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 .
The following action is in response to the original filing of 11/10/2022.
Claims 1-20 are pending and have been considered below.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to abstract ideas without significantly more.
Regarding claims 1, 10 and 19:
Step 1, MPEP 2106.03:
Claim 1. A method for detecting model performance [statutory category of invention]
Claim 10. An electronic device [statutory category of invention]
Claim 19. A computer program product that is tangibly stored on a non-transitory computer- readable medium and comprises machine-executable instructions [statutory category of invention]
Step 2A Prong One MPEP 2106.04, 2106.04(a):
determine a reconstruction error, the reconstruction error being a difference between the input feature before being reconstructed by the self-coding model and the input feature after being reconstructed by the self-coding model; [mathematical concepts such as mathematical relationships, mathematical formulas or equations, or mathematical calculations, MPEP 2106.04(a)(2), while broadly claimed, a BRI of the specification ¶32: determining reconstruction error rate of an encoder/decoder]
and determining a detection result of the target model at least based on a comparison between the confidence and a first threshold and a comparison between the reconstruction error and a second threshold; [mental processes such as concepts that can be practically performed in the human mind, or by a human using pen and paper as a physical aid, including observations, evaluations, judgments and opinions, MPEP 2106.04(a)(2)(III), for example a user judging a values have passed or not passed thresholds to be considered a result]
Step 2A Prong Two, MPEP 2106.04(d):
acquiring a prediction result of an input feature using a target model [represents mere data gathering, MPEP 2106.05]
to determine a confidence of the prediction result;
reconstructing the input feature using a self-coding model [represent mere instructions to apply, MPEP 2106.05]
Step 2B, MPEP 2106.05:
acquiring a prediction result of an input feature using a target model [insignificant extra-solution activity of data gathering/selecting a particular type of data, MPEP 2106.05(g)]
to determine a confidence of the prediction result;
reconstructing the input feature using a self-coding model [represent mere instructions to apply to obtain a solution/outcome, MPEP 2106.05(f)]
Regarding claims 2, 11 and 20:
Step 1, MPEP 2106.03:
Analysis of respective parent is incorporated.
Step 2A Prong One MPEP 2106.04, 2106.04(a):
wherein determining the detection result comprises:
determining the detection result as being normal in response to that the confidence is greater than the first threshold and the reconstruction error is less than the second threshold [mental processes such as concepts that can be practically performed in the human mind, or by a human using pen and paper as a physical aid, including observations, evaluations, judgments and opinions, MPEP 2106.04(a)(2)(III), for example a user judging a values have passed or not passed thresholds to be considered a result]
Step 2A Prong Two, MPEP 2106.04(d):
determining the detection result as being normal [represent mere instructions to apply, MPEP 2106.05]
Step 2B, MPEP 2106.05:
wherein each of the trajectories comprises information about the movement of one of the edge devices. [represent mere instructions to apply to obtain a solution/outcome, MPEP 2106.05(f)]
Regarding claims 3 and 12:
Step 1, MPEP 2106.03:
Analysis of respective parent is incorporated.
Step 2A Prong One MPEP 2106.04, 2106.04(a):
wherein determining the detection result comprises:
determining the detection result as a concept drift in response to that the confidence is greater than the first threshold, the reconstruction error is less than the second threshold, and the difference between the prediction result and a ground truth exceeds a predetermined range [mental processes such as concepts that can be practically performed in the human mind, or by a human using pen and paper as a physical aid, including observations, evaluations, judgments and opinions, MPEP 2106.04(a)(2)(III), for example a user judging a values have passed or not passed thresholds or ranges to be considered a result];
Step 2A Prong Two, MPEP 2106.04(d):
determining the detection result as a concept drift [represent mere instructions to apply, MPEP 2106.05]
and wherein the method further comprises:
retraining the target model [represent mere instructions to apply recited at a high level of generality, MPEP 2106.05]
on a training dataset that is different from a training dataset which was used for training the target mode [mere data gathering, MPEP 2106.05]
Step 2B, MPEP 2106.05:
determining the detection result as a concept drift [represent mere instructions to apply to obtain a solution/outcome, MPEP 2106.05(f)]
retraining the target model [represent mere instructions to apply recited at a high level of generality/generic computer tools for perform generic computing function, MPEP 2106.05(f)]
on a training dataset that is different from a training dataset which was used for training the target mode [insignificant extra-solution activity of data gathering/selecting a particular type of data, MPEP 2106.05(g)]
Regarding claims 4 and 13:
Step 1, MPEP 2106.03:
Analysis of respective parent is incorporated.
Step 2A Prong One MPEP 2106.04, 2106.04(a):
wherein determining the detection result comprises:
determining, in response to that the confidence is less than the first threshold and the reconstruction error is less than the second threshold, that the detection result indicates the target model being under fitted [mental processes such as concepts that can be practically performed in the human mind, or by a human using pen and paper as a physical aid, including observations, evaluations, judgments and opinions, MPEP 2106.04(a)(2)(III), for example a user judging a values have passed or not passed thresholds or ranges to be considered a result]
Step 2A Prong Two, MPEP 2106.04(d):
that the detection result indicates the target model being under fitted [represent mere instructions to apply, MPEP 2106.05]
and wherein the method further comprises:
decreasing a learning rate for training the target model; [mere data gathering, MPEP 2106.05] and
retraining the target model on a training dataset for training the target model at the decreased learning rate [represent mere instructions to apply recited at a high level of generality, MPEP 2106.05]
Step 2B, MPEP 2106.05:
that the detection result indicates the target model being under fitted [represent mere instructions to apply to obtain a solution/outcome, MPEP 2106.05(f)]
and wherein the method further comprises:
decreasing a learning rate for training the target model; [insignificant extra-solution activity of data gathering/selecting a particular type of data, MPEP 2106.05(g)]
and
retraining the target model on a training dataset for training the target model at the decreased learning rate [represent mere instructions to apply recited at a high level of generality/generic computer tools for perform generic computing function, MPEP 2106.05(f)]
Regarding claims 5 and 14:
Step 1, MPEP 2106.03:
Analysis of respective parent is incorporated.
Step 2A Prong One MPEP 2106.04, 2106.04(a):
further comprising:
determining a first Shapley value vector of the target model based on the input feature and the target model; and
determining a second Shapley value vector of the self-coding model based on the input feature and the self-coding model [mathematical concepts such as mathematical relationships, mathematical formulas or equations, or mathematical calculations, MPEP 2106.04(a)(2), specification ¶35]
Step 2A Prong Two, MPEP 2106.04(d):
All limitations are part of the abstract idea above.
Step 2B, MPEP 2106.05:
All limitations are part of the abstract idea above.
Regarding claims 6 and 15:
Step 1, MPEP 2106.03:
Analysis of respective parent is incorporated.
Step 2A Prong One MPEP 2106.04, 2106.04(a):
wherein determining the detection result comprises:
determining, in response to that the confidence is less than the first threshold and the reconstruction error is greater than the second threshold, that the detection result indicates appearance of a new feature pattern, [mental processes such as concepts that can be practically performed in the human mind, or by a human using pen and paper as a physical aid, including observations, evaluations, judgments and opinions, MPEP 2106.04(a)(2)(III), for example a user judging a values have passed or not passed thresholds or ranges to be considered a result]
determining the new feature pattern based on the second Shapley value vector; [mathematical concepts such as mathematical relationships, mathematical formulas or equations, or mathematical calculations, MPEP 2106.04(a)(2), specification ¶35, 43]
Step 2A Prong Two, MPEP 2106.04(d):
that the detection result indicates appearance of a new feature pattern [represent mere instructions to apply, MPEP 2106.05]
and wherein the method further comprises:
and incrementally training the target model on a training dataset that conforms to the new feature pattern [represent mere instructions to apply recited at a high level of generality, MPEP 2106.05]
Step 2B, MPEP 2106.05:
that the detection result indicates appearance of a new feature pattern [represent mere instructions to apply to obtain a solution/outcome, MPEP 2106.05(f)]
and wherein the method further comprises:
and incrementally training the target model on a training dataset that conforms to the new feature pattern [represent mere instructions to apply recited at a high level of generality/generic computer tools for perform generic computing function, MPEP 2106.05(f)]
Regarding claims 7 and 16:
Step 1, MPEP 2106.03:
Analysis of respective parent is incorporated.
Step 2A Prong One MPEP 2106.04, 2106.04(a):
wherein determining the detection result comprises:
when a similarity between the first Shapley value vector and the second Shapley value vector is greater than a third threshold, determining the detection result as a concept drift in response to that the confidence is greater than the first threshold and the reconstruction error is greater than the second threshold, [mental processes such as concepts that can be practically performed in the human mind, or by a human using pen and paper as a physical aid, including observations, evaluations, judgments and opinions, MPEP 2106.04(a)(2)(III), for example a user judging a values have passed or not passed thresholds or ranges to be considered a result]
Step 2A Prong Two, MPEP 2106.04(d):
and wherein the method further comprises:
retraining the target model on a training dataset that is different from a training dataset which was used for training the target model [represent mere instructions to apply recited at a high level of generality, MPEP 2106.05]
Step 2B, MPEP 2106.05:
and wherein the method further comprises:
retraining the target model on a training dataset that is different from a training dataset which was used for training the target model [represent mere instructions to apply recited at a high level of generality/generic computer tools for perform generic computing function, MPEP 2106.05(f)]
Regarding claims 8 and 17:
Step 1, MPEP 2106.03:
Analysis of respective parent is incorporated.
Step 2A Prong One MPEP 2106.04, 2106.04(a):
wherein determining the detection result further comprises:
determining the detection result as being undetermined when the similarity between the first Shapley value vector and the second Shapley value vector is less than the third threshold [mental processes such as concepts that can be practically performed in the human mind, or by a human using pen and paper as a physical aid, including observations, evaluations, judgments and opinions, MPEP 2106.04(a)(2)(III), for example a user judging a values have passed or not passed thresholds or ranges to be considered a result]
Step 2A Prong Two, MPEP 2106.04(d):
All limitations are part of the abstract idea above.
Step 2B, MPEP 2106.05:
All limitations are part of the abstract idea above.
Regarding claims 9 and 18:
Step 1, MPEP 2106.03:
Analysis of respective parent is incorporated.
Step 2A Prong One MPEP 2106.04, 2106.04(a):
Analysis of respective parent is incorporated.
Step 2A Prong Two, MPEP 2106.04(d):
wherein the first Shapley value vector and the second Shapley value vector are both determined with a model visualization tool [represents mere instructions to implement the abstract idea using generic computing tools, MPEP 2106.05]
Step 2B, MPEP 2106.05:
wherein the first Shapley value vector and the second Shapley value vector are both determined with a model visualization tool [represent mere instructions to apply recited at a high level of generality/generic computer tools for perform generic computing function, MPEP 2106.05(f)]
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1-3, 10-12 and 19-20 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Xu, Yiming, and Diego Klabjan. "Concept Drift and Covariate Shift Detection Ensemble with Lagged Labels." arXiv preprint arXiv:2012.04759 (2020) [“XU”].
Regarding claim 1, XU discloses a method for detecting model performance (XU page 2 ¶7: “The proposed method can detect both concept drift and covariate shift; it can determine when to retrain and what data to use to retrain automatically, by utilizing an ensemble of six different drift detectors.”), comprising:
acquiring a prediction result of an input feature using a target model to determine a confidence of the prediction result (XU pages 5 ¶5, pages 6 ¶: “Model uncertainty: It is also helpful to understand how confident the model is regarding the predictions. In the second score, we first predict every sample in the current batch Xn to obtain the predicted probability distributions. For each of these distributions, we construct the histogram of the largest probability and the histogram of the second largest probability, and fit each histogram using a Gaussian distribution to obtain probability density functions N1,N2. The model uncertainty score is obtained by
PNG
media_image1.png
57
202
media_image1.png
Greyscale
” – acquiring model uncertainty score q2 for input features of current batch Xn, confidence is reflected as inverse of model uncertainty score);
reconstructing the input feature using a self-coding model to determine a reconstruction error, the reconstruction error being a difference between the input feature before being reconstructed by the self-coding model and the input feature after being reconstructed by the self-coding model (XU page 6 ¶4: “Auto-encoder reconstruction error: Auto-encoders are widely used in tasks such as dimension reduction and embedding learning. In our work, we employ an auto-encoder to measure how different the two datasets or batches are. We first train an auto-encoder using training features Xtr and obtain the training reconstruction MSE loss Ltr = MSE(Xtr,AE(Xtr)). For the current batch Xn, we calculate the test reconstruction loss Ln = MSE(Xn,AE(Xn)). The auto-encoder reconstruction score is defined as
PNG
media_image2.png
46
140
media_image2.png
Greyscale
” – using an autoencoder to obtain reconstruction loss and provide a reconstruction score q4); and
determining a detection result of the target model at least based on (XU page 7 ¶6: “Each descriptive statistic defined in the previous subsection generates a time series over time. The remaining problem is how to detect drift based on these time series. We describe next for the current batch Xn, the resulting algorithm to report drift, warning or safe signals, which utilizes an ensemble of six independent drift detection modules to monitor drift.”, page 8 ¶2: “To detect a drift event, we employ a tailored majority voting rule as follows: if the drift detection module on q1 reports drift, then the system reports drift and retrains all models (classifier, auto-encoder and SPN), as the KPI is the ultimate goal; otherwise, if most of the remaining modules (i.e. not less than 3 modules) report a drift, CDCSDE reports drift and retrains all models.” – determining a drift detection result based on ensemble including at least the model uncertainty score and reconstruction score) a comparison between the confidence and a first threshold (XU page 6 ¶2: “When the overlapping area increases, the mean and variance of the two Gaussians are close to each other, indicating that the largest probability and the second largest probability from predictions are very similar and thus the model is more uncertain regarding the predictions. A significant increase of q2 is a signal for drift, since when the model predictions are uncertain, the likelihood of a change in distribution is high.” – thresholds are passed when considering a score as higher or lower, confidence is reflected as inverse of model uncertainty score, i.e. a higher model uncertainty is a lower confidence in the prediction result and lower model uncertainty is a higher confidence in prediction result) and a comparison between the reconstruction error and a second threshold (XU page 6 ¶4: “This score allows us to measure divergence by how large the test reconstruction error is compared to the training error. The increase of q4 indicates that the auto-encoder is unable to fully reconstruct the incoming data and thus the covariate shift has occurred.” – a higher reconstruction score indicates higher error and a lower reconstruction score indicates a lower error).
Regarding claim 2, XU discloses the method according to claim 1, wherein determining the detection result comprises:
determining the detection result as being normal in response to that the confidence is greater than the first threshold and the reconstruction error is less than the second threshold (XU page 6 ¶2: Model Uncertainty lower than threshold – high confidence, normal result, page 6 ¶4: Auto-encoder reconstruction error – lower than threshold, normal result).
Regarding claim 3, XU discloses the method according to claim 1, wherein determining the detection result comprises:
determining the detection result as a concept drift in response to that the confidence is greater than the first threshold and the reconstruction error is less than the second threshold (XU page 6 ¶2: Model Uncertainty lower than threshold – high confidence, normal result, page 6 ¶4: Auto-encoder reconstruction error – lower than threshold, normal result), and the difference between the prediction result and a ground truth exceeds a predetermined range (XU page 5 ¶3-4: “EWMA of a delayed classification indicator: Let kpi(Y,X) be the most important key performance indicator of cl f, e.g. error rate or 1 - F1. We calculate the exponentially weighted moving average of the delayed KPI as follows. Assume we trace back k batches to calculate the moving average, i.e. at time n, we use the KPI of batches n-k+1, n-k+2,..., n, as the labels are delayed by l based on our assumption. For weight decay w, 0 < w < 1, the score is calculated as
PNG
media_image3.png
65
362
media_image3.png
Greyscale
[..] Intuitively, when EWMA increases significantly, the model suffers from performance degradation and the drift has likely occurred.”); and
wherein the method further comprises:
retraining the target model on a training dataset that is different from a training dataset which was used for training the target model (XU page 8 ¶2: “To detect a drift event, we employ a tailored majority voting rule as follows: if the drift detection module on q1 reports drift, then the system reports drift and retrains all models (classifier, auto-encoder and SPN), as the KPI is the ultimate goal; otherwise, if most of the remaining modules (i.e. not less than 3 modules) report a drift, CDCSDE reports drift and retrains all models.”, ¶3: “If drift is reported, the data used to retrain is determined by CDCSDE as well. Suppose the latest warning zone (if a module exits the warning zone, then the previous warning zone is not included in future retraining) for each module is Wi, i = 1, 2, …, 6. The union of these warning batches
PNG
media_image4.png
49
55
media_image4.png
Greyscale
is used to retrain all models.”).
Regarding claims 10-12, claims 10-12 recite limitations similar to claims 1-3, respectively, and are similarly rejected.
Regarding claims 19-20, claims 19-20 recite limitations similar to claims 1-2, respectively, and are similarly rejected.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 4 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over XU in view of Gheibi, Omid, and Danny Weyns. "Lifelong self-adaptation: Self-adaptation meets lifelong machine learning." Proceedings of the 17th symposium on software engineering for adaptive and self-managing systems. (Aug 2022) [“GHEIBI”].
Regarding claim 4, XU discloses the method according to claim 1, wherein determining the detection result comprises:
determining, in response to that the confidence is less than the first threshold and the reconstruction error is less than the second threshold (page 6 ¶2: Model Uncertainty higher than threshold – low confidence, abnormal result, page 6 ¶4: Auto-encoder reconstruction error – lower than threshold, normal result), that the detection result indicates the target model being under fitted (page 6 ¶3: Hellinger distance – higher than threshold, abnormal result underfitted dataset), and
wherein the method further comprises:
retraining the target model on a training dataset for training the target model (page 8 ¶2: Drift detection - retrain if majority of modules report drift).
XU fails to disclose wherein the retraining of the target model is at a decreased learning rate.
GHEIBI discloses methods for determining drift in predictive models (GHEIBI page 1 col 1 ¶1: “We present a reusable architecture for lifelong self-adaptation and apply it to the case of concept drift caused by unforeseen changes of the input data of a learning model that is used for decision-making in self-adaptation.”). In particular, GHEIBI discloses retraining the model with a decreased learning rate (GHEIBI page 5 col 1 ¶3: “An SGD regressor estimates the gradient of the loss for each data sample and uses that to update the learning model with a decreasing learning rate.”). Therefore it would have been obvious to one having ordinary skill in the art and the teachings of XU and GHEIBI before them before the effective filing of the claimed invention to combine the retraining of the model by decreasing the learning rate, as taught by GHEIBI, with the retraining of the model when the low confidence and low reconstruction error yields an underfitted determination of the model of XU. One would have been motivated to make this combination to provide minimal packet loss methods of dealing with sudden co-variate drift, as suggested by GHEIBI (GHEIBI pages 4-6, Figure 6).
Regarding claim 13, claim 13 recites limitations similar to claim 4 and is similarly rejected.
Claims 5-9 and 14-18 are rejected under 35 U.S.C. 103 as being unpatentable over XU in view of Zheng, Shihao & Zon, Simon & Pechenizkiy, Mykola & Campos, Cassio & Ipenburg, Werner & Harder, Hennie.“Labelless Concept Drift Detection and Explanation.”, NeurIPS 2019 Workshop on Robust AI in Financial Services (2019) [“ZHENG”].
Regarding claim 5, XU discloses the method according to claim 1, and further discloses determining vectors for the target model and self-coding model based on the input feature (XU page 6 ¶3: Hellinger distance – encoded feature vectors).
XU fails to disclose the vectors are Shapley value vectors.
ZHENG discloses methods for determining drift in predictive models (ZHENG page 1 Abstract). In particular, ZHENG discloses determining Shapley value vectors for a classifier and explainer model (ZHENG page 2 ¶5: “Let (X1, …, XN) be a sequence of data with labels (y1, …, yN) to be used to train an initial classifier
f
^
, where X has d dimensions, y is restricted to binary classes and N is the predefined chunk size. Next, an explainer g is obtained based on
f
^
(we use the SHAP library [12]). Classifier
f
^
and explainer g stay fixed until any model adaptation takes place. By inputting X1, X2, … into g, corresponding Shapley values S1, S2, … can be obtained and Shapley feature space formed.”). Therefore it would have been obvious to one having ordinary skill in the art and the teachings of XU and ZHENG before them before the effective filing of the claimed invention to combine the use of Shapley value vectors in a classifier/explainer model system, as taught by ZHENG, with the use of vectors in the target model/self-coding model of XU, predictably resulting in one of the metrics of XU used to determine drift involving the Shapley value vectors of the models. One would have been motivated to make this combination to provide efficient root cause analysis in drift detection systems, as suggested by ZHENG (ZHENG page 1 ¶2-3).
Regarding claim 6, XU and ZHENG disclose the method according to claim 5, wherein determining the detection result comprises:
determining, in response to that the confidence is less than the first threshold and the reconstruction error is greater than the second threshold (XU page 6 ¶2: Model Uncertainty higher than threshold – low confidence, abnormal result, page 6 ¶4: Auto-encoder reconstruction error – higher than threshold, abnormal result), that the detection result indicates appearance of a new feature pattern (XU page 6 ¶3: Hellinger distances – detecting new feature patterns in dataset), and
wherein the method further comprises:
determining the new feature pattern based on the second Shapley value vector (ZHENG page 2 ¶4: “Several properties of the Shapley values justify their use for tracking concept drift in a univariate manner: 1) combining local explanations of each prediction allows us to capture global patterns between feature space X and the predictions ŷ.”); and
incrementally training the target model on a training dataset that conforms to the new feature pattern (XU page 8 ¶3 - training on new warning zone dataset).
Regarding claim 7, XU and ZHENG disclose the method according to claim 5, wherein determining the detection result comprises:
when a similarity between the first Shapley value vector and the second Shapley value vector is greater than a third threshold (ZHENG page 2 ¶6: “If there is a significant difference between (XRef, SRef) and (XCur, SCur), then we say that t is a drift position and XCur is a chunk with drift.”), determining the detection result as a concept drift in response to that the confidence is greater than the first threshold and the reconstruction error is greater than the second threshold (XU page 6 ¶2: Model Uncertainty lower than threshold – high confidence, normal result, page 6 ¶4: Auto-encoder reconstruction error – higher than threshold, abnormal result), and
wherein the method further comprises:
retraining the target model on a training dataset that is different from a training dataset which was used for training the target model (XU page 8 ¶2: “otherwise, if most of the remaining modules (i.e. not less than 3 modules) report a drift, CDCSDE reports drift and retrains all models”.
Regarding claim 8, XU and ZHENG disclose the method according to claim 7, wherein determining the detection result further comprises:
determining the detection result as being undetermined when the similarity between the first Shapley value vector and the second Shapley value vector is less than the third threshold (ZHENG page 2 ¶6: “If there was no drift alert, the reference window will be extended by adding (XCur, SCur).”).
Regarding claim 9, XU and ZHENG disclose the method according to claim 5, wherein the first Shapley value vector and the second Shapley value vector are both determined with a model visualization tool (ZHENG page 3 ¶7: “Explanation of detected drift. L-CODE provides three levels of drift information to assist end-users (see Figure 2).”).
Regarding claims 14-18, claims 14-18 recite limitations similar to claims 5-9, respectively, and are similarly rejected.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Wang; Weichen et al.
US 11768917 B2
SYSTEMS AND METHODS FOR ALERTING TO MODEL DEGRADATION BASED ON DISTRIBUTION ANALYSIS
Yu; Ye et al.
US 12019987 B1
SYSTEMS AND METHODS FOR FLEXIBLE REGULARIZED DISTILLATION OF NATURAL LANGUAGE PROCESSING MODELS TO FACILITATE INTERPRETATION
Shim; Dongsub et al.
US 12406483 B2
ONLINE CLASS-INCREMENTAL CONTINUAL LEARNING WITH ADVERSARIAL SHAPLEY VALUE
Zhang; Jun et al.
US 20160155136 A1
AUTO-ENCODER ENHANCED SELF-DIAGNOSTIC COMPONENTS FOR MODEL MONITORING
Draelos; Timothy J. et al.
US 20170177993 A1
ADAPTIVE NEURAL NETWORK MANAGEMENT SYSTEM
Erlandson; Erik J. et al.
US 20190147357 A1
AUTOMATIC DETECTION OF LEARNING MODEL DRIFT
Ghanta; Sindhu et al.
US 20200193313 A1
INTERPRETABILITY-BASED MACHINE LEARNING ADJUSTMENT DURING PRODUCTION
Che; Tong et al.
US 20200372339 A1
SYSTEMS AND METHODS FOR VERIFICATION OF DISCRIMINATIVE MODELS
Fu; Shengyu et al.
US 20200410390 A1
MACHINE LEARNING RETRAINING
Kolar; Vinay Kumar et al.
US 20210184958 A1
ANOMALY DETECTION OF MODEL PERFORMANCE IN AN MLOPS PLATFORM
Datta; Anupam et al.
US 20220012613 A1
SYSTEM AND METHOD FOR EVALUATING MACHINE LEARNING MODEL BEHAVIOR OVER DATA SEGMENTS
Allahdadian; Saeid et al.
US 20220156578 A1
STATISTICAL CONFIDENCE METRIC FOR RECONSTRUCTIVE ANOMALY DETECTION MODELS
Dalli; Angelo et al.
US 20220172050 A1
METHOD FOR AN EXPLAINABLE AUTOENCODER AND AN EXPLAINABLE GENERATIVE ADVERSARIAL NETWORK
Katsuki; Takayuki et al.
US 20220180204 A1
ADVERSARIAL SEMI-SUPERVISED ONE-SHOT LEARNING
Allahdadian; Saeid et al.
US 20220188410 A1
COPING WITH FEATURE ERROR SUPPRESSION: A MECHANISM TO HANDLE THE CONCEPT DRIFT
Mopur; Satish Kumar et al.
US 20220215289 A1
MANAGING DATA DRIFT AND OUTLIERS FOR MACHINE LEARNING MODELS TRAINED FOR IMAGE CLASSIFICATION
Vishwakarma; Rahul Deo et al.
US 20220230083 A1
STOCHASTIC RISK SCORING WITH COUNTERFACTUAL ANALYSIS FOR STORAGE CAPACITY
Rowe; Matthew Charles et al.
US 20220366280 A1
GENERATING CONFIDENCE SCORES FOR MACHINE LEARNING MODEL PREDICTIONS
Pushak; Yasha et al.
US 20220366297 A1
LOCAL PERMUTATION IMPORTANCE: A STABLE, LINEAR-TIME LOCAL MACHINE LEARNING FEATURE ATTRIBUTOR
Casserini; Matteo et al.
US 20230024884 A1
BALANCING FEATURE DISTRIBUTIONS USING AN IMPORTANCE FACTOR
Koulierakis; Eleftherios et al.
US 20230252347 A1
METHOD AND APPARATUS FOR CONCEPT DRIFT MITIGATION
Rama; Kiran et al.
US 20230316045 A1
DRIFT DETECTION USING AN AUTOENCODER WITH WEIGHTED LOSS
Teixeira Nogueira; Joao Gabriel et al.
US 20240086704 A1
MODEL UNDERSTANDABILITY
Fu; Chunyan et al.
US 20240362472 A1
AUTOMATED HANDLING OF DATA DRIFT IN EDGE CLOUD ENVIRONMENTS
Zhang; Cheng et al.
US 20250036947 A1
AUXILIARY MODEL FOR PREDICTING NEW MODEL PARAMETERS
Ross, Gordon J., et al. "Exponentially weighted moving average charts for detecting concept drift." Pattern recognition letters 33.2 (2012): 191-198.
Lundberg, Scott M., and Su-In Lee. "A unified approach to interpreting model predictions." Advances in neural information processing systems 30 (2017).
Demšar, Jaka, and Zoran Bosnić. "Detecting concept drift in data streams using model explanation." Expert Systems with Applications 92 (2018): 546-559.
Madireddy, Sandeep, et al. "Adaptive learning for concept drift in application performance modeling." Proceedings of the 48th International Conference on Parallel Processing. 2019.
Forward, Cloudera Fast. "Deep Learning for Anomaly Detection." URL: https://ff12. fastforwardlabs. com/(Last accessed: 24. 11. 2021).
Suprem, Abhijit, et al. "Odin: Automated drift detection and recovery in video analytics." arXiv preprint arXiv:2009.05440 (2020).
Halstead, Ben, et al. "Fingerprinting concepts in data streams with supervised and unsupervised meta-information." 2021 IEEE 37th International Conference on Data Engineering (ICDE). IEEE, 2021.
Hardt, Michaela, et al. "Amazon sagemaker clarify: Machine learning bias detection and explainability in the cloud." Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 2021.
Jacob, Vincent, et al. "Exathlon: A benchmark for explainable anomaly detection over time series." arXiv preprint arXiv:2010.05073 (2020).
Shanbhag, Aalok, Avijit Ghosh, and Josh Rubin. "Unified shapley framework to explain prediction drift." arXiv preprint arXiv:2102.07862 (2021).
Souza, Vinicius MA, et al. "Efficient unsupervised drift detector for fast and high-dimensional data streams." Knowledge and Information Systems 63.6 (2021): 1497-1527.
Stocco, Andrea, and Paolo Tonella. "Confidence‐driven weighted retraining for predicting safety‐critical failures in autonomous driving systems." Journal of Software: Evolution and Process 34.10 (2022): e2386.
Li, Bin, Chiara Balestra, and Emmanuel Müller. "Enabling the visualization of distributional shift using shapley values." NeurIPS 2022 Workshop on Distribution Shifts: Connecting Methods and Applications. 2022.
Che, Tong, et al. "Deep verifier networks: Verification of deep discriminative models with deep generative models." Proceedings of the AAAI conference on artificial intelligence. Vol. 35. No. 8. 2021.
Roshan, Khushnaseeb, and Aasim Zafar. "Using kernel shap xai method to optimize the network anomaly detection model." 2022 9th International Conference on Computing for Sustainable Global Development (INDIACom). IEEE, 2022.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ANDREW L TANK whose telephone number is (571)270-1692. The examiner can normally be reached Monday-Thursday 9a-6p.
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, Matthew Ell can be reached at 571-270-3264. 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.
/ANDREW L TANK/ Primary Examiner, Art Unit 2141