DETAILED ACTION
This office action is in response to amendments filed on 05/29/2026.
Claims 1, 10, and 19 have been amended. Claims 9, 18, and 24 have been canceled. Claims 1-4, 6-7, 10-13, 15-16, 19, and 21-23 are pending.
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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 06/08/2026 has been entered.
Response to Arguments
Rejections Under 35 U.S.C. § 101:
In light of applicant’s amendments to the claims (pg. 2-8) and the associated arguments (pg. 9-14), the rejections under 35 U.S.C. § 101 have been withdrawn.
Prior Art Rejections:
Applicant’s arguments regarding the prior art rejections have been fully considered but they are not persuasive.
Applicant argues (pg. 15-16) that Akbari does not teach or suggest tracking features using a Kalman filter, Mahalanobis distance as an association metric, or model compression based on Shapley values. Examiner respectfully notes that, as can be seen in the rejection below, Akbari is not relied upon to teach these features.
Applicant similarly argues (pg. 16) that Khaertdinov does not teach or suggest tracking features using a Kalman filter, Mahalanobis distance as an association metric, or model compression based on Shapley values. Examiner respectfully notes that, as can be seen in the rejection below, Khaertdinov is not relied upon to teach these features.
Applicant argues (pg. 16-17) that while Vaishnav teaches feature tracking using a Kalman filter, it does not teach or suggest (1) using Mahalanobis distance as an association metric with triplet/quadruplet-loss extracted features, (2) smoothing features toward an activity cluster centroid, or (3) using tracked feature data for Shapley value-based model compression. Examiner respectfully notes that, as can be seen in the rejection below, (1) Vaishnav does in fact teach using Mahalanobis distance as an association metric, and Khaertdinov teaches triplet-loss extracted features, (2) the Gogoglou reference has been brought in to teach that a Kalman filter smooths tracked features toward a cluster centroid, and (3) Marcílio teaches Shapley value-based model compression.
Applicant argues (pg. 17) that while Marcílio teaches Shapley value-based feature selection for model compression, it does not teach or suggest generating Shapley values based on confidence and uncertainty feature tracking data from a Kalman filter which smooths features toward a cluster centroid. Examiner respectfully notes that while the Kalman filter may output feature tracking data which is smoothed toward a cluster centroid, this data is not structurally different from any other classifier input data to which Marcílio’s feature selection method might be applied. The claimed Shapley value-based model compression may be performed subsequent to the Kalman filtering of the data and thus depend on its output, but these are essentially two independent data processing steps. Therefore, the motivation for one of ordinary skill in the art to apply Marcílio’s feature selection to the output of Vaishnav’s Kalman filter is no different from the motivation to apply Marcílio’s feature selection to any other data: to reduce the dimensionality and thereby avoid complications arising from the “curse of dimensionality” (Marcílio, pg. 340, section I).
Applicant argues (pg. 17) that Marcílio does not teach or suggest that the Shapley value-based feature selection would yield the claimed technical result of a compressed model which retains separability of both known and unknown classes. Examiner respectfully notes that, as can be seen in the rejection below, the Wang reference has been brought in to teach this feature.
Applicant argues (pg. 17-18) that the examiner's conclusion of obviousness is based upon improper hindsight reasoning. Examiner respectfully notes that it must be recognized that any judgment on obviousness is in a sense necessarily a reconstruction based upon hindsight reasoning. But so long as it takes into account only knowledge which was within the level of ordinary skill at the time the claimed invention was made, and does not include knowledge gleaned only from the applicant's disclosure, such a reconstruction is proper. See In re McLaughlin, 443 F.2d 1392, 170 USPQ 209 (CCPA 1971).
Applicant argues (pg. 18-19) that the cited references do not address the specific technical problem of achieving model compression while retaining separability between known and unknown classes. Examiner respectfully notes that the reason or motivation to modify the reference may often suggest what the inventor has done, but for a different purpose or to solve a different problem. It is not necessary that the prior art suggest the combination to achieve the same advantage or result discovered by applicant. See, e.g., In re Kahn, 441 F.3d 977, 987, 78 USPQ2d 1329, 1336 (Fed. Cir. 2006); MPEP 2144(IV). Further, as can be seen in the rejection below, this specific technical benefit is achieved by combination of the previously cited references with newly cited reference Wang.
Applicant argues (pg. 19) that the cited combination would require substantial modification of the references to achieve the claimed invention. Examiner respectfully notes that the test for obviousness is not whether the features of a secondary reference may be bodily incorporated into the structure of the primary reference; nor is it that the claimed invention must be expressly suggested in any one or all of the references. Rather, the test is what the combined teachings of the references would have suggested to those of ordinary skill in the art. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981).
The prior art rejections have been updated to include the amended limitations and to clarify
the reasoning given for the limitations that were not amended.
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.
Claims 1-4, 6-7, 10-13, 15-16, 19, and 21-23 are rejected under 35 U.S.C. 103 as being unpatentable over
Akbari et al. (hereinafter Akbari), “A Deep Learning Assisted Method for Measuring Uncertainty in Activity Recognition with Wearable Sensors” in view of
Khaertdinov et al. (hereinafter Khaertdinov), “Deep Triplet Networks with Attention for Sensor-based Human Activity Recognition”,
Vaishnav et al. (hereinafter Vaishnav), “Continuous Human Activity Classification With Unscented Kalman Filter Tracking Using FMCW Radar”,
Gogoglou et al. (hereinafter Gogoglou), U.S. Patent Application Publication US-20220292340-A1,
Marcílio et al. (hereinafter Marcílio), “From explanations to feature selection: assessing SHAP values as feature selection mechanism”, and
Wang et al. (hereinafter Wang), “Zero-shot Feature Selection via Transferring Supervised Knowledge”.
Regarding Claim 1,
Akbari teaches A system comprising:
obtain an input signal corresponding to data obtained from a data source; (Pg. 4, section V.A: “We used 3D acceleration and gyroscope sensors that results in 18 axis of data and segmented the data into windows of length 100 (one second as the sampling rate of the sensors is 100Hz) with 50% overlap.” Windows of data (i.e. an input signal) correspond to data obtained from 3D acceleration and gyroscope sensors (i.e. a data source).)
extract a set of features using the input signal and a feature extractor, […] wherein the set of features comprises a set of confidence features and a set of uncertainty features; (Pg. 5, section IV.B: “In Figure 2, the encoder, which serves as feature extractor, estimates the mean and standard deviation of a Gaussian distribution that is the approximation of the posterior of the features given data…” Mean and standard deviation features are extracted from each input using a feature extractor to obtain a set of mean (i.e. confidence) and standard deviation (i.e. uncertainty) features.)
make an activity prediction associated with an object. (Pg. 2, section I: “We design a unified framework for automatic feature extraction, classification, and estimation of uncertainty of the classifier for human activity recognition.” The framework is used to classify human activity (i.e. predict an activity associated with an object).)
Akbari does not appear to explicitly disclose wherein the feature extractor is a triplet-loss based feature extractor or a quadruplet-loss based feature extractor,
However, Khaertdinov teaches wherein the feature extractor is a triplet-loss based feature extractor or a quadruplet-loss based feature extractor, (Pg. 1, section I: “Deep Metric Learning (DML), also known as similarity learning, is a paradigm of learning deep feature embeddings which are extensively used in various problems, mostly coming from the Computer Vision domain… This approach requires specific loss functions which are based on distances between certain data points such as triplet loss [15], quadruplet loss [16] or contrastive loss [17]. In this paper, we are focused on the triplet loss function and its variations. This study aims to apply the DML concept to sensor-based HAR [human activity recognition]. The main motivation for exploiting DML is its powerful property of extracting robust deep feature embeddings.” Feature extraction is performed using triplet loss.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Akbari and Khaertdinov. Akbari teaches measuring uncertainty in human activity recognition via a variational autoencoder which extracts a Gaussian distribution representing latent features of sensor data. Khaertdinov teaches human activity recognition where feature extraction is performed using triplet loss. One of ordinary skill would have motivation to combine Akbari and Khaertdinov because “triplet networks not only improve the quality of [human activity] recognition but also are capable of constructing robust feature representations less affected by subject heterogeneity and inter-class similarities” (Khaertdinov, pg. 9, section V).
Akbari and Khaertdinov do not appear to explicitly disclose
perform classification gating to generate a classification gating output;
generate a set of feature tracking data by recursively tracking the set of features and associated uncertainty based on the classification gating output, wherein generating the set of feature tracking data comprises implementing a Kalman filter that assumes a state vector as a Gaussian random variable distribution and uses a Mahalanobis distance as an association metric to perform association between the set of features and activity classes, wherein the set of feature tracking data comprises a set of confidence feature tracking data and a set of uncertainty feature tracking data, and
However, Vaishnav teaches perform classification gating to generate a classification gating output; (Pg. 3, section III.E: “Gating is used to remove noisy outlier data from being associated to the states of the tracker.” The data that is not removed by the classification gating is the classification gating output.)
generate a set of feature tracking data by recursively tracking the set of features and associated uncertainty based on the classification gating output, wherein generating the set of feature tracking data comprises implementing a Kalman filter that assumes a state vector as a Gaussian random variable distribution and uses a Mahalanobis distance as an association metric to perform association between the set of features and activity classes, wherein the set of feature tracking data comprises a set of confidence feature tracking data and a set of uncertainty feature tracking data, and (Pg. 1, section I: “The classification output is fed into the tracker through classification gating, where the activity class probabilities are updated.” Pg. 2, section III.A: “The UKF [unscented Kalman filter] assumes a Gaussian random variable for the distribution of the state vector. Thus, the integration of the classifier output into the tracker facilitates to obtain not only the value of the current state of the classification but also the uncertainty associated with the state.” Pg. 2, section III.C: “The UKF is based on unscented transformation that tries to approximate the distribution of a random variable that undergoes a nonlinear transformation. Considering a Gaussian random variable
η
with mean
μ
and covariance
Ω
, on performing a nonlinear transformation
ψ
=
ϕ
(
η
)
also leads to another Gaussian distribution.” Pg. 3, section III.E: “Mahalanobis distance metric is used to calculate distance between the predicted state value and the actual incoming measurements.” Tracking is performed on the data that is fed into the tracker through classification gating (i.e. based on the classification gating output) using a Kalman filter that assumes a Gaussian random variable distribution for the state vector. As shown by the cyclical arrows between the blocks labeled ‘Tracker’, ‘Feature Extraction’, ‘Classifier’, and ‘Classification Gating’ in figure 1(b), this process is recursive (pg. 1). The UKF tracker generates feature tracking data based on the state vector, which is defined by a Gaussian random variable and thus necessarily includes mean (i.e. confidence) and variance (i.e. uncertainty) features. Mahalanobis distance is used to measure the distance between the predicted state and actual measurements (i.e. association between the activity classes and features).)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Akbari, Khaertdinov, and Vaishnav. Akbari teaches measuring uncertainty in human activity recognition via a variational autoencoder which extracts a Gaussian distribution representing latent features of sensor data. Khaertdinov teaches human activity recognition where feature extraction is performed using triplet loss. Vaishnav teaches measuring uncertainty in human activity classification by tracking the distribution of a Gaussian random variable using an unscented Kalman filter. One of ordinary skill would have motivation to combine Akbari, Khaertdinov, and Vaishnav because “The proposed integration of classifier and tracker improves the classification accuracy by smoothening several misclassifications arising due to the mentioned artifacts. Furthermore, the UKF provides the state estimation along with its associated uncertainty, thus providing a simple mechanism for Bayesian classification in terms of estimating the uncertainty associated with a predicted class probabilities. Furthermore, the integration of classification probabilities into the tracker enables rejection of ghost targets from nonhuman Doppler sources and better target association” (Vaishnav, pg. 1, section 1).
Akbari, Khaertdinov, and Vaishnav do not appear to explicitly disclose wherein the Kalman filter tracks features over time while smoothing features towards an activity cluster centroid;
However, Gogoglou teaches wherein the Kalman filter tracks features over time while smoothing features towards an activity cluster centroid; (0042: “Advantageously, by applying the Kalman filter, an entity may be placed in a different neighborhood. For example, in FIG. 2, entity D has no neighbors within a predefined distance in the plot 216. However, by applying the Kalman filter, entity D may join a group, or cluster, of entities, such as the group including entities A and B.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Akbari, Khaertdinov, Vaishnav, and Gogoglou. Akbari teaches measuring uncertainty in human activity recognition via a variational autoencoder which extracts a Gaussian distribution representing latent features of sensor data. Khaertdinov teaches human activity recognition where feature extraction is performed using triplet loss. Vaishnav teaches measuring uncertainty in human activity classification by tracking the distribution of a Gaussian random variable using an unscented Kalman filter. Gogoglou teaches that Kalman filters smooth tracked entities toward clusters in feature space. One of ordinary skill would have motivation to combine Akbari, Khaertdinov, Vaishnav, and Gogoglou in order to leverage the smoothing and clustering capabilities of Kalman filters for grouping and classification of human activity.
Akbari, Khaertdinov, Vaishnav, and Gogoglou do not appear to explicitly disclose
memory; and a processing device, operatively coupled to the memory, to:
determine, for each feature of the set of features based on the set of feature tracking data, a respective model compression parameter of a set of model compression parameters, wherein the set of model compression parameters comprises a set of Shapley values generated based on the set of feature tracking data;
generate a subset of the set of features based on the set of model compression parameters by:
determining, for each feature of the set of features, whether a respective Shapley value for the feature satisfies a threshold condition; and
in response to determining that the respective Shapley value satisfies the threshold condition, adding the feature to the subset of the set of features;
compress a machine learning model to obtain a compressed model based on the subset of the set of features,
use the compressed model to make an activity prediction associated with an object.
However, Marcílio teaches memory; and a processing device, operatively coupled to the memory, to: (Pg. 343, section IV: “The experiments were performed in a computer with the following configuration: Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz, 32GB RAM, Windows 10 64 bits.”)
determine, for each feature of the set of features based on the set of feature tracking data, a respective model compression parameter of a set of model compression parameters, wherein the set of model compression parameters comprises a set of Shapley values generated based on the set of feature tracking data; (Pg. 340, section I: “The approach assigns SHAP values, which are contribution values for a model’s output, for each feature of each data point. These SHAP values encode the importance that a model gives for a feature, so that, we use the contribution information of each feature to order the features based on its importance.” Pg. 342, section III.A: “SHAP values [1] is a model addictive explanation approach, in which each prediction is explained by the contribution of the features of the dataset to the model’s output. More specifically, SHAP approximate Shapley values…” Each feature is assigned a SHAP value (i.e. a model compression parameter comprising a Shapley value) based on its contribution to model output (i.e. based on the feature tracking data).)
generate a subset of the set of features based on the set of model compression parameters by:
determining, for each feature of the set of features, whether a respective Shapley value for the feature satisfies a threshold condition; and (Pg. 340, section I: “In this case, selecting a subset of
d
features based on SHAP values means to select the first
d
features after ordering them based on the feature contributions to the model’s prediction.” For each feature, it is determined whether that feature’s SHAP value falls within the top
d
SHAP values (i.e. satisfies a threshold condition).)
in response to determining that the respective Shapley value satisfies the threshold condition, adding the feature to the subset of the set of features; (Pg. 340, section I: “In this case, selecting a subset of
d
features based on SHAP values means to select the first
d
features after ordering them based on the feature contributions to the model’s prediction.” For each feature, if it is determined that its SHAP value falls within the top
d
SHAP values (i.e. it satisfies the threshold condition), it is selected for (i.e. added to) the subset of features.)
compress a machine learning model to obtain a compressed model based on the subset of the set of features, (Pg. 340, section I: “In this case, selecting a subset of
d
features based on SHAP values means to select the first
d
features after ordering them based on the feature contributions to the model’s prediction.” The model is compressed by based on the subset of selected features.)
use the compressed model to make an [activity] prediction [associated with an object]. (Pg. 342, section IV: “The algorithms were evaluated upon eight publicly available datasets, described in Table I, and based on the Keep Absolute metric [33], which computes a model score on varying number of features kept for classification/regression.” The model with varying number of kept features (i.e. the compressed model) is evaluated on classification and regression tasks (i.e. makes a prediction). Making an activity prediction associated with an object is taught by Akbari, as shown above.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Akbari, Khaertdinov, Vaishnav, Gogoglou, and Marcílio. Akbari teaches measuring uncertainty in human activity recognition via a variational autoencoder which extracts a Gaussian distribution representing latent features of sensor data. Khaertdinov teaches human activity recognition where feature extraction is performed using triplet loss. Vaishnav teaches measuring uncertainty in human activity classification by tracking the distribution of a Gaussian random variable using an unscented Kalman filter. Gogoglou teaches that Kalman filters smooth tracked entities toward clusters in feature space. Marcílio teaches reducing the dimensionality of a dataset via feature selection based on explanatory SHAP values. One of ordinary skill would have motivation to combine Akbari, Khaertdinov, Vaishnav, Gogoglou, and Marcílio because “dealing with high-dimensional data can be complicated due to the so-called curse of dimensionality… Other approaches to deal with high dimensionality is to use feature selection algorithms,” but “One problem with traditional feature selection algorithms is related to their explainability issues” (Marcílio, pg. 340, section I). Marcílio solves these problems by providing “a methodology and assessment for feature selection based on model agnostic explanations” (Marcílio, pg. 340, section I) which “demonstrated to be superior to other common feature selection mechanisms” (Marcílio, pg. 346, section VI).
Akbari, Khaertdinov, Vaishnav, Gogoglou, and Marcílio do not appear to explicitly disclose wherein the compressed model retains separability of both known activity classes and unknown activity classes;
However, Wang teaches wherein the compressed model retains separability of both known activity classes and unknown activity classes; (Pg. 2-3, section 1: “Therefore, the problem of Zero-Shot Feature Selection (ZSFS), i.e., building a feature selection model that generalizes well to unseen concepts with limited training data of seen concepts, deserves great attention. The major challenge in the ZSFS problem is how to deduce the knowledge of unseen concepts from seen concepts. In fact, the primary reason why existing studies fail to handle unseen concepts is that they only consider the discrimination among seen concepts (like the 0/1-form class labels illustrated in Fig. 1), such that little knowledge could be deduced for unseen concepts. To address this, as illustrated in Fig. 2, we adopt the class-semantic descriptions (i.e., attributes) as supervision for feature selection.” Feature selection is performed (i.e. the model is compressed) using class-semantic descriptions as supervision so that both seen and unseen concepts (i.e. known and unknown activity classes) are discriminable (i.e. separable).)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Akbari, Khaertdinov, Vaishnav, Gogoglou, and Marcílio. Akbari teaches measuring uncertainty in human activity recognition via a variational autoencoder which extracts a Gaussian distribution representing latent features of sensor data. Khaertdinov teaches human activity recognition where feature extraction is performed using triplet loss. Vaishnav teaches measuring uncertainty in human activity classification by tracking the distribution of a Gaussian random variable using an unscented Kalman filter. Gogoglou teaches that Kalman filters smooth tracked entities toward clusters in feature space. Marcílio teaches reducing the dimensionality of a dataset via feature selection based on explanatory SHAP values. Wang teaches model compression via feature selection, where class-semantic descriptions are used for supervision in order to retain discriminability of both known and unknown classes. One of ordinary skill would have motivation to combine Akbari, Khaertdinov, Vaishnav, Gogoglou, Marcílio, and Wang because Wang’s feature selection method is “an effective technique for dimensionality reduction” (Wang, pg. 1, abstract) which, in contrast to other feature selection methods, “generalizes well to unseen concepts” (Wang, pg. 4, section 1).
Regarding Claim 2, Akbari, Khaertdinov, Vaishnav, Gogoglou, Marcílio, and Wang teach The system of claim 1, as shown above.
Akbari also teaches wherein, to obtain the input signal, the processing device is to:
receive raw data from the data source; and generate the input signal from the raw data. (Pg. 2, section IV: “The network receives raw signal
x
as input and maps it to a latent variable
z
.” Pg. 4, section V.A: “We used 3D acceleration and gyroscope sensors that results in 18 axis of data and segmented the data into windows of length 100 (one second as the sampling rate of the sensors is 100Hz) with 50% overlap.” Raw data is received from the 3D acceleration and gyroscope sensors (i.e. data source) and segmented into windows (i.e. the input signal is generated).)
Regarding Claim 3, Akbari, Khaertdinov, Vaishnav, Gogoglou, Marcílio, and Wang teach The system of claim 1, as shown above.
Akbari also teaches wherein the data source comprises a sensor device comprising one or more sensors. (Pg. 4, section V.A: “We used 3D acceleration and gyroscope sensors that results in 18 axis of data and segmented the data into windows of length 100 (one second as the sampling rate of the sensors is 100Hz) with 50% overlap.”)
Regarding Claim 4, Akbari, Khaertdinov, Vaishnav, Gogoglou, Marcílio, and Wang teach The system of claim 1, as shown above.
Akbari also teaches wherein:
the set of confidence features comprises a set of mean-based features; the set of uncertainty features comprises a set of variance-based features; (Pg. 5, section IV.B: “In Figure 2, the encoder, which serves as feature extractor, estimates the mean and standard deviation of a Gaussian distribution that is the approximation of the posterior of the features given data…” Mean and standard deviation features are extracted from each input to obtain a set of mean (i.e. mean-based) and standard deviation (i.e. variance-based) features.)
Vaishnav also teaches wherein:
the set of confidence feature tracking data comprises a set of mean-based feature tracking data; and the set of uncertainty feature tracking data comprises a set of variance-based feature tracking data. (Pg. 2, section III.A: “The UKF [unscented Kalman filter] assumes a Gaussian random variable for the distribution of the state vector. Thus, the integration of the classifier output into the tracker facilitates to obtain not only the value of the current state of the classification but also the uncertainty associated with the state.” Pg. 2, section III.C: “The UKF is based on unscented transformation that tries to approximate the distribution of a random variable that undergoes a nonlinear transformation. Considering a Gaussian random variable
η
with mean
μ
and covariance
Ω
, on performing a nonlinear transformation
ψ
=
ϕ
(
η
)
also leads to another Gaussian distribution.” The UKF tracker generates feature tracking data based on the state vector, which is defined by a Gaussian random variable and thus necessarily includes mean (i.e. mean-based) and variance/covariance (i.e. variance-based) features.)
Regarding Claim 6, Akbari, Khaertdinov, Vaishnav, Gogoglou, Marcílio, and Wang teach The system of claim 1, as shown above.
Marcílio also teaches wherein, to use the compressed model to make the activity prediction, the processing device is to train the compressed model during a training stage to obtain a trained model. (Pg. 343, figure 3: “To evaluate how well a feature selection technique can select important features, the model is retrained with
d
features kept for classification and
m
-
d
features masked, where
d
is the number of features to select and
m
is the dimensionality of the dataset.” Before classification (i.e. during the training stage), the model with
d
features kept (i.e. the compressed model) is retrained (i.e. trained).)
Regarding Claim 7, Akbari, Khaertdinov, Vaishnav, Gogoglou, Marcílio, and Wang teach The system of claim 1, as shown above.
Marcílio also teaches wherein, to use the compressed model to make the activity prediction, the processing device is to make the activity prediction during an inference stage. (Pg. 342, section IV: “The algorithms were evaluated upon eight publicly available datasets, described in Table I, and based on the Keep Absolute metric [33], which computes a model score on varying number of features kept for classification/regression.” The model with varying number of kept features (i.e. the compressed model) is evaluated on classification and regression tasks (i.e. makes predictions during the inference stage).)
Claims 10-13 and 15-16 are method claims containing substantially the same elements as system claims 1-4 and 6-7, respectively. Akbari, Khaertdinov, Vaishnav, Gogoglou, Marcílio, and Wang teach the elements of claims 1-4 and 6-7, as shown above.
Claims 19 and 21-23 are product claims containing substantially the same elements as system claims 1-4, respectively. Akbari, Khaertdinov, Vaishnav, Gogoglou, Marcílio, and Wang teach the elements of claims 1-4, as shown above.
Marcílio also teaches A non-transitory computer-readable storage medium comprising instructions that, when executed by a processing device, cause the processing device to: (Examiner notes that this limitation is interpreted as implementation of the disclosed process in a generic computing environment. Pg. 343, section IV: “The experiments were performed in a computer with the following configuration: Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz, 32GB RAM, Windows 10 64 bits.”)
Conclusion
Claims 1-4, 6-7, 10-13, 15-16, 19, and 21-23 are rejected.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BENJAMIN M ROHD whose telephone number is (571)272-6445. The examiner can normally be reached Mon-Thurs 8:00-6:00 EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Viker Lamardo can be reached at (571) 270-5871. 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.
/B.M.R./Examiner, Art Unit 2147 /VIKER A LAMARDO/Supervisory Patent Examiner, Art Unit 2147
Prosecution Insights