DETAILED ACTION
This action is responsive to the claims filed on 09/03/2025. Claims 16-23 are pending for examination.
Response to Arguments
In response to Applicant’s arguments regarding the rejection of claim 16 under 35 U.S.C. § 112, the Examiner agrees and the 112 rejection has been withdrawn
In response to Applicant’s arguments regarding the rejection of claim 16 under 35 U.S.C. § 101, the Examiner respectfully disagrees. Although Applicant has amended claim 16 to further recite “training a neural network …” and “using the trained neural network …,” the claim as a whole still recites a judicial exception in the form of mathematical concepts/mental processes (e.g., computing emission values with a model, numerically altering inputs, and generating synthetic measurement series based on a relative change). The additional recitations of “training” and “using” a neural network do not integrate the exception into a practical application because they merely add generic instructions to apply the abstract calculations using a neural network at a high level of generality, without reciting any particular machine implementation, technical improvement to computer functionality, or other meaningful limitation that confines the claim to a specific technological solution. Rather, the claim continues to focus on manipulating data and performing calculations to produce synthetic training data and predicted outputs—activities that can be performed conceptually and that amount to applying the exception using generic computer/ML components. As such, the claim fails Step 2A, Prong 2. Further, the additional limitations do not amount to “significantly more” under Step 2B, because they do not add an inventive concept beyond using well-understood, routine, conventional neural-network training/inference to carry out the abstract mathematical operations. Accordingly, the §101 rejection is maintained.
In response to Applicant’s arguments regarding the rejection of claim 16 under 35 U.S.C. §103, the Examiner respectfully disagrees with Applicant’s assertion that the cited references, alone or in combination, fail to teach the amended claim language. Applicant’s statement that “none of the references” teach “using the neural network as recited” is not persuasive because the applied references expressly disclose the specific training and inference/use of a neural-network pollutant-prediction model, and the remaining limitations are taught by the combination as mapped. For example, with respect to the newly added limitations of “training a neural network to determine a pollutant concentration based on an input pollutant emission” and “using the trained neural network to determine a pollutant concentration in response to a user input pollutant emission,” Xipeng explicitly discloses creating/training a neural network model using plume-model based data (i.e., “Gaussian plume diffusion model-based experimental data were used to create a neural network model. The first 200 groups of data were used for a drill…” (Xipeng, p. 67, col. 2, Sec. 2.1)), which teaches training a neural network model for pollutant concentration prediction using plume-model derived training data (where plume models inherently incorporate source/emission terms as inputs to compute concentration). Xipeng further discloses applying the trained model to perform prediction at new/unknown conditions (“…available air pollution concentration data were used to predict the pollutant concentration at the unknown positions” (Xipeng, p. 67, col. 1, paragraph 2)), which teaches using the trained neural network for inference to output pollutant concentration in response to the provided input values (including source/emission-related inputs as part of the plume-model input set). Additionally, to the extent Applicant alleges the applied art does not teach the “threshold” first measurement series, Sharp explicitly teaches triggering acquisition/collection based on a predetermined criterion including a threshold value, thereby ensuring a measurement/sampling series includes values above a defined threshold (“…a control unit for generating a grab sample command in response to the acquired sensor data meeting a predetermined criteria, and a grab sample unit for acquiring an air sample…” (Sharp, col. 6, lines 13–20) and “The criteria for triggering a grab Sample may include concentrations… above a predetermined threshold value” (Sharp, col. 20, lines 33–37)), which directly corresponds to providing a measurement series containing a measured value above a defined threshold. Finally, regarding Applicant’s broader position that the combination does not teach the remaining computational/augmentation steps, Xipeng teaches computing pollutant emission values from measured values via an inversion/optimization framework (“…Gaussian plume diffusion model… quadratic optimization model and inversion method for inverse calculation…” (Xipeng, Abstract)), and Goodfellow teaches numerically altering inputs to generate perturbed examples (“…fast gradient sign method…” (Goodfellow, page. 3, paragraph 2)) and using such perturbed examples as a continually updated supply of adversarial/synthetic training examples (“continually update our supply of adversarial examples…” (Goodfellow, page. 5, paragraph 1)), which correspond to the claimed numerical alteration and synthetic measurement series generation steps as mapped. Accordingly, when the references are properly considered together as set forth in the rejection and mapping, the cited combination teaches or renders obvious each limitation of claim 16, and Applicant has not identified a specific deficiency in these teachings. Therefore, the §103 rejection is maintained.
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 16-23 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Statutory Categories
Claims 16-23 are directed to a method.
Independent Claims – Claims 16
Step 2A Prong 1: Does the claim recite an abstract idea, law of nature, or natural phenomenon?
Yes. Independent claims 16 and 24 recites limitations that are abstract ideas in the form of mental processes:
Claim 16 recites:
computing a first value Eo of the pollutant emission with the model using a value of the physical measured variable related to a value Co of the provided measured pollutant concentration; (this limitation merely amounts to performing mathematical calculations, algorithms, or formulas at least because emission models are mathematical models.)
computing a second value EI of the pollutant emission with the model by numerically altering the measured value of the physical measured variable used for computing the first value Eo of the pollutant emissions; (this limitation merely amounts to performing mathematical calculations, algorithms, or formulas with the numerical alteration being interpreted as a mental process of evaluation which can reasonably be permed in human mind or with aid of pen and paper)
and generating a synthetic measurement series as training data using an alteration of the value Co of the physical measured measurement series of the pollutant concentrations, wherein the alteration depends on the relative change zE/Eo in the computed values of the pollutant emissions. (this limitation merely amounts to performing mathematical calculations, algorithms, or formulas (See specification pages 9, 14, and 20 for the related mathematical disclosure))
Claim 16 also recites the following additional elements for the purposes of Step 2A Prong Two analysis:
A method for determining a pollutant concentration from a pollutant emission, the method comprising: (the recitation of a neural network used at a high level of generality is being considered as mere instructions to apply an exception, see MPEP 2106.05(f))
providing a first measurement series of the pollutant concentration containing a measured value above a defined threshold value; (providing data is mere data gathering/outputting and is considered insignificant extra-solution activity. See MPEP 2106.05(g))
providing a second measurement series for a physical measured variable related to the measured pollutant concentration; (providing data is mere data gathering/outputting and is considered insignificant extra-solution activity. See MPEP 2106.05(g))
providing a model for a relationship between the physical measured variable and the pollutant emission; (providing data is mere data gathering/outputting and is considered insignificant extra-solution activity. See MPEP 2106.05(g))
training a neural network to determine a pollutant concentration based on an input pollutant emission; (the recitation of a neural network used at a high level of generality is being considered as mere instructions to apply an exception under step 2A prong 2, see MPEP 2106.05(f))
and using the trained neural network to determine a pollutant concentration in response to a user input pollutant emission. (the recitation of a neural network used at a high level of generality is being considered as mere instructions to apply an exception under step 2A prong 2, see MPEP 2106.05(f))
The additional limitations fail step 2A Prong 2 of the 101 analysis because they do not transform the claim into a practical application. These limitations are too abstract or lack technical improvement that would make the concept practically useful. Without clear utility or integration into a specific field, the claim does not relate to any particular application. It does not meet the requirements of Step 2A Prong 2, as it fails to make the concept meaningfully applicable in practice.
Since the claim as a whole, looking at the additional elements individually and in combination, does not contain any other additional elements that are indicative of integration into a practical application, the claim is “directed” to an abstract idea.
This claim recites the following additional elements for the purposes of Step 2B analysis:
A method for determining a pollutant concentration from a pollutant emission, the method comprising: (the recitation of a neural network used at a high level of generality is being considered as mere instructions to apply an exception, see MPEP 2106.05(f))
providing a first measurement series of the pollutant concentration containing a measured value above a defined threshold value; (providing data is mere data gathering/outputting and is considered insignificant extra-solution activity. See MPEP 2106.05(g), it should be noted that the courts have recognized receiving or transmitting data over a network as well-understood, routine, and conventional activity. (See MPEP 2106.05(d)(ii) and Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information);))
providing a second measurement series for a physical measured variable related to the measured pollutant concentration; (providing data is mere data gathering/outputting and is considered insignificant extra-solution activity. See MPEP 2106.05(g), it should be noted that the courts have recognized receiving or transmitting data over a network as well-understood, routine, and conventional activity. (See MPEP 2106.05(d)(ii) and Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information);))
providing a model for a relationship between the physical measured variable and the pollutant emission; (providing data is mere data gathering/outputting and is considered insignificant extra-solution activity. See MPEP 2106.05(g), it should be noted that the courts have recognized receiving or transmitting data over a network as well-understood, routine, and conventional activity. (See MPEP 2106.05(d)(ii) and Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information);))
training a neural network to determine a pollutant concentration based on an input pollutant emission; (the recitation of a neural network used at a high level of generality is being considered as mere instructions to apply an exception under Step 2B, see MPEP 2106.05(f))
and using the trained neural network to determine a pollutant concentration in response to a user input pollutant emission. (the recitation of a neural network used at a high level of generality is being considered as mere instructions to apply an exception under Step 2B, see MPEP 2106.05(f))
The claim also fails Step 2B of the analysis because the additional limitations do not amount to significantly more than the abstract idea itself. The additional limitations do not enhance the claim in a way that would move it beyond its abstract ideas as they minimally elaborate on the core concept without adding any inventive or technical substance. Considering the additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
Dependents of Claim 16
The remaining dependent claims corresponding to independent claim 16 do not recite additional elements, whether considered individually or in combination, that are sufficient to integrate the judicial exception into a practical application or amount to significantly more than the judicial exception. The analysis of which is shown below:
The claims below recite additional limitations which fail step 2A Prong 2 of the 101 analysis because they do not transform the claim into a practical application. These limitations are too abstract or lack technical improvement that would make the concept practically useful. Without clear utility or integration into a specific field, the claim does not relate to any particular application. It does not meet the requirements of Step 2A Prong 2, as it fails to make the concept meaningfully applicable in practice.
The claims also fails Step 2B of the analysis because the additional limitations do not amount to significantly more than the abstract idea itself. The additional limitations do not enhance the claim in a way that would move it beyond its abstract ideas as they minimally elaborate on the core concept without adding any inventive or technical substance. The claims are unpatentable.
Claim 17 recites the additional limitation of:
The computer aided method as claimed in claim 16, wherein the alteration Ac of the value Co of the provided measurement series of the pollutant concentrations is additionally depends on a traffic-related proportion a of the pollutant concentration. (this limitation merely amounts to performing mathematical calculations, algorithms, or formulas (See specification pages 9, 14, and 20 for the related mathematical disclosure))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 18 recites the additional limitation of:
The computer aided method as claimed in claim 17, wherein the alteration Ac of the value Co of the provided measurement series of the pollutant concentration depends on AC/Co= αAE/Eo. (this limitation merely amounts to performing mathematical calculations, algorithms, or formulas (See specification pages 9, 14, and 20 for the related mathematical disclosure))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 19 recites the additional limitation of:
The computer aided method as claimed in claim 17, wherein the traffic- related proportion a is in the range from 0.3 to 0.5. (a mathematical concept including mathematical formulas, algorithms, or calculations)
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 20 recites the additional limitation of:
The computer aided method as claimed in claim 16, wherein the pollutant concentration is a nitrogen oxide concentration and the pollutant emission is a nitrogen oxide emission. (Under both step 2A prong II and step 2B this limitation is generally linking the use of a judicial exception to a particular technological environment or field, see MPEP 2106.05(h))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 21 recites the additional limitation of:
The computer aided method as claimed in claim 16, wherein the physical measured variable used is a temperature, a wind speed, and/or a traffic level. (Under both step 2A prong II and step 2B this limitation is generally linking the use of a judicial exception to a particular technological environment or field, see MPEP 2106.05(h))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 22 recites the additional limitation of:
The computer aided method as claimed in claim 16, further comprising capturing the measurement series of the pollutant concentration and the measurement series of the measured variable using a measurement station within a town. (Step 2A Prong II: providing data is mere data gathering/outputting and is considered insignificant extra-solution activity. See MPEP 2106.05(g). For Step 2B, it should be noted that the courts have recognized receiving or transmitting data over a network as well-understood, routine, and conventional activity. (See MPEP 2106.05(d)(ii) and Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information);))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 23 recites the additional limitation of:
The computer aided method as claimed in claim 16, wherein the model is a domain-based model. (For the purposes of Step 2A prong 2 and Step 2B: the recitation of a domain based model used at a high level of generality is being considered as mere instructions to apply an exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
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 text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
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 non-obviousness.
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 16, 18, and 21-23 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al. (US 2016/0125307 A1), hereafter referred to as Zhang, in view of Sharp et al., (US 7302313 B2), hereafter referred to as Sharp, and in further view ofXipeng et al. (Xipeng, Z., Shunsheng, Y., Wenchuan, X., & Yu, C. (2018). RBF Neural Network-Based Prediction and Inverse Calculation of Air Pollutant Emission Concentration. American Journal of Biological and Environmental Statistics, 4(2), 66-73.), hereafter referred to as Xipeng, Goodfellow et al. (Goodfellow, I. J., Shlens, J., & Szegedy, C. (2014). Explaining and harnessing adversarial examples. arXiv preprint arXiv:1412.6572.), hereafter referred to as Goodfellow, and Furukawa et al. (Furukawa, M., Matsuyama, A., & Ohkawa, Y. (2016). High-Accuracy Numerical Integration of Charged Particle Motion–with Application to Ponderomotive Force. Plasma and Fusion Research, 11, 1303003-1303003.), hereafter referred to as Furukawa.
Claim 16: Zhang teaches:
A method for determining a pollutant concentration from a pollutant emission, the method comprising: (Zhang, paragraph 23, “During the inference stage 108, the trained spatial classifier 116 and the trained temporal classifier 120 may be used to infer AQIs for areas in the region that do not have air quality monitor stations. The trained spatial classifier 116 may be applied to the spatial features that are extracted from observed data 122 for each area in the region to generate a corresponding spatial probability score. Likewise, the trained temporal classifier 120 may be applied to the temporal features that are extracted from the observed data 122 for each area in the region to generate a corresponding temporal prob ability score. The observed data 122 may include real-time spatially-related data and real-time temporally-related data.”, Zhang discloses a computer method involving a neural network model that uses emissions data (observed data) and predicts pollutant concentration (Air-Quality Index (AQI)).)
providing a second measurement series for a physical measured variable related to the measured pollutant concentration; (Zhang, paragraph 3, “The other data sources may provide meteorological data, traffic flow data, human mobility data, road structure data, points of interest data, and/or so forth.”,
Zhang, paragraph 4, “The temporal classifier may use temporally-related features, such as traffic flow data and meteorological data, to discover the temporal dependency of air qualities at different areas.”, Zhang’s “meteorological data” is a physical measured variable because it corresponds to measurable physical quantities (e.g. wind speed, wind direction, temperature, etc.) collected from sensors. Such physical measurements are related to measured pollutant concentration because they influence diffusion/dispersion and are used together with concentration measurements in pollution inference.)
providing a model for a relationship between the physical measured variable and the pollutant emission; (Zhang, paragraph 5, “The co-training framework may generate inference models, i.e., classifiers, which are used to interpolate air qualities for additional areas based on a limited set of measured air quality data from a small number of areas. The models may be used to infer the air quality of the additional areas based on real-time air quality data from existing air quality monitor stations and other forms of collected spatial or temporal data.”)
Sharp, in the same field of air quality tracking, teaches the following limitations which Zhang fails to teach:
providing a first measurement series of the pollutant concentration containing a measured value above a defined threshold value; (Sharp, col. 6, lines 13-20, “an air quality monitoring system comprises at least one air quality sensor for acquiring sensor data at a selected indoor location, a control unit for generating a grab sample command in response to the acquired sensor data meeting a predetermined criteria, and a grab sample unit for acquiring an air sample at the selected indoor location in response to the grab sample command from the control unit.”,
Sharp, col. 20, lines 33-37, “The criteria for triggering a grab Sample may include concentrations above the average concentration, above a predetermined threshold value,”, Sharp collects air samples (including air concentrations) when the acquired sensor data meet a predetermined threshold.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated the teachings of Sharp into that of Zhang, because Zhang relies on monitor-station measurement series for air-quality modeling/inference, and Sharp teaches a well-known technique for triggering acquisition/collection of air samples (and thus pollutant concentration measurements) when sensed air-quality data meets predetermined criteria (i.e., exceeds a defined threshold). A motivation for combining would have been to improve the relevance and efficiency of Zhang’s measurement-series collection by ensuring the collected/used concentration series includes pollutant events of interest (e.g., exceedances above a defined threshold), thereby focusing training/inference on higher-impact pollution episodes and reducing unnecessary sampling/processing when conditions are normal. Sharp supports this motivation by expressly disclosing “a control unit for generating a grab sample command in response to the acquired sensor data meeting a predetermined criteria” (Sharp, col. 6, lines 13-20).
Xipeng, in the same field of pollution inference, teaches the following limitation which the above fails to teach:
computing a first value Eo of the pollutant emission with the model using a value of the physical measured variable related to a value Co of the provided measured pollutant concentration; (Xipeng, abstract, “According to the measured values and the predicted data, Gaussian plume diffusion model for air pollution was created, and the quadratic optimization model and inversion method for inverse calculation of single pollution source and multi pollution source were built.”, the measured pollutant values (C) are used with the inference model to compute a pollution source (the emission value E) from those concentrations.)
training a neural network to determine a pollutant concentration based on an input pollutant emission; (Xipeng, page 67, col. 2, section 2.1, paragraph 3, “The Gaussian plume diffusion model-based experimental data were used to create a neural network model. The first 200 groups of data were used for a drill… The comparative simulation results from RBF neural network-based data prediction and BP neural network-based data prediction were shown in Figure 1~ 3.”, Xipeng states that “Gaussian plume diffusion model-based experimental data were used to create a neural network model” and that a defined portion of that data (“first 200 groups”) was used for a “drill,” i.e., training, and it further describes “RBF neural network-based data prediction” results. This teaches training a neural network on plume-model data to perform pollutant concentration prediction. Because a Gaussian plume diffusion model computes pollutant concentration as a function of source terms (i.e., emission/source intensity) and physical variables, the plume-model-based training dataset inherently includes emission/source-related input parameters corresponding to the concentration outputs, thereby teaching training the neural network to determine pollutant concentration based on an input pollutant emission.)
using the trained neural network to determine a pollutant concentration in response to a user input pollutant emission. (Xipeng, page 67, col. 1 paragraph 2, “To solve the problem, the available air pollution concentration data were used to predict the pollutant concentration at the unknown positions”, Xipeng explains that “available air pollution concentration data were used to predict the pollutant concentration at the unknown positions,” which is an inference/use step: the trained prediction model is applied to input values to output predicted pollutant concentration for new/unknown conditions. In the context of Xipeng’s plume-model-based framework (used to create/train the neural network model), supplying the relevant model inputs, including source/emission-related measurements, corresponds to providing an input pollutant emission (here, via an input value provided to the model), so the reference teaches using the trained neural network to output pollutant concentration in response to an input emission value.)
Zheng discloses a neural-network framework that takes pollutant emissions, meteorological parameters, and other measured variables as inputs to predict pollutant concentrations via co-trained spatial and temporal classifiers. In particular, Zheng et al. describe extracting spatial and temporal features from multiple sensor streams and feeding them into a neural network to output inferred air-quality indices and concentration values. Xi-Peng then shows how, “according to the measured values and the predicted data,” a Gaussian-plume diffusion model can be coupled with a quadratic optimization and inversion method to directly compute pollutant emission rates from measured concentration data. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Zhang and Sharp by incorporating the teachings of Xipeng to perturb input of the models training data. A motivation of which would be to use RBF (Radial Basis Function) neural networks to infer pollutant indexes, achieving higher accuracy when compared to other neural networks like BP (Back-propagation), (Goodfellow, page 5, paragraph 5, “The adversarial training procedure can be seen as minimizing the worst case error when the data is perturbed by an adversary. That can be interpreted as learning to play an adversarial game, or as minimizing an upper bound on the expected cost over noisy samples with noise from U(−ε, ε) added to the inputs. Adversarial training can also be seen as a form of active learning, where the model is able to request labels on new points. In this case the human labeler is replaced with a heuristic labeler that copies labels from nearby points.”).
Goodfellow, in the same field of data augmentation, teaches the following limitations which the above fails to teach:
computing a second value EI of the pollutant emission with the model by numerically altering the measured value of the physical measured variable used for computing the first value Eo of the pollutant emissions; (Goodfellow, page 3, paragraph 2, “Let θ be the parameters of a model, x the input to the model, y the targets associated with x (for machine learning tasks that have targets) and J(θ, x, y) be the cost used to train the neural network. We can linearize the cost function around the current value of θ, obtaining an optimal max-norm constrained perturbation of
PNG
media_image1.png
24
151
media_image1.png
Greyscale
We refer to this as the “fast gradient sign method” of generating adversarial examples. Note that the required gradient can be computed efficiently using backpropagation.”, Goodfellow defines an adversarial perturbation, η, which is an explicit alternation of the measured input vector x by taking the sign of the model’s gradient.)
and generating a synthetic measurement series as training data using an alteration of the value Co of the first measured measurement series of the pollutant concentrations, (Goodfellow, page 5, paragraph 1, “We found that training with an adversarial objective function based on the fast gradient sign method was an effective regularizer:
PNG
media_image2.png
22
371
media_image2.png
Greyscale
In all of our experiments, we used α = 0.5. Other values may work better; our initial guess of this hyperparameter worked well enough that we did not feel the need to explore more. This approach means that we continually update our supply of adversarial examples, to make them resist the current version of the model.”, Goodfellow synthesizes new training examples by combining the original input x (the values of C) with perturbed inputs. These perturbed inputs function analogously to a synthetic measurement series by altering Co.)
Zheng teaches using a neural network to predict pollutant concentrations from emission data, and Xipeng demonstrates inverting measured concentration values into emission rates via a Gaussian‐plume/RBF neural-network framework . Goodfellow uses the “fast gradient sign method” for numerically perturbing inputs, and the mixed-loss objective to synthesize new training examples from original and perturbed inputs. A person of ordinary skill in the art seeking to improve pollutant-emission prediction would have been motivated to apply these adversarial‐perturbation and synthetic-data techniques to the measured concentration inputs taught by Zheng, Sharp, and Xipeng—thereby generating perturbed concentration series (C₁), computing their corresponding emission outputs (E₁), and producing a synthetic measurement series, with each perturbation inherently tied to the model’s relative output change ΔE/E₀. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Zhang by incorporating the teachings of Goodfellow to perturb input of the models training data. A motivation of which would be to use adversarial examples to further minimize the worst case error and provide a form of active learning, (Goodfellow, page 5, paragraph 5, “The adversarial training procedure can be seen as minimizing the worst case error when the data is perturbed by an adversary. That can be interpreted as learning to play an adversarial game, or as minimizing an upper bound on the expected cost over noisy samples with noise from U(−ε, ε) added to the inputs. Adversarial training can also be seen as a form of active learning, where the model is able to request labels on new points. In this case the human labeler is replaced with a heuristic labeler that copies labels from nearby points.”).
Furukawa, in the same field of input perturbation, teaches the following limitation which the above fails to teach:
wherein the alteration depends on the relative change (ΔE/Eo) in the computed values of the pollutant emissions. (Furukawa, page 1303003, col. 1, paragraph 2, “The relative change of energy was measured by
PNG
media_image3.png
44
229
media_image3.png
Greyscale
”, Furukawa expressly defines and uses a normalized “relative change” metric in the form “ΔE/E0” by stating that “the relative change of energy was measured by” that ratio. This teaches the specific mathematical concept required by the claim: an alteration/condition that depends on a relative change expressed as a difference (ΔE) normalized by a baseline value (E0). Applying this conventional relative-change form to the computed emission values (E0 and Ei) taught elsewhere in the combination provides the claimed dependency on relative change (ΔE/E0) in the computed output quantity.)
It would have been obvious to one of ordinary skill in the art before the effective filing date to have modified the teachings of Zhang/Xipeng/Sharp’s prediction framework and Goodfellow’s perturbation-based synthetic data generation by incorporating Furukawa’s explicit use of a normalized relative-change metric (ΔE/E0) as a way to quantify output sensitivity and scale/condition perturbations accordingly. Furukawa expressly defines and uses “ΔE/E0” as a measured relative change ; furthermore, Goodfellow teaches numerically perturbing inputs to generate adversarial/synthetic examples via gradient-based perturbation η. A POSITA would have been motivated to use a normalized relative-change ratio (ΔE/E0) to govern how perturbations are applied (e.g., to normalize across magnitudes and focus perturbations where they meaningfully affect outputs), thereby arriving at “wherein the alteration depends on the relative change ΔE/E0 in the computed values…” (Furukawa, page 13003, col. 1, paragraph 2).
Claim 18: Zhang, Sharp, Xipeng, Goodfellow, and Furukawa teaches the limitations of claim 16, Goodfellow further teaches:
The computer aided method as claimed in claim 17, wherein the alteration Ac of the value Co of the provided measurement series of the pollutant concentration depends on ΔC/Co= αΔE/Eo. (Goodfellow, page 3, paragraph 2, “Let θ be the parameters of a model, x the input to the model, y the targets associated with x (for machine learning tasks that have targets) and J(θ, x, y) be the cost used to train the neural network. We can linearize the cost function around the current value of θ, obtaining an optimal max-norm constrained pertubation of
PNG
media_image1.png
24
151
media_image1.png
Greyscale
We refer to this as the “fast gradient sign method” of generating adversarial examples. Note that the required gradient can be computed efficiently using backpropagation.”, Here, η (the alteration Ac) is computed by taking the sign of the gradient ∇ₓ J, which tells exactly how much and in what direction the model’s output J will change for a small tweak in x. It is interpreted that the change ΔJ is ΔE of the claim.
Goodfellow, page 5, paragraph 1, “We found that training with an adversarial objective function based on the fast gradient sign method was an effective regularizer:
PNG
media_image2.png
22
371
media_image2.png
Greyscale
In all of our experiments, we used α = 0.5. Other values may work better; our initial guess of this hyperparameter worked well enough that we did not feel the need to explore more. This approach means that we continually update our supply of adversarial examples, to make them resist the current version of the model.”, By blending the original cost J(x) with the perturbed cost J(x+η), they explicitly use the difference J(x+η)–J(x) (ΔE) relative to J(x) (E₀) to determine η. In other words, η is chosen as a direct function of the ratio ΔE/E₀.)
Claim 21: Zhang, Sharp, Xipeng, Goodfellow, and Furukawa teaches the limitations of claim 16, Zhang further teaches:
The computer aided method as claimed in claim 16, wherein the physical measured variable used is a temperature, a wind speed, and/or a traffic level. (Zhang, paragraph 37, “The temporal feature extraction module 212 may extract temporal features from various environmental or human movement data of the unlabeled source data 112. The concentration of air pollutants may be influenced by meteorology. Accordingly, the temporal feature extraction module 212 may identify, for example, five categories of features: temperature, humidity, barometric pressure, wind speed, and weather (Such as cloudy, foggy, rainy, Sunny, and Snowy). For example, high wind speed may disperse the concentration of PM, and high humidity may result in a high concentration of Mo. High pressure may result in a relative good AQI. A relatively good AQI may also be present when pressure is high and temperature is low.”)
Claim 22: Zhang, Sharp, Xipeng, Goodfellow, and Furukawa teaches the limitations of claim 16, Zhang further teaches:
The computer aided method as claimed in claim 16, further comprising capturing the measurement series of the pollutant concentration and the measurement series of the measured variable using a measurement station within a town. (Zhang, paragraph 6, “In at least one embodiment, labeled air quality index data for a pollutant in a region may be obtained from one or more air quality monitor stations.”)
Claim 23: Zhang, Sharp, Xipeng, Goodfellow, and Furukawa teaches the limitations of claim 16, Zhang further teaches:
The computer aided method as claimed in claim 16, wherein the model is a domain-based model. (Zhang, paragraph 18, “The techniques may use a semi-supervised learning approach based on a co-training framework that trains two separate classifiers, such as a spatial classifier and a temporal classifier.”, By splitting learning into distinct spatial and temporal “domains” and then combining them, Zheng teaches a domain-based modeling approach.)
Claims 17 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang, Sharp, Xipeng, Goodfellow, and Furukawa in further view of Antoniou et al. (Antoniou, C., Azevedo, C. L., Lu, L., Pereira, F., & Ben-Akiva, M. (2015). W–SPSA in practice: Approximation of weight matrices and calibration of traffic simulation models. Transportation Research Procedia, 7, 233-253.), hereafter referred to as Antoniou.
Claim 17: Zhang, Sharp, Xipeng, Goodfellow, and Furukawa teaches the limitations of claim 16, Antoniou in the same field of traffic analysis, teaches the following which the above fails to teach:
The computer aided method as claimed in claim 16, wherein the alteration
A
c of the value Co of the provided measurement series of the pollutant concentrations is additionally depends on a traffic-related proportion α of the pollutant concentration. (Antoniou, page 238, last paragraph, “In each iteration, the gradient estimation process essentially tries to find a direction and amplitude for each parameter value in the decision vector to move. This is achieved by comparing the influence to the system caused by perturbing each of the parameter value in two opposite directions.”, in Antoniou, the W-SPA algorithm perturbs the input parameters iteratively by estimating a gradient that determines both the direction and magnitude of each perturbation.
Antoniou, page 248, last paragraph, “Regarding the specification of the objective function, the OLS assumption was assumed and the variance-covariance matrices ΩM and Ωθ are specified as a block diagonal matrix with ΩMk and Ωβ as its diagonal elements, respectively. The assumed fixed weights of the optimizing function are ΩMcounts k = 0.3, ΩMspeeds k = 0.5 and Ωβ = 0.2. These values were defined previously, based on the contribution of each information on the calibration process. As we focus on detailed traffic statistics a higher contribution was given to speed related data. A sensitivity analysis on these weight values may, however, enhance the calibration final results.”, the gradient estimation (which the perturbations are based on) are no uniform, it is guided by fixed weights (with weights between 0.3-0,5) assigned to different traffic related data streams (e.g. traffic count and speeds). These weights influence the objective function, and thereby directly affecting the computed gradients that determines how the model input parameters are perturbed.)
Zheng teaches a neural‐network framework that takes measured pollutant emissions, meteorological data, and other variables as inputs to predict pollutant concentrations via co-trained spatial and temporal classifiers. Antoniou, in turn, teaches traffic related coefficients that affect perturbation generation. It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the neural-network inputs of Zheng, Sharp, Xipeng, Goodfellow, and Furukawa by incorporating Antoniou’s traffic coefficients to the perturbation generation process. The clear motivation for this combination is to provide a way to generate input perturbations using calibrated traffic-aware adjustments. (Antoniou, page 240, last paragraph, “The accurate estimation of weight matrices is the key to successfully applying W–SPSA for the off–line calibration of traffic estimation models. The ideal weight matrix estimation approach should be easy to implement, efficient to execute, and able to provide accurate estimation for different types of parameters and measurements.”, having calibrated traffic related weights between 0.3-0.5 allow for accurate training of this model.)
Claim 19: Zhang, Sharp, Xipeng, Goodfellow, and Antoniou teaches the limitations of claim 17, Antoniou further teaches:
The computer aided method as claimed in claim 17, wherein the traffic- related proportion a is in the range from 0.3 to 0.5. (Antoniou, page 248, last paragraph, “The assumed fixed weights of the optimizing function are ΩMcounts k = 0.3, ΩMspeeds k = 0.5 and Ωβ = 0.2.”, the weights (with values between 0.3-0.5) are used to perturb the input parameters as disclosed above in claim 17.).)
The rationale for the combination of Zhang with Negrenti is substantially similar to that applied for claim 17 above.
Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Zhang, Sharp, Xipeng, Goodfellow, and Furukawa, in further view of Forehead et al. (Forehead, H., & Huynh, N. (2018). Review of modelling air pollution from traffic at street-level-The state of the science. Environmental Pollution, 241, 775-786.), hereafter referred to as Forehead.
Claim 20: Zhang, Sharp, Xipeng, Goodfellow, and Furukawa teaches the limitations of claim 16, Forehead in the same field of pollution analysis, teaches the following which the above fails to teach:
The computer aided method as claimed in claim 16, wherein the pollutant concentration is a nitrogen oxide concentration (Forehead, page 776, col. 1, paragraph 2, “Diesel vehicles produce most of the particles of 2.5 microns and smaller (PM2.5) and oxides of nitrogen (NOx).”)
and the pollutant emission is a nitrogen oxide emission. (Forehead, page 776, col. 2, paragraph 2, “The composition of the mixture of gases and particles changes with time after release from the exhaust pipe. For example, the concentrations of particular species, such as NOx, can determine the production of secondary pollutants such as ozone”)
Zheng teaches a neural‐network based scheme for inferring air quality indexes by co‐training spatial and temporal classifiers on sensor‐derived concentration inputs. However, it does not address how to select or weight specific pollutant species like nitrogen oxide. Forehead highlights the critical role of identifying NOₓ in traffic‐related air pollution. It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to have combined the teachings of Zheng, Sharp, Xipeng, Goodfellow, and Furukawa with Forehead (i.e. identifying NO levels in emission/concentrations). A motivation would have been to provide a way to track specific environmental harming pollutants. (Forehead, page 776, col. 2, paragraph 3, “Although numbers of vehicles on roads continue to increase, emissions regulations have mandated increased efficiency of engine technologies to reduce outputs of harmful emissions. Older, carbureted cars released 10 times the HC, 4 times the CO and 3 times the NOx of newer multi-point ignition engines (Qu et al., 2015). However, while newer cars release less pollution, the expected reduction in emissions from modern vehicles will only be realized if their emissions control equipment is properly maintained”, tracking specific emissions like NOx provides further detailed analysis of the pollution concentration of an environment.)
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Negrenti, E. (1999). The ‘corrected average speed’approach in ENEA’s TEE model: an innovative solution for the evaluation of the energetic and environmental impacts of urban transport policies. Science of the total environment, 235(1-3), 411-413.
Li, X., Peng, L., Yao, X., Cui, S., Hu, Y., You, C., & Chi, T. (2017). Long short-term memory neural network for air pollutant concentration predictions: Method development and evaluation. Environmental pollution, 231, 997-1004.
Freeman, B. S., Taylor, G., Gharabaghi, B., & Thé, J. (2018). Forecasting air quality time series using deep learning. Journal of the Air & Waste Management Association, 68(8), 866-886.
Bui, T. C., Le, V. D., & Cha, S. K. (2018). A deep learning approach for forecasting air pollution in South Korea using LSTM. arXiv preprint arXiv:1804.07891.
Stockie, J. M. (2011). The mathematics of atmospheric dispersion modeling. Siam Review, 53(2), 349-372.
Tramèr, F., Kurakin, A., Papernot, N., Goodfellow, I., Boneh, D., & McDaniel, P. (2017). Ensemble adversarial training: Attacks and defenses. arXiv preprint arXiv:1705.07204.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HYUNGJUN B YI whose telephone number is (703)756-4799. The examiner can normally be reached M-F 9-5.
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, Usmaan Saeed can be reached on (571) 272-4046. 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.
/H.B.Y./Examiner, Art Unit 2146
/USMAAN SAEED/Supervisory Patent Examiner, Art Unit 2146