (encoding and decoding time series) to a known device (a non-transitory computer readable storage medium) ready for improvement to yield predictable results (encoded and decoded time series) (MPEP 2143 I. (C) Use of known technique to improve similar devices (methods, or products) in the same way).
Regarding claim 15, claim 15 recites substantially similar limitations to claim 2, and is therefore rejected under the same analysis.
Regarding claim 16, claim 16 recites substantially similar limitations to claim 3, and is therefore rejected under the same analysis.
Regarding claim 17, claim 17 recites substantially similar limitations to claim 4, and is therefore rejected under the same analysis.
Regarding claim 19, claim 19 recites substantially similar limitations to claim 7, and is therefore rejected under the same analysis.
Claim(s) 6 and 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Garsdal in view of Kum, in further view of Blaiotta et al. (US 12,367,377 B2) (hereafter referred to as Blaiotta).
Regarding claim 6, Garsdal in view of Kum teaches the AI system of claim 1. Garsdal in view of Kum does not teach, but Blaiotta does teach
wherein the device is a mobile robot including a Wi- Fi receiver (Blaiotta, page 10, column 2, lines. 61-65, “Various features of the present invention relate to making time-series predictions relating to a computer-controlled system, such as a robot, a (semi-) autonomous vehicle, a manufacturing machine, a personal assistant, or an access control system” where “as also illustrated in FIG. 1, the data interface may be constituted by a data storage interface 120 which may access the data 030, 041, 042 from a data storage 021. For example, the data storage interface 120 may be a memory interface or a persistent storage interface, e.g., a hard disk or an SSD interface, but also a personal, local or wide area network interface such as Bluetooth, Zigbee or Wi-Fi interface or an ethernet or fiberoptic interface. The data storage 021 may be an internal data storage of the system 100, such as a hard drive or SSD, but also an external data storage, e.g., a network-accessible data storage. In some embodiments, the data 030, 041, 042 may each be accessed from a different data storage, e.g., via a different subsystem of the data storage interface 120. Each subsystem may be of a type as is described above for the data storage interface 120” (Blaiotta, page 14, column 10, lines 1-16). Examiner notes that the device is a robot that has a Wi-Fi interface or Wi-Fi receiver.),
wherein the signal space is parameterized on Wi-Fi measurements of the Wi-Fi receiver (Blaiotta, page 11, column 4, lines 45-50, “The function can take as input previous values of the one or more measurable quantities for the object and the further object, and in some cases (but not necessarily) also the values of the time-invariant laten features for the object and the further object” where “Thus, the trainable function that determines the pairwise contributions, in such cases takes as input values of the one or more measurable quantities only for the previous state” (Blaiotta, page 11, column 4, lines 62-65) where “FIG. 1 shows a system for training a decoder model for making time-series predictions of multiple interacting objects” (Blaiotta, page 14, column 9, lines 13-15) and “as also illustrated in FIG. 1, the data interface may be constituted by a data storage interface 120 which may access the data 030, 041, 042 from a data storage 021. For example, the data storage interface 120 may be a memory interface or a persistent storage interface, e.g., a hard disk or an SSD interface, but also a personal, local or wide area network interface such as Bluetooth, Zigbee or Wi-Fi interface or an ethernet or fiberoptic interface” (Blaiotta, page 14, column 10, lines 1-8). Examiner notes that the measurable quantities taken as input are the Wi-Fi measurements since the Wi-Fi receiver or Wi-Fi interface accessed them. Examiner further notes that the signal space is the Wi-Fi interface passing the input.),
and wherein location space is parametrized on coordinates of the mobile robot (Blaiotta, page 18, column18, lines 35-55, “As shown in the figure, the decoder model may comprise a trained graph model GM, 602, and a trained local function LF, 603. The graph model GM may be applied to obtain first prediction contributions FPCi, 670, for respective objects, while the local function LF may be used to obtain second prediction contributions SPCi, 671, for the respective objects….The first and second contributions FPCi, SPCi, may then be combined, in a combination operation CMB, 680, to obtain predicted values xi,t+1, 611, of the one or more measurable quantities” where “The measurable quantities may comprise positional information about an object. The measurable quantities may for example comprise a position of the object. For example, the measurable quantities at a point in time may be represented as 2D coordinates or 3D coordinates” (Blaiotta, page 17, column 16, lines 54-58) and where “for a robot, the objects can for example be components of the robot itself, e.g., different components of a robot arm connected to each other by joints, and/or objects in the environment of the robot” (Blaiotta, page 11, column 3, lines 6-9). Examiner notes that the predicted values of the measurable quantities are the output state space. Examiner further notes that the 2D coordinates of the object are the coordinates of the mobile robot.).
Garsdal, Kum, and Blaiotta are analogous to the claimed invention because they use encoders and decoders with time series data. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date to have implemented the autoencoder of Garsdal in view of Kum on the robot of Blaiotta. Thus, this would be applying a known technique (encoding and decoding time series) to a known device (a robot) ready for improvement to yield predictable results (encoded and decoded time series) (MPEP 2143 I. (C) Use of known technique to improve similar devices (methods, or products) in the same way).
Regarding claim 12, claim 12 recites substantially similar limitations to claim 6, and is therefore rejected under the same analysis.
Regarding claim 18, claim 18 recites substantially similar limitations to claim 6, and is therefore rejected under the same analysis.
Response to Arguments
The claim objections have been overcome in light of the instant amendments.
The 112 rejections have been overcome in light of the instant amendments.
On pages 9-10, Applicant argues:
At the outset, Applicant respectfully submits that the human mind is not equipped to encode continuous-time, high-dimensional time series data into a latent space representation. The claimed encoding is not a conceptual labeling of data, but execution of a trained neural network having an autoencoder architecture comprising weighted connections, nonlinear activation functions, and learned parameters stored in memory and executed by a processor. A human cannot apply learned weight parameters to transform each time-indexed data point from a signal state space into a structured latent space in a deterministic and repeatable manner. The production of "latent data points indexed in time" is a computational artifact of neural-network inference, not a mental observation or evaluation.
Furthermore, claim 1 recites propagating the latent data points backward in time with a neural Ordinary Differential Equation (ODE) approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space, and thereafter propagating the estimated initial point of latent dynamics of the device forward in time using the same neural ODE until a time index of interest to produce a state of latent dynamics of the device at the time index of interest. These limitations are not directed to a human performing abstract reasoning in the mind, but to a specific machine-implemented dynamical modeling framework tied to a neural network architecture. The backward propagation requires operating on multiple latent data points indexed according to corresponding time indices and integrating the learned dynamics in reverse temporal order to infer an initial point of latent dynamics. The forward propagation then uses that inferred initial point to compute, through the same learned dynamical model, a latent state at a specific time index of interest. This ordered combination presupposes storage of time series input data, execution of trained neural-network parameters, and numerical integration across potentially high-dimensional latent vectors-operations that depend on processor-executed computation over digital representations and trained model parameters.
Finally, claim 1 recites decoding the state of latent dynamics into an output state space different from the input state space, wherein the input state space is a signal space and the output state space is a location space. This decoding step requires transforming a latent vector into a physically meaningful representation in a different representational domain to enable sensing a state of a device with continuous time dynamics. The transformation from signal-space measurements to location-space outputs involves learned nonlinear mappings between distinct state spaces. A human cannot mentally execute such high-dimensional nonlinear transformations defined by stored neural- network parameters so as to produce quantitatively accurate location space
output data representing the state of the device at a time index of interest.
As in SRI Int 'l v. Cisco, where the Federal Circuit recognized that humans are not equipped to analyze network packet data to detect suspicious activity, the same rationale applies here: a human is not equipped to encode time-indexed signal data into latent vectors using trained neural networks, integrate neural ODEs backward and forward in continuous time, and decode latent dynamics into a different state space representing device location. These are computational dynamical-system operations executed by a processor for sensing a state of a device with continuous-time dynamics, not cognitive mental steps.
Regarding the Applicant’s argument that claim 1 does not recite an abstract idea, Examiner respectfully disagrees. Specifically, Examiner notes that the claims do not specify how much data is being encoded, propagated, and decoded. As such, under broadest reasonable interpretation, a human can very well encode, propagate, and decode data points as claim 1 claims. Examiner additionally notes that the propagating steps in claim 1 recite using an Ordinary Differential Equation to propagate data points and does not need to rely upon a neural network architecture as the Applicant is describing. Examiner further notes that encoding, propagating, and decoding data points are evaluations and are thus mental processes.
On pages 12-14, Applicant argues:
As described in the Specification, conventional device tracking techniques employing autoencoders are inherently limited to producing output data belonging to the same state space as the input data which prevents transformation between heterogeneous state variables (e.g., transforming signal measurements into physical location). Further, conventional frame-based or sequence-based localization approaches suffer from technical deficiencies, including inability to handle irregular sampling, imbalance between signal-domain and location-domain training data, and lack of a continuous-time latent representation capable of querying device state at arbitrary time indices. These deficiencies create a concrete technological problem in sensing and tracking devices with continuous-time dynamics, and accordingly there is a need to extend the autoencoders to data transformation among different state spaces to enable location tracking by transforming signals (e.g. Wi-Fi signals). See paragraphs [0002]-[0014] and [003 l]-[0033] of the as-filed Specification.
…
As described in paragraphs [0046], [0070], and [0072] of the Specification, the claimed invention allows Wi-Fi measurements (signal space) or acceleration measurements to be transformed into physical location coordinates (location space) for real-time tracking of a device (e.g. mobile robot or vehicle). Amended independent claim 1 enables decoding the state of latent dynamics of the device into an output state space different from the input state space, wherein the input state space is a signal space (e.g., Wi-Fi measurements, acceleration signals) and the output state space is a location space (e.g., physical coordinates). As stated in paragraph [0005] of the Specification, this overcomes the fundamental limitation of conventional autoencoders, in which the input data and the output data belong to the same state space, regardless of the applications.
Applicant respectfully submits that, the claimed decoding operation produces output data that includes the state of the device at a time index of interest, wherein the output state space different from the input state space, thereby enabling transformation from signal-domain measurements to spatial-domain coordinates. This cross-domain decoding directly addresses the technical problem identified in the Specification, namely, the inability of conventional architectures to perform state transformation between signal variables and physical location variables while maintaining a continuous-time latent representation. This represents a concrete technological improvement in device state sensing and tracking: Wi-Fi or inertial signal measurements (signal space) are dynamically transformed into physical position coordinates (location space) through a learned continuous-time latent dynamical model.
Regarding the Applicant’s argument that claim 1 provides an improvement, Examiner respectfully disagrees. Specifically, Examiner notes that claim 1 does not recite any Wi-Fi measurements, acceleration signals, or physical coordination coordinates. Even if Applicant was arguing this for claim 6 and claim 7, which recite said Wi-Fi measurements, acceleration signals, and physical coordination coordinates, these measurements, signals, and coordinates specify a particular technological environment to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(h)).
On page 14, Applicant argues:
Similar to the Desjardins case, the improvement here lies in the internal operation of the claimed AI system itself, specifically, in the processor-executed autoencoder architecture that (i) encodes time-series input data from a signal-space input state space into a latent space, (ii) propagates the latent data points backward in time with a neural (ODE) to estimate an initial point of latent dynamics of the device in the latent space, (iii) propagates the initial point of latent dynamics of the device forward in time until a time index of interest using the neural ODE to produce a state of latent dynamics of the device at the time index of interest, and (iii) decodes the state of latent dynamics into an output state space different from the input state space, including transforming signal measurements into physical location coordinates.
This ordered combination enables continuous-time estimation of a device's physical state at a time index of interest and produces output data representing the actual state of the device (e.g., location or trajectory) rather than merely reconstructing input signals. The claimed architecture therefore improves how a computer system senses, models, and outputs real-world device state information, resulting in enhanced real-time tracking and state determination of mobile devices such as robots or vehicles. Accordingly, the claims are directed to a specific technological improvement in device state sensing and tracking, not to an abstract concept.
Regarding the Applicant’s argument that processor-executed autoencoder architecture provides an improvement, Examiner respectfully disagrees. Specifically, Examiner notes that points (i), (ii), and both points (iii) are mental processes. Performing these steps on a processor does not provide an improvement because it recites generic computing components on which to perform the abstract idea (see MPEP 2106.05(f)).
On page 16, Applicant argues:
This ordered combination of elements is not a generic implementation of "encoding," "propagating," or "decoding" on a general-purpose computer. In particular, the encoding step is not generic data transformation, but part of an autoencoder architecture specifically adapted to produce time-indexed latent data points suitable for continuous-time dynamical modeling. The backward propagation step is not a mere application of mathematics in the abstract; rather, it requires configuring a neural ODE to approximate physical device dynamics in latent space and to infer an initial latent condition consistent with observed signal measurements. The subsequent forward propagation step is not routine numerical processing, but a structured use of the same neural ODE to produce a state of latent dynamics of the device at the time index of interest, thereby enabling continuous-time sensing beyond discrete sampling constraints.
Regarding the Applicant’s argument that the autoencoder provides significantly more, Examiner respectfully disagrees. Specifically, Examiner notes that the AI system as claimed in claim 1 uses a computer as a tool to perform the abstract idea and cannot provide significantly more (see MPEP 2106.05(f)). Examiner further notes that “the AI system including a neural network having an autoencoder architecture adapted for dynamic transformation of time series input data from an input state space indicative of the state of the device into an output state space indicative of the state of the device, comprising” is the well understood, routine, and conventional activity of “transmitting or receiving data over a network” (see MPEP 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network)). Neither of these limitations regarding the autoencoder can provide significantly more.
On pages 16-17, Applicant argues:
Further, the decoding step is not a generic output operation. The claim requires decoding the latent state into an output state space different from the input state space, specifically transforming from a signal space (e.g., Wi-Fi or acceleration measurements) into a location space ( e.g., physical coordinates). This cross-domain transformation overcomes the technical limitation of conventional autoencoders that reconstruct outputs only within the same state space as the input. The claimed architecture therefore extends autoencoder functionality to heterogeneous state-space transformation in a continuous-time framework, enabling real-time indoor localization and device tracking under irregular sampling conditions. Such an arrangement is neither well-understood nor routine, but instead reflects a specifically configured sensing and tracking system that improves the functioning of localization technology itself.
Regarding the Applicant’s argument that the decoding step provides significantly more, Examiner respectfully disagrees. Specifically, Examiner notes that “decode the state of latent dynamics of the device into the output state space different from the input state space to produce output data including the state of the device at the time index of interest” is a mental process based on mathematical concepts as it encompasses a human mentally decoding the state of latent dynamics of the device. As such, it cannot provide an improvement.
Examiner further notes that “wherein the input state space is a signal space, and wherein the output state space is a location space” recites a technological environment in which to apply a judicial exception and cannot provide significantly more (see MPEP 2106.05(h)).
On pages 18-19, Applicant argues:
Thus, Garsdal merely discloses encoding time series data in reverse using neural network to obtain a latent state. However, Garsdal does not disclose propagating latent data points indexed in time according to time indices of corresponding input data points backward in time using the neural ODE itself to estimate an initial point of latent dynamics of a device in the latent space, as recited in amended independent claim 8. In Garsdal, the initial latent state is produced directly by reverse sequential encoding of the time series data performed by RNN/LSTM encoder. The neural ODE in Garsdal does not operate on temporally indexed latent data points and is not employed to propagate such latent data points backward in time to estimate the initial point of latent dynamics of the device in the latent space.
Garsdal further discloses that "[i]nstead, when backpropagating through the network f to update the weights, 8, a second augmented ODE is solved, which can return the gradient of the parameters with respect to the loss function. The augmented ODE is solved backwards in time, starting from the final time step t1 and going to the initial time step to". The augmented ODE occurs within the context of the adjoint sensitivity method used for gradient computation of parameters with respect to a loss function during training of the NODE framework. See sub-section "B" of Section "Introduction" of Garsdal. At the outset, the augmented ODE does not operate on latent data points indexed in time according to time indices of corresponding input data points. Therefore, even during training, the ODE is not "propagating latent data points" backward in time to determine an initial point of latent dynamics of the device in the latent space. Instead, it serves only as a mathematical mechanism to compute gradients of the parameters with respect to the loss function during training of the NODE framework.
Regarding Applicant’s argument that Garsdal does not explicitly teach “propagating the latent data points backward in time with a neural Ordinary Differential Equation (ODE) approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space”, Examiner respectfully disagrees. Specifically, Garsdal does teach this (Garsdal, page 3, 1st paragraph of Section C. Models, “The VAE NODE models follow the structure seen in [1] and Figure 2 where time series data is encoded into a latent space using a time variant neural net such as a Recurrent Neural Network (RNN), or in our case a Long-Short Term Memory (LSTM) network. The latent state is represented from zt0 to ensure that decoding happens from the beginning of the time series. In order to enable this, we encode the time series in reverse thus ending up in zt0” where “This augmented ODE is solved backwards in time, starting from the final time step t1 and going to the initial time step t0” (Garsdal, page 2, 1st column, 1st paragraph) and Garsdal, page 2, Figure 2,
PNG
media_image2.png
542
1186
media_image2.png
Greyscale
Examiner notes that solving the ODE backwards in time is propagating the latent data points backward in time. Examiner further notes the boxed section in Figure 2 is the neural Ordinary Differential Equation approximating dynamics of the device in the latent space to estimate an initial point of latent dynamics of the device in the latent space. Examiner notes that the estimated initial point of latent dynamics is the circle zt0 that points down to t0.).
Examiner further notes that under broadest reasonable interpretation, propagating latent points backwards in time with a neural ODE includes solving the ODE backwards in time. Examiner additionally notes that the augmented ODE as recited under section B. Adjoint-state method is used as part of the NODE framework for the VAE NODE and is thus included.
On page 13, Applicant argues:
Applicant respectfully submits that the extrapolated points tN+l and tM shown in Figure 2 are merely future predictions of the same observed variable that was originally provided as input to the model. As shown in Figure 2, the extrapolated points tN+l and tM lie on the same continuous curve and reside in the identical data space as the observed input points at t0 through tN. A change in temporal index does not constitute a change in state space, rather it constitutes prediction along the time axis within the same representational domain. Thus, the extrapolated points remain expressed in the identical data space as the input observations. Accordingly, Garsdal merely discloses extrapolation within a single state space. There is no disclosure in Garsdal of transforming signal-domain measurements into spatial coordinate representations, in a manner as recited in amended independent claim 1. Thus, Garsdal fails to disclose decoding the state of latent dynamics of a device into an output state space different from the input space, let alone disclose that the input state space is a signal space, and the output state space is a location space.
Regarding the Applicant’s argument that the prior art does not disclose “decoding the state of latent dynamics of the device into an output state space different from the input state space to produce output data including the state of the device at the time index of interest, wherein the input state space is a signal space, and wherein the output state space is a location space”, Examiner respectfully disagrees. Specifically, Examiner notes that Garsdal teaches “decoding the state of latent dynamics of the device into an output state space different from the input state space to produce output data including the state of the device at the time index of interest” (Garsdal, page 2, Figure 2,
PNG
media_image2.png
542
1186
media_image2.png
Greyscale
and “Taking the NODE one step further, it can be utilized in continuous time extra- and interpolation for sequential data, by implementing the NODE as a decoder of a Variational Auto Encoder” (Garsdal, page 1, Introduction). Examiner notes that the boxed area in Figure 2 decodes the state of latent dynamics and the output state space different from the input state space is the Extrapolation of tn+1 and tm. Examiner further notes that the state of the device at tn+1 and tm are the dots on the curve within the data space and the time index of interest is tm.) Examiner additionally notes that under broadest reasonable interpretation, the extrapolation as the output state space is different than the observed data which was inputted under the input state space.
Kum goes on to teach
wherein the input state space is a signal space (Kum, page 17, paragraph 0046, “Referring to FIG. 9, at operation 910, the electronic device 100 may detect input data X. In this case, the input data X may be a time-series data. The processor 180 may detect the input data X having a first time interval. For example, the electronic device 100 may be related to the vehicle 300. In such a case, the processor 180 may check moving trajectories of the surrounding vehicles 301. The processor 180 may collect information on a surrounding situation of the electronic device 100. In this case, the processor 180 may collect the information on a surrounding situation of the electronic device 100, based on at least one of image data obtained through the camera module 120 or sensing data obtained through the sensor module 130. Accordingly, the processor 180 may check moving the trajectories of the surrounding vehicles 301 based on the information on a surrounding situation of the electronic device 100. That is, the moving trajectories of the surrounding vehicles 301 may be detected as the input data X.” Examiner notes that the moving trajectories of the surrounding vehicles is the signal space.)
and wherein the output state space is a location space (Kum, page 17, paragraph 0044, “The first decoder 710 may detect a lateral movement of each surrounding vehicle 310. The second decoder 720 may detect a longitudinal movement of each surrounding vehicle 310. Accordingly the decoder may generate the first prediction data (Yinitial;
Y
i
^
) or the second prediction data (Yfinal;
Y
i
^
) by combining the lateral movement and longitudinal movement of each surrounding vehicle 310” where “according to various embodiments, the processor 180 may be configured to update the future trajectory based on the moving trajectory and future trajectory of the surrounding vehicle 301 using the recursive network 200 and to output the updated future trajectory as the second prediction data (Yfinal)” (Kum, page 19, paragraph 0078). Examiner notes that the output state space is the future trajectory and the location space is the future trajectory of the surrounding vehicles based on the latitudinal and longitudinal movements.)
On page 19, Applicant argues:
Garsdal fails to anticipate dependent claims 9-11 at least by virtue of their dependency on amended independent claim 8.
Regarding the Applicant’s argument that the dependent claims are allowable at least due in part to their dependency on the independent claims, the Examiner respectfully disagrees and notes the instant rejections and response to arguments regarding the independent claims above.
On page 20, Applicant argues:
Independent claims 1 and 14 recite analogous features as independent claim 8 and are therefore not rendered obvious over the teachings of Garsdal, Kum, and Blaiotta, either alone or in combination, at least for the same reasons as presented above for independent claim 8.
Accordingly, amended independent claims 1, 8, and 14 are allowable.
Dependent claims 2-4, 6, 7, 9-13, and 15-19 are also allowable at least due to their dependence on amended independent claims 1, 8, and 14. The rejection pertaining to claim 5 is rendered moot in view of its cancellation.
Reconsideration and allowance of the claims 1-4 and 6-19 is respectfully requested.
Regarding the Applicant’s argument that claims 1 and 14 are allowable for the same reasons as claim 8, Examiner respectfully disagrees and notes that claims 1 and 14 are rejected for similar reasons as claim 8. Regarding the Applicant’s argument that the dependent claims are allowable at least due in part to their dependency on the independent claims, the Examiner respectfully disagrees and notes the instant rejections and response to arguments regarding the independent claims above.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Chen et al. (“Neural Ordinary Differential Equations”) also discusses ODE used in neural networks and decoders.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KAITLYN R LAU whose telephone number is (571)272-1429. The examiner can normally be reached Monday - Thursday: 7:15 am - 5:15 pm EST.
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, Michelle Bechtold can be reached at (571) 431-0762. 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.
/K.R.L./Examiner, Art Unit 2148 /MICHELLE T BECHTOLD/Supervisory Patent Examiner, Art Unit 2148
Prosecution Insights