DETAILED ACTION
This office action is in response to amendments filed on 02/19/2026.
Claims 1, 8, and 15 have been amended. Claims 3, 4, 10, 11, 17, and 18 were canceled previously. Claims 1, 2, 5-9, 12-16, 19, and 20 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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 02/19/2026 has been entered.
Response to Arguments
35 U.S.C. 112(b) Rejection:
In light of applicant’s after final amendment filed on 01/13/2026, the previous rejections under 35 USC § 112(b) have been withdrawn. However, a new rejection under 35 USC § 112(b) has been introduced in light of the claim amendments filed on 02/19/2026.
Prior Art Rejection:
Applicant's arguments regarding the prior art rejections under 35 USC § 103 (pg. 20-26) have been fully considered but they are not persuasive.
Applicant argues (pg. 22) that the examiner's conclusion of obviousness is based upon improper hindsight reasoning. However, 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). The rejection under 35 U.S.C. 103 includes explanations as to why one of ordinary skill in the art would have been motivated to combine the references, and applicant provides no specific argument as to why these explanations might be insufficient.
Applicant argues (pg. 23-24) that the amended independent claim limitations "detecting data variations in each of the layers", "diluting the data variations", "retraining the computer data model… using the diluted data variations ", and "wherein the principal component represents a direction of strong variability in data features for types of classifiers" are not disclosed or suggested by any of the cited references or their combination. Examiner respectfully notes that these limitations are taught by Liu and Artursson. Specifically, Liu (section 2.1 & 2.4.3) teaches selecting drifted data instances with high uncertainty (i.e. detecting data variations), expert labeling of the drifted data instances (i.e. diluting the data variations by removing the uncertainty), and renewing/retraining the learner with the labeled data instances (i.e. retraining the model using the diluted data variations). Artursson (pg. 712) teaches that the principal component captures the dominating type of variability in the data features used for classification (i.e. that the principal component represents a direction of strong variability in the features).
Applicant argues (pg. 23) that Liu teaches updating a training data set, but not retraining the model using the diluted data variations. Examiner respectfully notes that Liu does in fact teach retraining the model: “retrains the ‘learner’ iteratively by the ‘selected instance’ and its ‘label’” (Liu, section 2.1). A selected instance and its label represent a diluted data variation, so iterative retraining using the selected instances and labels amounts to retraining the model using the diluted data variations.
Applicant argues (pg. 24) that Zur teaches a method for noise injection, and thus fails to teach diluting data variations or adding the principal component of the data variations to each of the layers, as recited in the independent claim. Examiner respectfully notes that Zur is not relied on to teach these limitations. Diluting data variations is taught by Liu, and adding the principal component of the data variations to the training data is taught by Artursson,
Applicant argues (pg. 24) that while Artursson teaches identifying a principal component of data variability, Artursson fails to teach adding the principal component evenly to each training data layer as part of the dilution process. Examiner respectfully notes that Artursson (pg. 713) teaches adding the principal component evenly to the training data, and since the training data in Liu is represented in layers, it will be obvious to one of ordinary skill in the art that modifying the active learning framework of Liu with the principal component-based drift correction of Artursson would result in adding the principal component evenly to each layer of training data. As stated in the rejection below, one of ordinary skill in the art would have been motivated to make this modification because drift correction based on principal components “reduces the drift in such a way that the lifetime of the pattern recognition models is increased” (Artursson, pg. 712, paragraph 2).
Applicant argues (pg. 24) that Zur teaches noise injection, but not for the purpose of diluting data variations. Examiner respectfully notes that the noise injection step as currently claimed is not required to be for the purpose of diluting data variations.
Applicant argues (pg. 24) that Liu teaches labeling data, which is different from diluting data variations. Examiner respectfully notes that diluting data variations is not a term or concept known in the art, nor is it further defined in the specification, and thus removing the uncertainty associated with data variations by labeling them, as taught by Liu, falls within the broadest reasonable interpretation of diluting the data variations.
Applicant argues (pg. 25) that Artursson teaches away from “adding” the principal component to the training data because Artursson’s drift correction is expressed as a subtraction of the principal component. Examiner respectfully notes that while Artursson’s drift correction is written as a “subtraction” and the claimed drift dilution is written as “adding”, one of ordinary skill in the art will recognize that both Artursson and the instant application effectively perform the same operation to achieve the same technical effect: neutralizing data drift/variation by combining the data with the principal component of the drift using an additive operation.
The prior art rejections under 35 USC § 103 are maintained for the pending claims, and 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 § 112
Claims 1, 2, 5-9, 12-16, 19, and 20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claims 1, 8, and 15 recite the limitation “wherein confidence of the principal component is higher and a reason for a data mutation”. The term “higher” is a relative term which renders the claim indefinite. The term “higher” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. Further, the phrasing of the limitation seems to suggest that the principal component’s confidence is a reason for a data mutation, rather than the principal component itself. It is unclear how confidence associated with a principal component can be a reason for a data mutation. For examination purposes, the limitation will be interpreted as specifying that the principal component is confidently considered to be a reason for a data mutation.
Claims 2, 5-7, 9, 12-14, 16, 19, and 20 are additionally rejected due to their dependence on rejected claims 1, 8, and 15 for the reasons outlined above.
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 1-2, 6-9, 13-16, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over
Liu et al. (hereinafter Liu), “Active Learning on Dynamic Clustering for Drift Compensation in an Electronic Nose System” in view of
Zur et al. (hereinafter Zur), “Noise injection for training artificial neural networks: A comparison with weight decay and early stopping” and
Artursson et al. (hereinafter Artursson), “Drift correction for gas sensors using multivariate methods”.
Regarding Claim 1,
Liu teaches A computer-implemented method for training data models using machine learning, (pg. 9, section 4.1: “We imported the dataset and established the proposed methodology model in MATLAB (2014a). The computation was executed on a desktop computer…”)
Liu teaches the method comprising:
training a computer data model of data distribution using a training data set, the training data set including training data and additional training data, the training data and the additional training data being represented by layers of data representing the data distribution of the training data set; (pg. 5, section 2.4.1: “…a learner has initially been established by the training set with all kinds of samples.” The learner is a computer data model, and it has been established (i.e. trained) by the training set. Pg. 6, Algorithm 1, steps 1-2: “Compute the mean value m̄ks of each class based on (6); Cluster the instances in data pool according to m̄ks by (7) and (8) iteratively;” The data pool is additional training data, and the clusters are layers representing the data distribution.)
iteratively training the computer data model using the additional training data for each of the layers of the training data set; (pg. 3, section 2.1: “It is an obvious closed-loop structure, which retrains the ‘learner’ iteratively by the ‘selected instance’ and its ‘label’. The ‘selected instance’ is chosen from the ‘data pool’ full of drifted instances…” (i.e., the selected instance is an instance of additional training data). Pg. 6, section 2.4.3: “…the instances should be selected from each cluster alternately to maintain the balance of sample category of the training set” (i.e., additional training data is selected from each layer).)
detecting data variations in each of the layers of the additional training data; (pg. 6, section 2.4.3: “the instances should be selected from each cluster…Equations (10) makes larger informative (uncertainty) instances preferred in the selection…” Instances of additional training data with a high degree of uncertainty are data variations, and selecting these instances is detection of a data variation.)
diluting the data variations in each of the additional layers of the training data;(pg. 3, section 2.1: “The ‘experts’ implement manual labelling to provide a label to the selected instance.” Labeling the selected instance dilutes the data variation by removing the uncertainty associated with the additional training data.)
retraining the computer data model for the training data set using the diluted data variations in each of the layers of the additional training data. (pg. 6, Algorithm 1, step 5: “Renew the learner with the updated training set;”)
Liu does not appear to explicitly disclose adding statistical noise randomly to each of the layers of the training data set;
However, Zur teaches adding statistical noise randomly to each of the layers of the training data set; (pg. 4811-4812, section II.D: “The method of noise injection refers to adding ‘noise’ artificially to the ANN input data during the training process. Jitter is one particular method of implementing noise injection. With this method, a noise vector is added to each training case in between training iterations.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Liu and Zur. Liu teaches a method for drift compensation in an electronic nose system using active machine learning. Zur teaches noise injection during iterative training of a machine learning model. One of ordinary skill would have motivation to combine Liu and Zur because training machine learning models with noise injection “reduces overfitting and produces greater AUC [area under the receiver operating characteristic curve] values with smaller standard deviations compared with training ANNs without regularization” (Zur, pg. 4816, section IV).
Liu and Zur do not appear to explicitly disclose identifying a principal component of the data variations, wherein confidence of the principal component is higher and a reason for a data mutation; and
adding the principal component of the data variations to each of the layers of the training data set, as at least part of the diluting of the data variations in each of the additional layers of the training data set, wherein the principal component represents a direction of strong variability in data features for types of classifiers, and wherein the principal component is added evenly to each of the layers corresponding to categories to dilute a tendency caused by the principal component during classification.
However, Artursson teaches identifying a principal component of the data variations, wherein confidence of the principal component is higher and a reason for a data mutation; and (pg. 713, paragraph 1: “From the reference measurements a loading vector p is calculated by a PCA [principal component analysis]. The direction in the response space captured by the first loading vector can be attributed to the drift, since the gas sensors are exposed to the same gas and ought to give the same response with the exception of some random noise. This drift direction is assumed to be the same also for the samples. Projecting the samples X on the first loading
p
gives scores values t for the samples.” The data drift is a data variation/mutation, and the scores t are a principal component of that data variation. The direction captured by the principal component’s loading vector can be confidently attributed to the drift (i.e. confidence of the principal component is higher and a reason for a data mutation – see interpretation in light of rejection under 35 USC § 112(b) above).)
adding the principal component of the data variations to each of the layers of the training data set, as at least part of the diluting of the data variations in each of the additional layers of the training data set, wherein the principal component represents a direction of strong variability in data features for types of classifiers, and wherein the principal component is added evenly to each of the layers corresponding to categories to dilute a tendency caused by the principal component during classification. (pg. 712, paragraph 6: “The correlated information in the measurements often has some dominating types of variability that can be captured in terms of two smaller data matrices
P
and
T
. The loading matrix
P
is the projection matrix onto which the
X
data, e.g. measurements, have been projected, and the score matrix
T
is the object co-ordinate matrix in the space represented by the columns of
P
.” pg. 713, paragraph 2: “The drift correction is then expressed as a subtraction of the bilinear expression tpT from the original data: Xcorrected = X – tpT.” The principal component of the data variation is added to the original data (i.e. added evenly to each layer of training data) to correct the drift (i.e. dilute the data variation tendency) caused by the principal component. The principal components represented by matrices
P
and
T
capture the dominating types of variability in the data features used for classification (i.e. the principal component represents a direction of strong variability in the 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 Liu, Zur, and Artursson. Liu teaches a method for drift compensation in an electronic nose system using active machine learning. Zur teaches noise injection during iterative training of a machine learning model. Artursson teaches a method for drift correction in an electronic nose system using principal component analysis. One of ordinary skill would have motivation to combine Liu, Zur, and Artursson because drift correction based on principal components “reduces the drift in such a way that the lifetime of the pattern recognition models is increased” (Artursson, pg. 712, paragraph 2).
Regarding Claim 2, Liu, Zur, and Artursson teach The method of claim 1, as shown above.
Liu also teaches wherein the additional data is selected using parameters, for each of the layers of data, respectively. (pg. 6, section 2.4.3: “We define a binary flag vector f = {f1, f2,…, fK} and set all elements of f to 1, where K denotes the number of clusters. Then, we select the most valuable cluster… We set fk* = 0 to avoid successive selecting in the same cluster. Then, we pick up the finest instance x* in the most valuable cluster…” Flag vector f includes a parameter for each cluster (layer) of additional data, and the additional data is selected from the data pool based on these parameters.)
Regarding Claim 6, Liu, Zur, and Artursson teach The method of claim 1, as shown above.
Liu also teaches wherein the detecting of data variations in each of the layers of the additional training data includes detecting outlier data points in response to generating iterations of the computer model. (pg. 3, section 2.1: “The ‘instance selection strategy’ regularly determines the instance near the classification boundary with a certain rule for learner retraining.” Instances near the cluster classification boundary are outliers from their clusters. Figure 2(b) illustrates the identification and selection of these outliers, indicated by black circles on the outer edges of the cluster boundaries.)
Regarding Claim 7, Liu, Zur, and Artursson teach The method of claim 1, as shown above.
Zur also teaches wherein the statistical noise is Gaussian noise. (pg. 4812, section II.D: “The noise vector is typically drawn from some probability density function, known as a ‘kernel.’ We used a zero-mean Gaussian kernel…”)
Claims 8-9 and 13-14 are system claims, containing substantially the same elements as method claims 1-2 and 6-7, respectively. Liu, Zur, and Artursson teach the elements of claims 1-2 and 6-7, as shown above.
Liu also teaches A system for training data models using machine learning, which comprises: a computer system comprising; a computer processor, a computer-readable storage medium, and program instructions stored on the computer-readable storage medium being executable by the processor, to cause the computer system to perform the following functions (Examiner notes that this limitation is interpreted as a general-purpose computing environment. Liu, pg. 9, section 4.1: “We imported the dataset and established the proposed methodology model in MATLAB (2014a). The computation was executed on a desktop computer…”)
Claims 15-16 and 20 are product claims, containing substantially the same elements as method claims 1-2 and 6, respectively. Liu, Zur, and Artursson teach the elements of claims 1-2 and 6, as shown above.
Liu also teaches A computer program product for training data models using machine learning, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform functions, by the computer, comprising the functions (Examiner notes that this limitation is interpreted as a general-purpose computing environment. Liu, pg. 9, section 4.1: “We imported the dataset and established the proposed methodology model in MATLAB (2014a). The computation was executed on a desktop computer…”)
Claims 5, 12, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Liu in view of Zur and Artursson and further in view of
Cherian et al. (hereinafter Cherian), U.S. Patent Application Publication US 20210397970 A1.
Regarding Claim 5, Liu, Zur, and Artursson teach The method of claim 1, as shown above.
Liu, Zur, and Artursson do not appear to explicitly disclose wherein the adding of the statistical noise is implemented using an adversarial generation network, wherein a generator randomly generates the statistical noise and merges it with the parameters at each layer of the training data set.
However, Cherian teaches wherein the adding of the statistical noise is implemented using an adversarial generation network, wherein a generator randomly generates the statistical noise and merges it with the parameters at each layer of the training data set. ([0012]: “the extracted features are corrupted using noise data generated by an adversarial noise generator. The adversarial noise generator may be trained based on a Generative Adversarial Network (GAN). The GAN includes a generator and a discriminator. The adversarial noise generator generates the noise data from statistical distribution of the extracted features… The corrupted features are generated by combining the noise data with each replicated feature.” The replicated features are parameters of training data, and the generative adversarial network generates random statistical noise and merges it with these parameters.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Liu, Zur, Artursson, and Cherian. Liu teaches a method for drift compensation in an electronic nose system using active machine learning. Zur teaches noise injection during iterative training of a machine learning model. Artursson teaches a method for drift correction in an electronic nose system using principal component analysis. Cherian teaches improving classification accuracy of a machine learning model by corrupting training data with noise generated by a GAN. One of ordinary skill would have motivation to combine Liu, Zur, Artursson, and Cherian because corrupting training data via GAN improves robustness of the model “in real world applications, [where] presence of the perturbations in the input data may violate statistical assumptions” (Cherian, paragraph 0003).
Claim 12 is a system claim, containing substantially the same elements as method claim 5. Liu, Zur, Artursson, and Cherian teach the elements of claim 5, as shown above.
Claim 19 is a product claim, containing substantially the same elements as method claim 5. Liu, Zur, Artursson, and Cherian teach the elements of claim 5, as shown above.
Conclusion
Claims 1, 2, 5-9, 12-16, 19, and 20 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