DETAILED ACTION
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 .
Status of the Claims
The office action is in response to the claims filed on February 12, 2026, for the application filed on October 30, 2023, which is a national stage of International Application No. PCT/EP2021/062977 filed on May 17, 2021. In Applicant’s Response filed October 1, 2025, claims 1 – 2, 14 – 15, and 18 were amended, claim 9 was cancelled, and new claims 19 – 21 were added. Claims 1 – 8 and 10 – 21 are currently pending and have been examined as discussed below.
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 set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
Determining the scope and contents of the prior art.
Ascertaining the differences between the prior art and the claims at issue.
Resolving the level of ordinary skill in the pertinent art.
Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 10 – 15, and 18 – 20 are rejected under 35 U.S.C. 103(a) as being unpatentable over Moridi (U.S. Pub. No. 2021/0192356 A1) in view of Achin (U.S. Pub. No. 2022/0199266 A1), NPL Akmandor, and Sobol (U.S. Pub. No. 2019/0209022 A1).
Regarding independent claims 1, 14, and 15, Moridi teaches the limitations of representative claim 14 identified in bold as:
A system for training a machine learning model for monitoring of a proneness of an indoor environment to infectious disease transmission, the system comprising a functional unit having one or more computational processors with access to memory, which, alone or in combination, are configured to provide for execution of the following steps (Paragraphs [0003] and [0081] of Moridi. In the instant application, the broadest reasonable interpretation of “the system comprising a functional unit having one or more computational processors with access to memory, which, alone or in combination, are configured to provide for execution of the following steps” reads on the system of Moridi (Paragraphs [0003] and [0081]) including the non-transitory machine-readable storage medium storing instructions that may be read and executed by at least one processor to perform the operations described in detail herein.).
obtaining unlabeled sensor data from sensors of the indoor environment in order to provide a set of sensor features (Paragraphs [0023] and [0056] and FIGS. 1 – 3 and 5 of Moridi. In the instant application, the broadest reasonable interpretation of “obtaining unlabeled … data” reads on the step in Moridi (Paragraph [0023]) of providing a weak supervision setting that utilizes unlabeled data Xu=x1, … , xN.);
generating, by applying situation labeling functions, a labeling matrix that is fed to a generative model, wherein the generative model feeds a discriminative classifier model with probabilistic labels for the sensor features, wherein the probabilistic labels of the generative model are used for training the discriminative classifier model (Paragraphs [0042] – [0044] of Moridi. In the instant application, the broadest reasonable interpretation of “generating, by applying situation labeling functions, a labeling matrix that is fed to a generative model” reads on the label matrix of Moridi (Paragraph [0042]) include results of each of the labeling functions respectively, with the labeling functions being inputted from the label matrix to the generative model of the Generative Adversarial Network (GAN). The broadest reasonable interpretation of “the generative model feeds a discriminative classifier model with probabilistic labels for the sensor features, wherein the probabilistic labels of the generative model are used for training the discriminative classifier model” reads on the generative model of Moridi (Paragraphs [0042] – [0044]) that aggregates and analyzes the results of the labeling functions from the label matrix to thereby determine probability labels associated with two or more classes and output those probability labels to the discriminative model for training.); and
determining, by a feature selection optimizer entity, a subset of the sensor features to be used by the discriminative classifier model to detect predefined situations which make the indoor environment prone to infectious disease transmission, the subset of the sensor features being determined based on an optimization procedure that is built based on a feature cost vector, wherein the feature cost vector includes feature cost information for the sensor features and at least one of an energy consumption cost and a privacy cost.
Regarding claim 15, Mondini further recites the limitation identified in bold as “a non-transitory, computer-readable storage medium having instructions thereon, which, upon execution on one or more processors provide for training a machine learning model for monitoring proneness of an indoor environment to infectious disease transmission by execution of the following steps” (Paragraph [0081] of Mondini. In the instant application, the broadest reasonable interpretation of “a non-transitory, computer-readable storage medium having instructions thereon, which, upon execution on one or more processors provide for training a machine learning model for monitoring proneness of an indoor environment to infectious disease transmission by execution of the following steps:” reads on the non-transitory machine-readable storage medium in Mondini (Paragraph [0081]) storing instructions that may be read and executed by at least one processor to perform the operations described in detail herein.).
Moridi does not appear to disclose, but Sobol teaches the limitation of claim 14 identified in bold as “a system for training a machine learning model for monitoring of a proneness of an indoor environment to infectious disease transmission, the system comprising a functional unit having one or more computational processors with access to memory, which, alone or in combination, are configured to provide for execution of the following steps,” “obtaining unlabeled sensor data from sensors of the indoor environment in order to provide a set of sensor features,” and “determining, by a feature selection optimizer entity, a subset of the sensor features to be used by the discriminative classifier model to detect predefined situations which make the indoor environment prone to infectious disease transmission, the subset of the sensor features being determined based on an optimization procedure that is built based on a feature cost vector, wherein the feature cost vector includes feature cost information for the sensor features and at least one of an energy consumption cost and a privacy cost” in claim 14 and “a non-transitory, computer-readable storage medium having instructions thereon, which, upon execution on one or more processors provide for training a machine learning model for monitoring proneness of an indoor environment to infectious disease transmission by execution of the following steps” in claim 15 (Paragraphs [0009], [0017], and [0042] of Sobol. In the instant application, the broadest reasonable interpretation of “an indoor environment” and “sensors of the indoor environment” reads on the BLE beacons in Sobol (Paragraphs [0009], [0017], and [0042], see also [0030], [0130] – [0133], and [0301]) for sending indoor location-detecting signal transmissions to the wearable electronic device (i.e., particularly when the assisted living community, hospital, home or other dwelling or place of recovery is of older construction where one or more of building layout, choice of construction materials or other factors can degrade a WiFi signal or Bluetooth signal relative to a BLE signal) and sensors that are formed as part of the wearable electronic device for acquiring environmental data, with the sensors including an ambient temperature sensor, an ambient pressure sensor, an ambient humidity sensor, an ambient light sensor, a motion sensor, a carbon monoxide sensor, a carbon dioxide sensor, a smoke detector or a microphone.).
Moridi does not appear to disclose, but Achin teaches the limitation of claim 14 identified in bold as “a system for training a machine learning model for monitoring of a proneness of an indoor environment to infectious disease transmission, the system comprising a functional unit having one or more computational processors with access to memory, which, alone or in combination, are configured to provide for execution of the following steps” (Abstract and Paragraph [0072], [0084] – [0085], and [0177] of Achin. In the instant application, the broadest reasonable interpretation of “a system for training a machine learning model for monitoring of a proneness … to infectious disease transmission” reads on the system in Achin (Abstract and Paragraph [0072], [0084] – [0085], and [0177], see also [0106]) of epidemiological modeling using artificial intelligence and machine learning for any infectious or communicable disease and training the selected predictive model.).
Moridi does not appear to disclose, but Sobol teaches the limitation of claim 14 identified in bold as “obtaining unlabeled sensor data from sensors of the indoor environment in order to provide a set of sensor features” (Paragraph [0233] of Sobol. In the instant application, the broadest reasonable interpretation of “sensor data from sensors … in order to provide a set of sensor features” reads on the raw data, e.g., environmental data, (i.e., sensor data) of Sobol (Paragraph [0233], see also [0017] and [0042]) acquired from the sensors to form derived features (i.e., set of sensor features) for use in subsequent learning and generalization activities within the algorithms.).
Moridi does not appear to disclose, but Achin teaches the limitation of claim 14 identified in bold as “determining, by a feature selection optimizer entity, a subset of the sensor features to be used by the discriminative classifier model to detect predefined situations which make the indoor environment prone to infectious disease transmission, the subset of the sensor features being determined based on an optimization procedure that is built based on a feature cost vector, wherein the feature cost vector includes feature cost information for the sensor features and at least one of an energy consumption cost and a privacy cost” (Paragraph [0295] of Sobol. In the instant application, the broadest reasonable interpretation of “determining, by a feature selection optimizer entity, a subset of the sensor features to be used by the discriminative classifier model to detect predefined situations … prone to infectious disease transmission, the subset of the sensor features being determined based on an optimization procedure that is built based on a feature cost vector” reads on the activity in Sobol (Paragraph [0295], see also [0233], [0292] – [0293], [0328] – [0330], [0334]) of finding, via feature selection (i.e., feature selection optimizer entity), a subset of the original variables (i.e., subset of sensor features) that contain accurate, relevant input data (e.g., sequential feature selection) and feature extraction (also called feature transformation, where the data is placed into a feature space of fewer dimensions while maintaining accuracy) as a way to spot rules, trends or related likelihoods. One common form of dimensionality reduction algorithm is the PCA that through search optimization (i.e., an optimization procedure) transforms feature variables into a smaller number of principal component variables (i.e., subset of sensor features). The Conditional Random Field (CRF) probabilistic modeling approach is a graphical-based discriminative probabilistic model that may be used as a classification model (i.e., a discriminative classifier) for diagnosis, e.g., a diagnosis of community-acquired pneumonia (CAP) when an individual has spent time in a location where the likelihood coming into contact with another who may have pneumonia is heightened and predicting the onset of influenza by analyzing whether an individual shows the associated predictors.).
Moridi does not appear to disclose, but NPL Akmandor teaches the limitation identified in bold as “determining, by a feature selection optimizer entity, a subset of the sensor features to be used by the discriminative classifier model to detect predefined situations which make the indoor environment prone to infectious disease transmission, the subset of the sensor features being determined based on an optimization procedure that is built based on a feature cost vector, wherein the feature cost vector includes feature cost information for the sensor features and at least one of an energy consumption cost and a privacy cost” (Second Paragraph in Second Column on Page 915 to First Paragraph in Second Column on Page 916, Second Paragraph to Third Paragraph in First Column on Page 918, Second Paragraph in Second Column on Page 919 to Fifth Paragraph in Second Column on Page 920; and Third Paragraph in Second Column on Page 926 to Last Paragraph in First Column on Page 927 of NPL Akmandor. In the instant application, the broadest reasonable interpretation of “the feature cost vector includes feature cost information for the sensor features” reads on the encryption and hashing (i.e., and associated computational cost) in NPL Akmandor (Second Paragraph to Third Paragraph in First Column on Page 918) used by the IoT sensors and their corresponding applications. The broadest reasonable interpretation of “the feature cost vector includes … at least one of an energy consumption cost and a privacy cost” reads on the energy efficiency (i.e., decreased energy consumption via both compression and machine learning inferences) and privacy/security (i.e., increased energy consumption via encryption and hashing) in NPL Akmandor (Second Paragraph in Second Column on Page 915 to First Paragraph in Second Column on Page 916, Second Paragraph in Second Column on Page 919 to Fifth Paragraph in Second Column on Page 920; and Third Paragraph in Second Column on Page 926 to Last Paragraph in First Column on Page 927) associated with inference outcomes generated from informative values (i.e., features) extracted from data collected from the IoT sensors for their corresponding applications.).
Therefore, it would have been obvious to one of ordinary skill in the art of artificial intelligence in epidemiology at the time of filing to modify the system and method of Moridi to implement: the indoor environment, the sensors of the indoor environment, the activity of obtaining unlabeled sensor data from sensors of the indoor environment in order to provide a set of sensor features, and the activity of determining, by a feature selection optimizer entity, a subset of the sensor features to be used by the discriminative classifier model to detect predefined situations which make the indoor environment prone to infectious disease transmission, the subset of the sensor features being determined based on an optimization procedure that is built based on a feature cost vector, wherein the feature cost vector includes feature cost information for the sensor features and at least one of an energy consumption cost and a privacy cost, as taught by Sobol (Paragraphs [0009], [0017], [0042], [0233] – [0236], [0243], [0295], [0328] – [0330], and [0334]) in order to improve the accuracy of predicting which room the wearer is likely to occupy next in order to form one or both of baseline wearer state of health and changes to such health (Paragraph [0223] of Sobol); implement the system for training a machine learning model for monitoring of a proneness of an indoor environment to infectious disease transmission, the system comprising a functional unit having one or more computational processors with access to memory, which, alone or in combination, are configured to provide for execution of the following steps, as taught by Achin (Abstract and Paragraph [0072], [0084] – [0085], and [0177]), in order to help communities assist at-risk individuals, stop event spread, monitor mitigation progress and policies, and predict future occurrences (Paragraph [0003] of Achin), and implement the activity of determining, by a feature selection optimizer entity, a subset of the sensor features to be used by the discriminative classifier model to detect predefined situations which make the indoor environment prone to infectious disease transmission, the subset of the sensor features being determined based on an optimization procedure that is built based on a feature cost vector, wherein the feature cost vector includes feature cost information for the sensor features and at least one of an energy consumption cost and a privacy cost, as taught by NPL Akmandor (Second Paragraph in Second Column on Page 915 to First Paragraph in Second Column on Page 916, Second Paragraph to Third Paragraph in First Column on Page 918, Second Paragraph in Second Column on Page 919 to Fifth Paragraph in Second Column on Page 920; and Third Paragraph in Second Column on Page 926 to Last Paragraph in First Column on Page 927) in order to enable the IoT system to solve data security and energy efficiency problems simultaneously in healthcare and environmental monitoring applications (Abstract on Page 914 and Table 1 on Page 917 of NPL Akmandor).
Regarding claim 10, Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1, teaches the limitation identified in bold as “the optimization procedure includes an optimization function, wherein the optimization function is built based on training and/or prediction times of the generative model and/or of the discriminative classifier model” (Paragraph [0369] of Achin, The system can use a multi-step optimization process where parameters θt=[αt, βt, γt] are tuned each day. To find the parameters θt that minimize the error between simulated and actual cases and deaths during [t, t+30 days] when θi∈[t,t+30] is held constant (i.e., θi∈[t,t+30 days]=θt), sequential bounded Brent minimization (γ) can first be performed on αt, then on βt, and finally on γt using the following loss functions:
PNG
media_image1.png
659
849
media_image1.png
Greyscale
Paragraph [0370] of Achin, A Gaussian process minimization can then be performed on all three parameters simultaneously, according to Equation (13). Finally, the variance Σt+1 can be updated using a Laplace approximation.
PNG
media_image2.png
103
822
media_image2.png
Greyscale
In the instant application, the broadest reasonable interpretation of “the optimization procedure includes an optimization function” reads on the multi-step optimization process of Achin (Paragraphs [0369] and [0370]) built on the loss functions and the Laplace approximation (i.e., a feature cost vector). The broadest reasonable interpretation of “the optimization function is built based on training and/or prediction times of the generative model and/or of the discriminative classifier model” reads on the loss functions and the Laplace approximation (i.e., the feature cost vector) of Achin (Paragraphs [0369] and [0370]) during [t, t+30 days] (i.e., training and/or prediction times of the generative model and/or the discriminative classifier model).).
Regarding claim 11, Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1, teaches the limitation identified in bold as “the optimization procedure includes an optimization function, wherein the optimization function is built based on prediction accuracy and/or confidence values of the discriminative classifier model” (Paragraph [0369] of Achin, The system can use a multi-step optimization process where parameters θt=[αt, βt, γt] are tuned each day. To find the parameters θt that minimize the error between simulated and actual cases and deaths during [t, t+30 days] when θi∈[t,t+30] is held constant (i.e., θi∈[t,t+30 days]=θt), sequential bounded Brent minimization (γ) can first be performed on αt, then on βt, and finally on γt using the … loss functions: Paragraph [0370] of Achin, A Gaussian process minimization can then be performed on all three parameters simultaneously, according to Equation (13). Finally, the variance Σt+1 can be updated using a Laplace approximation. Paragraph [0335] of Achin, Similarly, certain observations may represent particularly useful or indicative events to which the user wants to assign additional weight. Thus, an additional variable inserted into the dataset may indicate the relative weight of each observation. The predictive modeling exploration engine 210 may then use this weight when training models and calculating their accuracy, with the goal being to produce more accurate predictions under higher-weighted conditions. In the instant application, the broadest reasonable interpretation of “the optimization procedure includes an optimization function” reads on the multi-step optimization process of Achin (Paragraphs [0369] and [0370]) built on the loss functions and the Laplace approximation (i.e., a feature cost vector). The broadest reasonable interpretation of “the optimization function is built based on prediction accuracy and/or confidence values of the discriminative classifier model” reads on the predictive modeling exploration engine 210 of Achin (Paragraph [0335]) using an additional variable inserted into a data set for calculating their accuracy, with the goal being to produce more accurate predictions under higher-weighted conditions.).
Regarding claim 12, Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1 teaches the limitation identified in bold as “in the training phase, the feature selection optimizer entity iteratively interacts with the discriminative classifier model such that the subset of sensor features is iteratively updated based on feedback information that is provided by the discriminative classifier model” (Paragraph [0033] of Moridi, [T]he Generative Adversarial Network (GAN) training application 106 may be configured to utilize the DBSCAN algorithm to reduce the spatial data set size of the trip log. Paragraph [0077] of Moridi, [T]he discriminative module 308 may input the set(s) of probabilistic labels to the discriminative model 216 of the GAN 104 to train a wide variety of machine learning models (i.e., the GAN and the DBSCAN algorithm). In the instant application, the broadest reasonable interpretation of “the feature selection optimizer entity interacts with the discriminative classifier model such that the subset of sensor features is updated based on feedback information that is provided by the discriminative classifier model” reads on the DBSCAN algorithm of the GAN 104 (i.e., the feature selection optimizer entity) in Moridi (Paragraph [0033]) reducing a spatial data set size (thereby leaving a subset of the original data set remaining) and the discriminative module 308 inputting the set(s) of probabilistic labels to the discriminative model 216 to train a wide variety of machine learning models (i.e., providing feedback to the feature selection optimizer entity, e.g., the GAN and the DBSCAN algorithm, to update the subset of sensor features, e.g., modify reduction of the spatial data set))
Moridi as applied to claim 1 does not appear to explicitly disclose, but Achin teaches the limitation identified in bold as “in the training phase, the feature selection optimizer entity iteratively interacts with the discriminative classifier model such that the subset of sensor features is iteratively updated based on feedback information that is provided by the discriminative classifier model” (Paragraph [0175] of Achin, [S]electing the predictive model for the prediction problem may comprise iteratively selecting a subset of the predictive models and training the selected predictive models on larger or different portions of the dataset. This iterative process may continue until a predictive model is selected for the prediction problem or until the processing resources budgeted for generating the predictive model are exhausted. In the instant application, the broadest reasonable interpretation of iterative interaction between the feature selection optimizer entity and the discriminative classifier model iteratively updating the subset of sensor features based on feedback information that is provided by the discriminative classifier model reads on the activity in Achin (Paragraph [0175]) of selecting for the prediction problem by “iteratively selecting” a subset of the predictive models and training the selected predictive models on larger or different portions of the dataset.)
Regarding claim 13, Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1, teaches the limitation identified in bold as “the probabilistic labels of the generative model are based on the unlabeled sensor data and predetermined situation labeling thresholds” (Paragraph [0063] of Achin, [T]he data points associated with locational coordinates associated with the parking locations of the vehicle 102 may be clustered, re-clustered, and analyzed to determine a subset of aggregated locational coordinates that may be within a distance threshold of the center points of a cluster associated with a regularly visited location of the vehicle occupant and/or another regularly visited location. The aggregated locational coordinates may thereby be associated with regularly visited locations of the vehicle 102. The data filtering module 302 may be configured to classify the aggregated locational coordinates that may pertain to a particular point of interest location that may be regularly visited by the vehicle 102 as a center point of a particular cluster that may correspond to the particular point of interest location (e.g., regularly visited location, home location). In the instant application, the broadest reasonable interpretation of “the probabilistic labels of the generative model are based on the unlabeled sensor data and predetermined situation labeling thresholds” reads on the subset of aggregated locational coordinates of Achin (Paragraph [0063]) that may be within a distance threshold of the center points of a cluster associated with a regularly visited location of the vehicle occupant and/or another regularly visited location.).
Regarding claim 18, Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1, teaches the limitation identified in bold as “using, by the discriminative classifier model, the subset of sensor features for detecting predefined situations which make the indoor environment prone to infectious disease transmission” (Paragraph [0042]) include results of each of the labeling functions 208 respectively, with the labeling functions 208 being inputted from the label matrix 212 to the generative model 214 of the Generative Adversarial Network (GAN) 104. Paragraph [0397] of Achin. FIG. 13D illustrates an example user interface to present a social distancing factor, further to the example user interface of FIG. 13A. … The Social Distancing graph tracks state-based social distancing policies for the largest and most influential changes to social behavior. This is useful because social distancing behaviors, as mandated by each state, as well as influenced by the responsible citizen factor, have a considerable impact on the spread of the virus. The graph helps to illustrate the effects of a policy on virus case count for a location. In the instant application, the broadest reasonable interpretation of “the discriminative classifier model” reads on the discriminative model 216 of Moridi (Paragraph [0044]). The broadest reasonable interpretation of “using … the subset of sensor features for detecting predefined situations which make the indoor environment prone to infectious disease transmission” reads on the social distancing factors in Achin (Paragraph [0397]), such as social distancing behaviors, and the responsible citizen factor having a considerable impact on the spread of the virus.).
Regarding claim 19, Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1, teaches the limitation identified in bold as “the indoor environment is a room of a building, and wherein the sensor features include data measured by the sensors that are located in the room” (Paragraph [0291] of Sobol. In the instant application, the broadest reasonable interpretation of “the indoor environment is a room of a building, and wherein the sensor features include data measured by the sensors that are located in the room” reads on the rooms in Sobol (Paragraph [0291]). in the house or other building where the individual resides and a wearable electronic device (i.e., having a plurality of sensors for measuring data) is worn by the individual residing in the rooms in the house.).
Regarding claim 20, Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 19, teaches the limitation identified in bold as “the sensors include at least one of a humidity sensor, thermometer, a carbon dioxide (C02) sensor, a stereoscopic camera, an acoustic sensor, an accelerometer or a gyroscope of a device located in the room, a barometer sensor, or an infrared sensor” (Paragraphs [0017]). In the instant application, the broadest reasonable interpretation of “the sensors include at least one of a humidity sensor, thermometer, a carbon dioxide (C02) sensor, a stereoscopic camera, an acoustic sensor, an accelerometer or a gyroscope of a device located in the room, a barometer sensor, or an infrared sensor” reads on the at least one sensor in Sobol (Paragraphs [0017]) is selected from the group consisting of an ambient temperature sensor, an ambient pressure sensor, an ambient humidity sensor, an ambient light sensor, a motion sensor, a carbon monoxide sensor, a carbon dioxide sensor, a smoke detector and a microphone.).
Claims 2 and 16 are rejected under 35 U.S.C. 103(a) as being unpatentable over Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1, and further in view of Crabtree (U.S. Pub. No. 2025/0175456 A1) and Goldberg (U.S. Pub. No. 2020/0051694 A1).
Regarding claim 2, Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1 the limitation identified in bold as “receiving feature cost information for the sensor features from a knowledge base, wherein the feature cost information is used for considering a predetermined constraint metric for the indoor environment” (The broadest reasonable interpretation of “feature cost information for the sensor features” reads on the encryption and hashing (i.e., and associated computational cost) in NPL Akmandor (Second Paragraph to Third Paragraph in First Column on Page 918) used by the IoT sensors and their corresponding applications.).
Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1 does not appear to explicitly disclose, but Crabtree teaches the limitation identified in bold as “receiving … the feature cost information is used for considering a predetermined constraint metric for the indoor environment” (Paragraph [0280] of Crabtree, A privacy scoring engine 3750 handles real-time evaluation of privacy exposure levels by analyzing overlapping sensor coverage zones and calculating contextual privacy metrics. For each vantage point along a user's path, the system estimates detection capability from various sensors, considering factors such as camera resolution, signal strength, and potential for triangulation. The scoring engine 3750 aggregates data to produce a numerical privacy score. In the instant application, the broadest reasonable interpretation of “feature cost information for the sensor features …, wherein the feature cost information is used for considering a predetermined constraint metric for the indoor environment” reads on the contextual privacy metrics in Crabtree (Paragraph [0280]) for overlapping zone coverage zones, camera resolution, signal strength, and potential for triangulation, wherein the contextual privacy metrics are used for producing the numerical privacy score.).
Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1 does not appear to explicitly disclose, but Goldberg teaches the limitation identified in bold as “receiving feature cost information for the sensor features from a knowledge base” (Paragraph [0072] of Goldberg, In a preferred embodiment a mobile app or web interface enables browsing the above mentioned Associative Medical Knowledge Base and enables accessing any of the data elements within the Associative Medical Knowledge Base along with its associations with other features. In the instant application, the broadest reasonable interpretation of “receiving … from a knowledge base” reads on the activity in Goldberg (Paragraph [0072]) of accessing data elements within the Associative Medical Knowledge Base.).
Therefore, it would have been obvious to one of ordinary skill in the art of artificial intelligence in epidemiology at the time of filing to modify the system and method of Moridi in view of Achin and NPL Akmandor to: implement the feature cost information for considering a predetermined constraint metric for the indoor environment, as taught by Crabtree (Paragraph [0280]), in order to include physical proximity and real-world cues merging physical and network data for holistic, dynamic risk assessment (Paragraph [0150] of Crabtree); and include the step of receiving data from a knowledge base, as taught by Goldberg (Paragraph [0072]), in order to improve healthcare, cut costs of care, and reduce the number of unnecessary tests and diagnostic errors (Paragraph [0008] of Goldberg).
Regarding claim 16, Moridi in view of Achin, NPL Akmandor, Sobol, Crabtree, and Goldberg and applied to claim 2 teaches the limitation identified in bold as “wherein domain experts interact with the knowledge base to provide a cost value of each feature in the knowledge base” (Paragraph [0054] of Goldberg, [T]he Associative Medical Knowledge Base will include data which represents the strength of data sources. Such data will be either entered directly by the community experts or calculated using dedicated code to assign strength weights based on several factors e.g. strength of evidence and published research, academic strength of contributor, number of contributions by such member, her internal ranking, etc. In the instant application, the broadest reasonable interpretation of “a cost value of each feature” reads on the numerical privacy score in Crabtree (Paragraph [0280] as applied to claim 2 above). The broadest reasonable interpretation of “domain experts interact with the knowledge base to provide … in the knowledge base” reads on the community experts in Goldberg (Paragraph [0054]) entering data directly into the Associative Medical Knowledge Base.).
Claims 3 – 4 are rejected under 35 U.S.C. 103(a) as being unpatentable over Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1, and further in view of Jakobsson (U.S. Pub. No. 2020/0106637 A1).
Regarding claim 3, Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1 does not appear to explicitly disclose, but Jakobsson teaches the limitation identified in bold as “receiving sensor feature predicates from a knowledge base, wherein the sensor feature predicates specify characteristics of the sensor features” (Paragraph [0074] of Jakobsson, The disclosed system processes often sensitive sensor data and generates from it user predicates. Paragraph [0078] of Jakobsson, The heuristic techniques work as a rule-based system, where the computation of predicates is based on sensor output, … already computed predicates, … and profile data determined for the user(s) associated with the sensors of a space. Paragraph [0084] of Jakobsson, Once a location (corresponding to the “presence” predicate) and a pseudonym are computed, these two predicates are preferably used to compute other predicates, along with raw sensor data. Many of these derived predicates use multiple previous predicates in the form of a history. Paragraph [0093] of Jakobsson, The disclosed system creates a platform for sharing of non-sensitive predicate data. Paragraph [0094] of Jakobsson, sensor data is locally processed… or centrally processed, e.g., all sensor data being transmitted to a cloud computing resource or centralized server. Paragraph [0096] of Jakobsson, [S]uch information is preferably transmitted, to the extent it is possible, and stored in a log associated with the end-user system. In the instant application, the broadest reasonable interpretation of “receiving sensor feature predicates from a knowledge base, wherein the sensor feature predicates specify characteristics of the sensor features” reads on the activity in Jakobsson (Paragraphs [0078], [0084], [0093], [0094], and [0096]) of receiving, from the log via the platform, predicates generated from sensor data.).
Therefore, it would have been obvious to one of ordinary skill in the art of artificial intelligence in epidemiology at the time of filing to modify the system and method of Moridi in view of Achin and NPL Akmandor, to further include the step of, in the training phase, receiving sensor feature predicates from a knowledge base, wherein the sensor feature predicates specify characteristics of the sensor features, as taught by Jakobsson (Paragraphs [0078], [0084], [0093], [0094], and [0096]), in order to have techniques that can both provide an understanding of an observed space as well as react to the derived understanding in a manner that provides optimal benefit to users (Paragraph [0003] of Jakobsson).
Regarding claim 4, Moridi in view of Achin, NPL Akmandor, Sobol, and Jakobsson and applied to claim 3 teaches the limitation identified in bold as “a match score is employed for considering a level of matches of sensor feature predicates between two sensor features” (Paragraph [0084] of Jakobsson, Once a location (corresponding to the “presence” predicate) and a pseudonym are computed, these two predicates are preferably used to compute other predicates, along with raw sensor data. Many of these derived predicates use multiple previous predicates in the form of a history. For example, the determination of what is likely to be a bedroom is performed based on detecting a long-term presence with a very weak signal (not much movement, not much sound, and with periodic lack of signal), preferably during the same time of the day (or night) for a series of days. For example, the system may determine that a user enters an area at 10:35 pm, and that a user—presumably the same—exits the area at 6:00 am, and that there is intermittent minor movement and sound in between. On a second day, the system, may detect the arrival of a user at 10:25 pm and the departure of a user at 5:52 am, with low-strength sensor signal associated with the space in between. After a few days of largely consistent determinations of this type, this area, defined as the space covered by a small set of nodes and associated sensors, is determined tentatively to be a bedroom. As more classifications that are consistent with this [classification] are made, the certainty of the classification improves. This can be expressed as a score, a probability, or as a variance computed from different individual observations. In the instant application, the broadest reasonable interpretation of “a match score is employed for considering a level of matches of sensor feature predicates between two sensor features” reads on the score in Jakobsson (Paragraph [0084]) employed as an expression of the certainty of a classification based on predicates, e.g., two predicates and other predicates computed by the two predicates.).
Claims 5 – 8 are rejected under 35 U.S.C. 103(a) as being unpatentable over Moridi in view of Achin, NPL Akmandor, Sobol, and Jakobsson as applied to claim 3, and further in view of Narlikar (U.S. Pub. No. 2021/0406601 A1).
Regarding claim 5, Moridi in view of Achin, NPL Akmandor, Sobol, and Jakobsson and applied to claim 3 does not appear to explicitly disclose, but Narlikar teaches the limitation identified in bold as “generating a feature node dependency graph based on the sensor feature predicates of the sensor features, wherein a match between the sensor feature predicates of two sensor features constitutes a dependency between the two sensor features” (Paragraph [0100] of Narlikar, To identify points similar to labeled negatives, label propagation over a graph induced by the created common feature representation is used (e.g., by the data processing system 105 or any of its components). In this graph, each vertex can represent a data point, and edges between data points can be constructed based on features shared between them, where these features can be either categorical or quantitative. The edges between data points can be weighted based on the similarities between their feature representations. These weights can be determined either based on the number of shared feature values between them if the features are categorical, or based on a pre-specified distance metric if the features are quantitative. Paragraph [0014] of Narlikar, In some implementations of the system, applying a set of content rules may further include identifying a first text portion of the content item. In some implementations of the system, training the discrimination model may further include matching the first text portion of the content item to a set of conditional instructions to determine that a feature is present in the content item. In some implementations of the system. In some implementations of the system, training the discrimination model further includes updating the feature space based on the determination that the feature may be present in the content item. In the instant application, the broadest reasonable interpretation of “generating a feature node dependency graph based on the sensor feature predicates of the sensor features, wherein a match between the sensor feature predicates of two sensor features constitutes a dependency between the two sensor features” reads on the step in Narlikar (Paragraph [0100]) of inducing a graph based on the created common feature representation, with each vertex representing a data point, and edges between data points constructed based on features shared between them, and the edges weighted based on: similarities between their feature representations; the number of shared feature values between them if the features are categorical; or a pre-specified distance metric if the features are quantitative.).
Therefore, it would have been obvious to one of ordinary skill in the art of artificial intelligence in epidemiology at the time of filing to further modify the system and method of Moridi in view of Achin, NPL Akmandor, Sobol, and Jakobsson and applied to claim 3, to further include the step, in the training phase, of generating a feature node dependency graph based on the sensor feature predicates of the sensor features, wherein a match between the sensor feature predicates of two sensor features constitutes a dependency between the two sensor features, as taught by Narlikar (Paragraphs [0014] and [0100]), in order to adapt the same predictive tasks to data modalities that arise over the application life cycle and reduce the time needed to develop models for new modalities (Paragraph [0041] of Narlikar).
Regarding claim 6, Moridi in view of Achin, NPL Akmandor, Sobol, Jakobsson, and Narlikar and applied to claim 5 teaches the limitation identified in bold as “the feature node dependency graph is generated in such a way that the sensor features are represented by vertices, connections between sensor features are represented by edges, and match scores are represented by edge weights” (Paragraph [0100] of Narlikar, To identify points similar to labeled negatives, label propagation over a graph induced by the created common feature representation is used (e.g., by the data processing system 105 or any of its components). In this graph, each vertex can represent a data point, and edges between data points can be constructed based on features shared between them, where these features can be either categorical or quantitative. The edges between data points can be weighted based on the similarities between their feature representations. These weights can be determined either based on the number of shared feature values between them if the features are categorical, or based on a pre-specified distance metric if the features are quantitative. Paragraph [0101] of Narlikar, The algorithm can propagate the labels of data points of existing modalities to neighboring unlabeled points of the new modality using this graph structure. An unlabeled data point that shares edges with labeled data points can be assigned a weighted combination of its neighbor's labels. The algorithm can iteratively update this assignment until convergence, and return the final weighted score. This score can be used as part of a threshold-based LF. Similar to generating normal LFs, this threshold can be selected using a development set in the target modality. However, in the cross-modal setting, a development set consisting of labeled examples in existing modalities can be leveraged. In the instant application, the broadest reasonable interpretation of “the feature node dependency graph is generated in such a way that the sensor features are represented by vertices, connections between sensor features are represented by edges, and match scores are represented by edge weights” reads on the graph in Narlikar (Paragraph [0100] and [0101]) based on the created common feature representation, with each vertex representing a data point, and edges between data points constructed based on features shared between them, and the edges weighted based on: similarities between their feature representations; the number of shared feature values between them if the features are categorical; or a pre-specified distance metric if the features are quantitative.).
Regarding claim 7, Moridi in view of Achin, NPL Akmandor, Sobol, Jakobsson, and Narlikar and applied to claim 5 teaches the limitation identified in bold as “the optimization procedure of the feature selection optimizer entity is based on a traversal of the feature node dependency graph” (Paragraph [0101] of Narlikar, The algorithm can propagate the labels of data points of existing modalities to neighboring unlabeled points of the new modality using this graph structure. An unlabeled data point that shares edges with labeled data points can be assigned a weighted combination of its neighbor's labels. The algorithm can iteratively update this assignment until convergence, and return the final weighted score. This score can be used as part of a threshold-based LF. Similar to generating normal LFs, this threshold can be selected using a development set in the target modality. However, in the cross-modal setting, a development set consisting of labeled examples in existing modalities can be leveraged. In the instant application, the broadest reasonable interpretation of “the optimization procedure of the feature selection optimizer entity is based on a traversal of the feature node dependency graph” reads on the algorithm of Narlikar (Paragraph [0101]) propagating the labels of data points of existing modalities to neighboring unlabeled points of the new modality using this graph structure.).
Regarding claim 8, Moridi in view of Achin, NPL Akmandor, Sobol, Jakobsson, and Narlikar and applied to claim 5 teaches the limitation identified in bold as “the optimization procedure includes an optimization function, and wherein the optimization function is built based on the feature node dependency graph based on the edge weight values” (Paragraph [0369] of Achin, The system can use a multi-step optimization process where parameters θt=[αt, βt, γt] are tuned each day. To find the parameters θt that minimize the error between simulated and actual cases and deaths during [t, t+30 days] when θi∈[t,t+30] is held constant (i.e., θi∈[t,t+30 days]=θt), sequential bounded Brent minimization (γ) can first be performed on αt, then on βt, and finally on γt using the … loss functions: Paragraph [0370] of Achin, A Gaussian process minimization can then be performed on all three parameters simultaneously, according to Equation (13). Finally, the variance Σt+1 can be updated using a Laplace approximation. Paragraph [0101] of Narlikar, The algorithm can propagate the labels of data points of existing modalities to neighboring unlabeled points of the new modality using this graph structure. An unlabeled data point that shares edges with labeled data points can be assigned a weighted combination of its neighbor's labels. The algorithm can iteratively update this assignment until convergence, and return the final weighted score. This score can be used as part of a threshold-based LF. Similar to generating normal LFs, this threshold can be selected using a development set in the target modality. However, in the cross-modal setting, a development set consisting of labeled examples in existing modalities can be leveraged. In the instant application, the broadest reasonable interpretation of “the optimization procedure includes an optimization function” reads on the multi-step optimization process of Achin (Paragraphs [0369] and [0370]) built on the loss functions and the Laplace approximation (i.e., a feature cost vector). The broadest reasonable interpretation of “the optimization procedure includes an optimization function, and wherein the optimization function is built based on the feature node dependency graph based on the edge weight values” reads on the algorithm of Narlikar (Paragraph [0101]) propagating the labels of data points of existing modalities to neighboring unlabeled points that sharing edges assigned a weighted combination of its neighbor's labels.).
Claim 17 is rejected under 35 U.S.C. 103(a) as being unpatentable over Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1, and further in view of Narlikar.
Regarding claim 17, Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1 teaches the limitation identified in bold as “wherein determining, by the feature selection optimizer entity, the subset of the sensor features based on the optimization procedure comprises taking, as input, the set of sensor features and the feature cost vector for the sensor features, and determining the subset of the sensor features by iteratively interacting with the discriminative classifier model to iteratively update the subset of features based on feedback information” (See NPL Akmandor (Second Paragraph in Second Column on Page 915 to First Paragraph in Second Column on Page 916) as applied to claim 1 for teaching the limitation “the subset of the sensor features being determined based on an optimization procedure that is built based on a feature cost vector.”).
Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1 does not appear to explicitly disclose, but Narlikar teaches the limitation identified in bold as “wherein … determining the subset of the sensor features by iteratively interacting with the discriminative classifier model to iteratively update the subset of features based on feedback information” (Paragraph [0092] of Narlikar, [T]he probabilistic labels can be used to train an end discriminative model for the target task that operates over a noise-aware loss function (LF). Paragraph [0100] of Narlikar, To identify points similar to labeled negatives, label propagation over a graph induced by the created common feature representation is used (e.g., by the data processing system 105 or any of its components). Paragraph [0101] of Narlikar, The algorithm can propagate the labels of data points of existing modalities to neighboring unlabeled points of the new modality using this graph structure. An unlabeled data point that shares edges with labeled data points can be assigned a weighted combination of its neighbor's labels. The algorithm can iteratively update this assignment until convergence, and return the final weighted score. In the instant application, the broadest reasonable interpretation of “iteratively interacting with the discriminative classifier model to iteratively update the subset of features based on feedback information” reads on the activity in Narlikar (Paragraphs [0092], [0100] and [0101]) of propagating the labels of data points of existing modalities to neighboring unlabeled points of the new modality using this graph structure such that each unlabeled data point that shares edges with labeled data points can be assigned a weighted combination of its neighbor's labels and this assignment is iteratively updated until convergence.).
Therefore, it would have been obvious to one of ordinary skill in the art of artificial intelligence in epidemiology at the time of filing to modify the system and method of Moridi in view of Achin, and NPL Akmandor and applied to claim 1, to include the step of iteratively interacting with the discriminative classifier model to iteratively update the subset of features based on feedback information, as taught by Narlikar (Paragraphs [0092], [0100] and [0101]), in order to adapt the same predictive tasks to data modalities that arise over the application life cycle and reduce the time needed to develop models for new modalities (Paragraph [0041] of Narlikar).
Claim 21 is rejected under 35 U.S.C. 103(a) as being unpatentable over Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1, and further in view of Risbeck (U.S. Pub. No. 2021/0011444 A1).
Regarding claim 21, Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1 teaches the limitation identified in bold as “activating a heating, ventilation, and air conditioning (HVAC) system to control the indoor environment based on an output of the discriminative classifier model that detects a predefined situation of the predefined situations that make the indoor environment prone to infectious disease transmission” (Paragraph [0295] of Sobol. In the instant application, the broadest reasonable interpretation of “an output of the discriminative classifier model that detects a predefined situation of the predefined situations that make the indoor environment prone to infectious disease transmission” reads on the activity in Sobol (Paragraph [0295], see also [0233], [0292] – [0293], [0328] – [0330], [0334]) of finding, via feature selection (i.e., feature selection optimizer entity), a subset of the original variables (i.e., subset of sensor features) that contain accurate, relevant input data (e.g., sequential feature selection) and feature extraction (also called feature transformation, where the data is placed into a feature space of fewer dimensions while maintaining accuracy) as a way to spot rules, trends or related likelihoods. One common form of dimensionality reduction algorithm is the PCA that through search optimization (i.e., an optimization procedure) transforms feature variables into a smaller number of principal component variables (i.e., subset of sensor features). The Conditional Random Field (CRF) probabilistic modeling approach is a graphical-based discriminative probabilistic model that may be used as a classification model (i.e., a discriminative classifier) for diagnosis, e.g., a diagnosis of community-acquired pneumonia (CAP) when an individual has spent time in a location where the likelihood coming into contact with another who may have pneumonia is heightened and predicting the onset of influenza by analyzing whether an individual shows the associated predictors.).
Moridi in view of Achin, NPL Akmandor, and Sobol and as applied to claim 1 does not appear to explicitly disclose, but Risbeck teaches the limitation identified in bold as “activating a heating, ventilation, and air conditioning (HVAC) system to control the indoor environment based on an output of the discriminative classifier model that detects a predefined situation of the predefined situations that make the indoor environment prone to infectious disease transmission” (Claims 7, 13, and 20 of Risbeck. In the instant application, the broadest reasonable interpretation of “activating a heating, ventilation, and air conditioning (HVAC) system to control the indoor environment” reads on the activities in Risbeck (Claims 7, 13, and 20) of generating control decisions using the dynamic temperature model, the dynamic infectious quanta model, and the infection probability; and using the control decisions to operate at least one of disinfection lighting, a variable air volume (VAV) unit, or an air handling unit (AHU) of the HVAC system.).
Therefore, it would have been obvious to one of ordinary skill in the art of artificial intelligence in epidemiology at the time of filing to modify the system and method of Moridi in view of Achin, and NPL Akmandor and applied to claim 1, to include the step of activating a heating, ventilation, and air conditioning (HVAC) system to control the indoor environment based on an output of the discriminative classifier model that detects a predefined situation of the predefined situations that make the indoor environment prone to infectious disease transmission, as taught by Risbeck (Paragraphs [0092], [0100] and [0101]), in order to maintain occupant comfort in a building and provide sufficient disinfection for multiple environmental conditions in the building while reducing expenses related to maintaining occupant comfort and disinfection (Paragraph [0003] of Risbeck).
Response to Arguments
Applicant’s arguments (Sixth Paragraph on Page 7 to Second Paragraph on Page 12 of the Amendment filed January 26, 2026) and amendment with respect to the rejections of claims 1 – 18 under 35 U.S.C. § 103 have been considered and are moot in view of the new grounds of rejection necessitated by the amendment.
Independent claims 1, 14, and 15 have been amended to further recite in part the limitation identified in bold as “determining, by a feature selection optimizer entity, a subset of the sensor features to be used by the discriminative classifier model to detect predefined situations which make the indoor environment prone to infectious disease transmission.” This limitation was not recited in any one of previously pending claims 1 – 8 and 10 – 18, and thus raises a new issue necessitating the new grounds of rejection.
Conclusion
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 VINCENT CAESAR ILAGAN whose telephone number is (703) 756-1639. The examiner can normally be reached Monday - Friday 8:30 am - 6:00pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Jason B. Dunham, can be reached on (571) 272-8109. The fax phone number for the organization where this application or proceeding is assigned is 571-272-8300.
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.
/V.C.I./Examiner, Art Unit 3686
/DEVIN C HEIN/Examiner, Art Unit 3686
