UNSUPERVISED MODEL DRIFT ESTIMATION SYSTEM FOR DATASET SHIFT DETECTION AND MODEL SELECTION

Application No. 18/573,527

• Filed: Dec 22, 2023
• Priority: Nov 03, 2021 — nonprovisional of PCTKR2021015819
• Examiner: TRIEU, EM N
• Assignee: Intel Corporation
• OA Round: 1 (Non-Final)

Office 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 .
DETAILED ACTION
This office action is in response to the claims filed on 12/22/2023.
Claims 1-25 are presented for examination.
Priority
The following claimed benefit is acknowledged: the instant application, filed 12/22/2023 claims priority from foreign application PCT/KR2021/015819, filed 11/03/2021.
Information Disclosure Statement
The information disclosure statements (IDS) filed12/22/2023 is in compliance with the provisions of 37 CFR 1.97 and 1.98. Accordingly, the information disclosure statement is being considered by the examiner.

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.

The claims 14-19 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. Specifically, claims 14-19 are toward " computer readable storage medium”. The broadest reasonable interpretation of “computer readable storage medium” covers transitory propagating signals, which are non-statutory. The disclosure in [0037] discloses that “The method 320 may be implemented in one or more modules as a set of logic instructions stored in a machine- or computer-readable storage medium such as RAM, ROM, PROM, firmware, flash memory, etc., in configurable logic such as, for example, PLAs, FPGAs, CPLDs, in fixed- functionality logic hardware using circuit technology such as, for example, ASIC, CMOS or TTL technology, or any combination thereof.” Although this disclosure may exclude “transmission media such as wireless signals and the like” from “a storage media,” it does not make it clear that “a storage device” is a “storage media,” and thus “a storage device” may exclude “transmission media such as wireless signals and the like.” Further, even if “a storage device” were considered as a “storage media,” the statement only explicitly excludes “wireless signals and the like.” It is unclear whether other types of signal, e.g., signal in wired transmission, is excluded from a storage media.
Therefore, a broadest reasonable interpretation of claims 14-19 covers a transitory signal. When the broadest reasonable interpretation of a claim covers a signal per se, the claim must be rejected under 35 U.S.C. 101 as covering non-statutory subject matter. See In re Nuijten, 500 F.3d 1346, 1356-57 (Fed. Cir. 2007) (transitory embodiments are not directed to statutory subject matter); MPEP 9th Ed., § 2106.I. To overcome this rejection, applicant should insert –- non-transitory — before “machine readable storage device”. Such an amendment is not considered new matter. See the “Subject Matter Eligibility of Computer Readable Media” memo dated January 26, 2010 (OG Cite: 1351 OG 212; OG Date: 23 Feb 2010).

Claims 1-25 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 analysis:
In the instant case, the claims are directed to a system (claims1-6), apparatus (claims 7-13) and the method (claims 20-25). Thus, each of the claims falls within one of the four statutory categories (i.e., process, machine, manufacture, or composition of matter). However, the claims 14-19 recite the computer readable storage medium, these are not fall within one of the four statutory categories ((i.e., process, machine, manufacture, or composition of matter). While the claims 14-19 are not fall within one of the four statutory categories ((i.e., process, machine, manufacture, or composition of matter), they can be amended to be.
Step 2A analysis:
Based on the claims being determined to be within of the four categories (Step 1), it must be determined if the claims are directed to a judicial exception (i.e., law of nature, natural phenomenon, and abstract idea), in this case the claims fall within the judicial exception of an abstract idea. Specifically the abstract idea of “Mental Processes/Concepts performed in the human mind (including an observation, evaluation, judgment, opinion)” and mathematical concept.

The claim 1 recites:
Step 2A: prong 1 analysis:
-“ identify a first neural network, This is a mental process, the human mind can identify/select particular neural network during the training of the neural network (observation/Evaluation)
-“ estimate a first drift of the first neural network based on the first neural network process, the first training parameter and the first population data to determine whether to retrain the first neural network.” This is a mental process, as the human can estimate the drift of the neural network, such as the error rate of the training of the neural network based on parameter and the particular dataset, is increasing, then the developer can retrain the neural network (observation/Evaluation).
a) Step 2A: Prong 2 analysis:
-“ a network controller to communicate with edge nodes; a processor coupled to the network controller; and a memory coupled to the processor, the memory including a set of executable program instructions, which when executed by the processor, cause the computing system to:” The additional limitation is recited at high level of generality and amounts to no more than mere instructions to apply the judicial exception using a generic computer component (See MPEP 2106.05(f)).
-“ wherein the first neural network is associated with a first training parameter and first population data that are generated during a process to train the first neural network”, “ execute a first neural network process to serve input data with the first neural network” These/this additional limitation(s) are/is recited at a high-level of generality such that it amounts to necessary data gathering, gathering the data that is necessary to perform the "estimate" mental process. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity of data gathering to a judicial exception do not amount to significantly more than the judicial exception and cannot integrate a judicial exception into a practical application.
b) Step 2B analysis:
-“ a network controller to communicate with edge nodes; a processor coupled to the network controller; and a memory coupled to the processor, the memory including a set of executable program instructions, which when executed by the processor, cause the computing system to:” The additional limitation is recited at high level of generality and amounts to no more than mere instructions to apply the judicial exception using a generic computer component (See MPEP 2106.05(f)).
-“ wherein the first neural network is associated with a first training parameter and first population data that are generated during a process to train the first neural network”, -“ execute a first neural network process to serve input data with the first neural network” These/this limitation(s) are/is recited at a high-level of generality such that it amounts to necessary data gathering in order to perform the abstract idea steps. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity of data gathering to a judicial exception do not amount to significantly more than the judicial exception itself.
Further, the specification of the instant application teaches that training a neural network on training data and executing that neural network are well-understood, routine, and conventional actions ([0002], "Some deep learning methods optimize neural networks on training data ... performance ... in real-world applications" denotes that training and executing neural networks is a common action). Well-known, routine, and conventional activity cannot provide an inventive concept nor significantly more than the abstract idea itself.
The claim 2 recites:
Step 2A: prong 1 analysis:
-“ determine a plurality of distances for a plurality of layers of the first neural network based on the first neural network process, the first training parameter and the first population data” this is a mental process, the human can determine the distance between the layer of the neural network ( observation/Evaluation)
-“and estimate the first drift based on the plurality of distances.” This is a mental process, the human mind can estimate the drift based on the distances, for example, if the distance of the input and the output is higher than the threshold, then the drift is occurred (Observation/Evaluation).
Step 2A: Prong 2 analysis and Step 2B analysis
No additional element that provides a practical application or amount to significantly more than the abstract idea.
The claim 3 recites:
Step 2A: prong 1 analysis:
-“ identify a second neural network” this is a mental process, the human can identify the second neural network (observation/Evaluation).
- “estimate a second drift of the second neural network based on the second neural network process, the second training parameter and the second population data” this is a mental process, the human can estimate the drift based on the particular parameter and the particular set of datasets, (observation/evaluation).
-“and select one of the first neural network and the second neural network for deployment based on the first drift and the second drift.” This is a mental process, the human can select the first or the second neural network to deploy based on the identified drift, (observation/Evaluation).
a) Step 2A: Prong 2 analysis:
-“ wherein the second neural network is associated with a second training parameter and second population data that are generated during a process to train the second neural network; The additional limitation is recited at high level of generality and amounts to no more than mere instructions to apply the judicial exception using a generic computer component (See MPEP 2106.05(f)).
“execute a second neural network process to serve the input data with the second neural network;” These/this additional limitation(s) are/is recited at a high-level of generality such that it amounts to necessary data gathering, gathering the data that is necessary to perform the "estimate" mental process. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity of data gathering to a judicial exception do not amount to significantly more than the judicial exception and cannot integrate a judicial exception into a practical application.
b) Step 2B analysis:
-“ wherein the second neural network is associated with a second training parameter and second population data that are generated during a process to train the second neural network; The additional limitation is recited at high level of generality and amounts to no more than mere instructions to apply the judicial exception using a generic computer component (See MPEP 2106.05(f)).
“execute a second neural network process to serve the input data with the second neural network;” These/this limitation(s) are/is recited at a high-level of generality such that it amounts to necessary data gathering. Gathering the data that is necessary to perform the "estimate" mental process. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity of data gathering to a judicial exception do not amount to significantly more than the judicial exception itself.
Further, the specification of the instant application teaches that training a neural network on training data and executing that neural network are well-understood, routine, and conventional actions ([0002], "Some deep learning methods optimize neural networks on training data ... performance ... in real-world applications" denotes that training and executing neural networks is a common action). Well-known, routine, and conventional activity cannot provide an inventive concept nor significantly more than the abstract idea itself.
The claim 4 recites:
Step 2A: prong 1 analysis:
-“ determine whether to retrain the first neural network based on a comparison of the first drift to a threshold” this is a mental process, the human mind can determine when the retrained is needed based on the comparison of the drift to the threshold, (observation/Evaluation).
Step 2A: Prong 2 analysis and Step 2B analysis
No additional element that provides a practical application or amount to significantly more than the abstract idea.
The claim 5 recites:
a) Step 2A: Prong 2 analysis:
-“ the first neural network includes a plurality of batch normalization layers; the first training parameter is to include a scale parameter and a shift parameter associated with the plurality of batch normalization layers, wherein the scale parameter and the shift parameter are generated during the process to train the first neural network; and the first population data is to include a moving mean and moving standard deviation generated during the process to train the first neural network.” These/this additional limitation(s) are/is recited at a high-level of generality such that it amounts to necessary data gathering. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity of data gathering to a judicial exception do not amount to significantly more than the judicial exception and cannot integrate a judicial exception into a practical application.
b) Step 2B analysis:
-“ the first neural network includes a plurality of batch normalization layers; the first training parameter is to include a scale parameter and a shift parameter associated with the plurality of batch normalization layers, wherein the scale parameter and the shift parameter are generated during the process to train the first neural network; and the first population data is to include a moving mean and moving standard deviation generated during the process to train the first neural network.” These/this limitation(s) are/is recited at a high-level of generality such that it amounts to necessary data gathering. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity of data gathering to a judicial exception do not amount to significantly more than the judicial exception itself.
The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
The claim 6 recites:
Step 2A: prong 1 analysis:
“ estimate the first drift of the first neural network based on an input vector into a batch normalization layer of the first neural network and an output vector of the batch normalization layer” this is a mental process, the human can estimate the drift of the first neural network based on the input vector, for example, the human can tell if the drift is occurred based on the input data, (observation/Evaluation).
Step 2A: Prong 2 analysis and Step 2B analysis
No additional element that provides a practical application or amount to significantly more than the abstract idea.
The claims 7-12 are rejected for the same reason as the claims 1-6, since these claims recite the same limitations.
The claim 7 further recites
a) Step 2A: Prong 2 analysis:
“a semiconductor apparatus comprising: one or more substrates; and logic coupled to the one or more substrates, wherein the logic is implemented in one or more of configurable or fixed-functionality hardware, the logic to:” The additional limitation is recited at high level of generality and amounts to no more than mere instructions to apply the judicial exception using a generic computer component (See MPEP 2106.05(f)).
b) Step 2B analysis:
“a semiconductor apparatus comprising: one or more substrates; and logic coupled to the one or more substrates, wherein the logic is implemented in one or more of configurable or fixed-functionality hardware, the logic to:” The additional limitation is recited at high level of generality and amounts to no more than mere instructions to apply the judicial exception using a generic computer component (See MPEP 2106.05(f)).
The claim 13 recites:
Step 2A: Prong 2 analysis:
“wherein the logic coupled to the one or more substrates includes transistor channel regions that are positioned within the one or more substrates.” These/this additional limitation(s) are/is recited at a high-level of generality such that it amounts to necessary data gathering. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity of data gathering to a judicial exception do not amount to significantly more than the judicial exception and cannot integrate a judicial exception into a practical application.
b) Step 2B analysis:
“wherein the logic coupled to the one or more substrates includes transistor channel regions that are positioned within the one or more substrates.” These/this limitation(s) are/is recited at a high-level of generality such that it amounts to necessary data gathering. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity of data gathering to a judicial exception do not amount to significantly more than the judicial exception itself.
The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
The claims 14-19 are rejected for the same reason as the claims 1-6, since these claims recite the same limitations.
The claims 20-25 are rejected for the same reason as the claims 1-6, since these claims recite the same limitations.

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 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.
Claims 1-3, 14-16, 20-22 are rejected under 35 U.S.C. 103 as being unpatentable over Farchi et al. (Pub. No. US 20200342310-hereinafter, Farchi) and further in view of Ni et al. (Patent. No. US 12406181 -hereinafter, Ni).
Regarding claim 1, Farchi teaches a computing system comprising: a processor coupled to the network controller; and a memory coupled to the processor, the memory including a set of executable program instructions, which when executed by the processor, cause the computing system to (Farchi, [Par.0005], “Another exemplary embodiment of the disclosed subject matter is a computerized apparatus having a processor and coupled memory, the processor being adapted to perform the steps”).;
identify a first neural network, wherein the first neural network is associated with a first training parameter and first population data that are generated during a process to train the first neural network (Farchi, [par.0016], “One technical solution is to compare a statistical metric of the seen dataset with the statistical metric in the unseen dataset. In some exemplary embodiments, a seen dataset may be obtained. In some exemplary embodiments, the seen dataset may comprise seen instances. Each seen instance may comprise feature values in a feature space. In some exemplary embodiments, the feature space may be an N dimensional space, where N is the number of features used to represent the instance. It is noted that some features may be obtained directly from the raw data, some features may be engineered features that are computed based on the raw data, or the like. In some exemplary embodiments, a first measurement of a statistical metric of the seen dataset may be determined.,”, [Par.0026], “As an example, the encoding decoding function may be determined using an autoencoder. The autoencoder may be an artificial neural network used to learn efficient data codings in an unsupervised manner. In some exemplary embodiments, the aim of an autoencoder may be to learn a representation (encoding) for a dataset, for dimensionality reduction, by training the network to ignore “noisy” signals. In some exemplary embodiments, along with the reduction side, a reconstructing side is learnt, where the autoencoder tries to generate from the reduced encoding a representation as close as possible to its original input, hence its name. In some exemplary embodiments, for each or some of the seen instances...” Examiner’s note, the first measurement of statistical metric is determined by autoencoder (neural network) is trained on the seen dataset, therefore, the autoencoder is considered as the first neural network. Each of an instance of the seen dataset including the feature (parameter). Using the autoencoder to train on the seen dataset is corresponding to the identifying the first neural network, and the seen dataset is considered as the first population data.);
execute a first neural network process to serve input data with the first neural network (Farchi, [Par.0026, Fig.3, Par.0062-0063], “As an example, the encoding decoding function may be determined using an autoencoder. The autoencoder may be an artificial neural network used to learn efficient data codings in an unsupervised manner. In some exemplary embodiments, the aim of an autoencoder may be to learn a representation (encoding) for a dataset, for dimensionality reduction, by training the network to ignore “noisy” signals... In some exemplary embodiments, the values of the delta function in the seen dataset may be viewed as a random variable. Statistical metric of the random variable, such as average value, variance, or the like, may be computed with respect to the seen dataset. Similarly, the same statistical metric may be computed with respect to the unseen dataset, based on the values of the delta function in the unseen dataset. In some exemplary embodiments, the data drift may be identified based on a difference between the different measurements of the statistical metric being above a predetermined threshold.” And [Par.0062-0063], “On Step 340, similarly to Step 320, unseen instances may be encoded and decoded to obtain decoded unseen instances. [0063] On Step 350, similarly to Step 320, a second measurement may be computed. The second measurement may be based on the same statistical metric of Step 320, measuring a statistical difference between unseen instances and their corresponding unseen decoded instances.” Examiner’s note, the autoencoder is trained on the unseen dataset to computer the second measurement, therefore, the unseen dataset is considered as the input data is trained on the first neural network.);
and estimate a first drift of the first neural network based on the first neural network process, the first training parameter and the first population data to determine whether to retrain the first neural network (Farchi, [Par.0005], “Another exemplary embodiment of the disclosed subject matter is a computerized apparatus having a processor and coupled memory, the processor being adapted to perform the steps of: obtaining a seen dataset, wherein the seen dataset comprises seen instances, each of which comprising feature values in a feature space; determining a first measurement of a statistical metric of the seen dataset; obtaining an unseen dataset, wherein the unseen dataset comprises unseen instances, each of which comprising features values in the feature space; determining a second measurement of the statistical metric of the unseen dataset; identifying a data drift in the unseen dataset with respect to the seen dataset based on the first and second measurements of the statistical metric; and performing a responsive action based on the identification of the data drift.” And [Par.0022], “For example, a variance that is above 0.5 points, may be indicative of a data drift. In case that a data drift was identified, the responsive action may comprise re-training the predictive model. In some cases, the predictive model may be re-trained using at least a portion of the unseen data.” Examiner’s note, determining the drift based on the first measurement on the seen dataset (first population data), wherein, the seen dataset include the feature (parameter), if the drift identified then the retraining action is taken.) .
However, Farchi does not clarify that a network controller to communicate with edge nodes

On the other hand, NI teaches a network controller to communicate with edge nodes (Ni, [Col. 3, lines 60-65, and Fig.3, Col.5], “FIG. 1 illustrates a schematic diagram of environment 100 in which embodiments of the present disclosure can be implemented. In the example of FIG. 1, computing device 110 is configured to train machine learning model 112 with training data set 102.” And “FIG. 3 illustrates an example structure of machine learning model 112. In the example of FIG. 3, machine learning model 112 shows an example structure of a deep neural network (DNN). DNN is an artificial neural network with multiple layers between an input layer and an output layer. As shown in FIG. 3, machine learning model 112 may include input layer 310, a plurality of hidden layers 320, 330, etc., and output layer 340… In the example of DNN, machine learning model 112 is a typical forward network, and the to-be-processed data (i.e., model inputs) flows from the input layer to the output layer without looping back. DNN creates a virtual neuron map. During training, the connections between neurons are assigned random values or “weights,” referred to as parameter values.” Examiner’s note, the device generating the machine learning model, wherein, the machine learning model is the neural network that including the plurality of neurons and the weight (edge) connection between the neurons.
Farchi and Ni are analogous in arts because they have the same field of endeavor of training data in neural network.
Accordingly, it would have been obvious to one of the ordinary skills in the art before the effective filing date of the claimed invention to have the modified the a processor coupled to the network controller, and a memory coupled to the processor, the memory including a set of executable program instructions, which when executed by the processor, cause the computing system to: identify a first neural network, wherein the first neural network is associated with a first training parameter and first population data that are generated during a process to train the first neural network, as taught by Farchi, to include the network controller to communicate with edge nodes, as taught by Ni. The modification would have been obvious because one of the ordinary skills in art would be motivated to determine the type of the drift, (NI, [Col. 9, lines 8-30], “At block 230, computing device 120 determines a type of a conceptual drift of machine learning model 112 based on the performance metric and the homogeneity degree. In machine learning, a conceptual drift means that statistical characteristics of a target variable that the model is trying to predict change in an unforeseen way over time. This will lead to problems, because the prediction by the model becomes increasingly inaccurate over time. In the embodiments of the present disclosure, the conceptual drifts of machine learning are classified into different types. In some embodiments, the types of conceptual drifts include drifts related to the output part of machine learning model 112, denoted as P(Y).”).
Regarding claim 2, Farchi teaches the computing system of claim 1, wherein the executable program instructions, when executed, cause the computing system to: determine a plurality of distances for a plurality of layers of the first neural network based on the first neural network process, the first training parameter and the first population data (Farchi, [Par.0058-0060], “On Step 310, seen instances may be encoded and decoded to obtain decoded seen instances. The encoding and decoding may be performed using an encoding function encoding the instance from a feature space to an encoded feature space, such as a feature space having a reduced number of dimensions. The encoding and decoding may be performed using a decoding function to decode the encoded instance, whereby obtaining a decoded instance represented back in the feature space. In some exemplary embodiments, the encoding and decoding functions may be determined using an autoencoder, as is illustrated in FIG. 4...[0060] In some exemplary embodiments, Encoding Function 422 may be an ANN. Additionally or alternatively, Decoding Function 428 may be an ANN. In some exemplary embodiments, the ANNs may be trained so as to provide a minimization of a distance between Input Vector 410 and Output Vector 430. After the training is performed, Encoded Vector 425 may be considered as providing an encoding of the features of the instance, using a reduced number of instances, but while retaining an embodiment of the important qualities thereof. In some exemplary embodiments, Encoding-Decoding Function 420 may be configured to filter non-important data and retain important data, using the features of Encoded Vector 425” Examiner’s note, calculate the distance between the input layer and the output layer of the autoencoder (first neural network);
and estimate the first drift based on the plurality of distances (Farchi, [Par.0042], “] On Step 160, it may be determined whether a data drift occurred. In some exemplary embodiments, a data drift may be identified based on a substantial change in the second measurement (of Step 150) from the first measurement (of Step 120). In some exemplary embodiments, a data drift may be identified based on a difference between the first measurement and the second measurement that is above a predetermined threshold, such as an absolute threshold, a relative threshold, or the like.” Examiner’s note, the drift is determined based on the difference of the first measure and the second measurement, wherein, the first measurement or the second measurement is determined based on the distance between the input layer and output layer of each respective dataset (the seen dataset or the unseen dataset).
Regarding claim 3, Farchi teaches the computing system of claim 1, wherein the executable program instructions, when executed, cause the computing system to: identify a second neural network, wherein the second neural network is associated with a second training parameter and second population data that are generated during a process to train the second neural network (Farchi, [Par.0020-0022]], “In some exemplary embodiments, the responsive action may comprise obtaining a new seen dataset. Referring again to the above mentioned signs example, in case that new signs were introduced, it may be desired to retrain the predictive model with the new signs as the seen data. The new seen dataset may comprise a portion of the seen dataset, a portion of the unseen data set, a combination thereof, or the like… The statistical measurement may be the resulting variance of the PCA. A variance that is above a predetermined threshold may be indicative of a data drift. For example, a variance that is above 0.5 points, may be indicative of a data drift. In case that a data drift was identified, the responsive action may comprise re-training the predictive model. In some cases, the predictive model may be re-trained using at least a portion of the unseen data.” Examiner’s note, the predictive model is re-trained on the new training data, therefore, the feature (new sign) is considered as the second training parameter, and the new training data is considered as the second population data, determining the retraining of the predictive model is considered as the identifying the second neural network.) ;
However, Farchi does not teach the second neural network but it does not teach execute a second neural network process to serve the input data with the second neural network, estimate a second drift of the second neural network based on the second neural network process, the second training parameter and the second population data, select one of the first neural network and the second neural network for deployment based on the first drift and the second drift,
On the other hand, Ni teaches execute a second neural network process to serve the input data with the second neural network (Ni, [Col.10, lines 1-23], “In some embodiments, if computing device 120 determines that the homogeneity degree of machine learning model 112 exceeds the homogeneity threshold, that is, there is little difference between the verification feature representation and the training feature representation, this means that the feature extraction of the model input by machine learning model 112 may be accurate. The unsatisfactory performance metric of the model may be due to the output part of machine learning model 112. This situation may occur because the labels used as supervision information in training data set 102 are not accurate enough. For example, all the categories in training data set 102 are not labeled, or the labeling of the classification or regression results deviates from the actual situation. This situation can be classified as a type of concept drifts (referred to as a “second type” of concept drift P(X)). For this type of conceptual drift, machine learning model 112 can retrain the output part of machine learning model 112 without training other parts of machine learning model 112, such as the feature extraction part. Such retraining can be referred to as local retraining, and sometimes referred to as a second local update.” Examiner’s note, The model is retrained on the portion of the data cause the drift.) ;
estimate a second drift of the second neural network based on the second neural network process, the second training parameter and the second population data (Ni, [Claim 1], “executing the updated machine learning model in the machine learning system to process one or more additional data sets; and repeating the determining of the performance metric of the machine learning model at runtime, the determining of the homogeneity degree, the determining of the type of conceptual drift, the performing of the update and the executing of the updated machine learning model in each of a plurality of iterations to provide automated adaptive updating of the machine learning model in the machine learning system, with different ones of the partial update and the global update being performed in different ones of the iterations based on the type of conceptual drift determined in each such iteration; wherein determining a type of conceptual drift of the machine learning model based on the performance metric and the homogeneity degree comprises distinguishing between multiple distinct types of conceptual drift including at least conceptual drift in an output of the machine learning model, conceptual drift in an input of the machine learning model, and conceptual drift in a relationship between the input and the output of the machine learning model;” ;
and select one of the first neural network and the second neural network for deployment based on the first drift and the second drift (Ni, [Claim 1], “and wherein performing an update of the machine learning model based on the type of the conceptual drift comprises updating different portions of the machine learning model utilizing respective distinct update processes responsive to determining respective different ones of the multiple distinct types of conceptual drift, a first one of the update processes applied responsive to determining a first type of conceptual drift comprising locally updating an output layer of a neural network of the machine learning model through calibration performed in a model calibrator deployed after the output layer of the neural network and configured to automatically adjust outputs of the output layer responsive to detection of the first type of conceptual drift, a second one of the update processes applied responsive to determining a second type of conceptual drift comprising locally updating the output layer of the neural network through parameter adjustment in which one or more parameters of the output layer but no hidden layers of the neural network are automatically updated responsive to detection of the second type of conceptual drift, and a third one of the update processes applied responsive to determining a third type of conceptual drift comprising globally retraining the neural network including the output layer and at least one or more hidden layers of the neural network, the global retraining automatically updating one or more parameters of the output layer and the one or more hidden layers of the neural network responsive to detection of the third type of conceptual drift.” Examiner’s note, the machine is deployed based on the determination of the plurality of type of the concept drifts (first type of the concept drift and second type of the concept drift.).
Farchi and Ni are analogous in arts because they have the same field of endeavor of training data in neural network.
Accordingly, it would have been obvious to one of the ordinary skills in the art before the effective filing date of the claimed invention to have the modified the identify a second neural network, wherein the second neural network is associated with a second training parameter and second population data that are generated during a process to train the second neural network, as taught by Farchi, to include execute a second neural network process to serve the input data with the second neural network, estimate a second drift of the second neural network based on the second neural network process, the second training parameter and the second population data, select one of the first neural network and the second neural network for deployment based on the first drift and the second drift, as taught by Ni. The modification would have been obvious because one of the ordinary skills in art would be motivated to determine the type of the drift, (NI, [Col. 9, lines 8-30], “At block 230, computing device 120 determines a type of a conceptual drift of machine learning model 112 based on the performance metric and the homogeneity degree. In machine learning, a conceptual drift means that statistical characteristics of a target variable that the model is trying to predict change in an unforeseen way over time. This will lead to problems, because the prediction by the model becomes increasingly inaccurate over time. In the embodiments of the present disclosure, the conceptual drifts of machine learning are classified into different types. In some embodiments, the types of conceptual drifts include drifts related to the output part of machine learning model 112, denoted as P(Y).”).
The claim 14 is rejected for the same reason as the claim 1, since these claims recite the same limitations.
The claim 15 is rejected for the same reason as the claim 2, since these claims recite the same limitations.
The claim 16 is rejected for the same reason as the claim 3, since these claims recite the same limitations.
The claim 20 is rejected for the same reason as the claim 1, since these claims recite the same limitations.
The claim 21 is rejected for the same reason as the claim 2, since these claims recite the same limitations.
The claim 22 is rejected for the same reason as the claim 3, since these claims recite the same limitations.
Claims 4, 17, 23 are rejected under 35 U.S.C. 103 as being unpatentable over Farchi et al. (Pub. No. US 20200342310-hereinafter, Farchi) further in view of Ni et al. (Patent. No. US 12406181 -hereinafter, Ni) and further in view of Hearty et al. (Pub. No. US 20200280578 -hereinafter, Hearty).
Regarding claim 4, Farchi teaches the computing system of the claim 1, wherein the executable program instructions, when executed, cause the computing system to, but it does not teach determine whether to retrain the first neural network based on a comparison of the first drift to a threshold,
On the other hand, Hearty teaches determine whether to retrain the first neural network based on a comparison of the first drift to a threshold (Hearty, [Par.0035], “Example machine learning engines include decision tree learning, association rule learning, artificial neural networks, classifiers, inductive logic programming, support vector machines, clustering, Bayesian networks, reinforcement learning, representation learning, similarity and metric learning, sparse dictionary learning, and genetic algorithms. Using one or more of the approaches described above, a computer program can ingest, parse, and understand data and progressively refine algorithms for data analytics. In some examples, the machine learning performed by the fraud prevention server 135 in executing the OAO service is an ensemble machine learning model named XGBoost (eXtreme Gradient Boosting trees), a gradient boosting algorithm implemented for speed and performance.” And [Par.0093], “These combinations may be problem specific or automatically defined (e.g., every combination of features may be assessed). The OAO drift monitoring component 1208 may also compare drift against a defined threshold and trigger an alert and retraining activity when the comparison determines that the drift exceeds a certain threshold.” .) .
Farchi and Hearty are analogous in arts because they have the same field of endeavor of training data in neural network.
Accordingly, it would have been obvious to one of the ordinary skills in the art before the effective filing date of the claimed invention to have the modified the estimating the drift, as taught by Farchi, to include determine whether to retrain the first neural network based on a comparison of the first drift to a threshold, as taught by Hearty. The modification would have been obvious because one of the ordinary skills in art would be motivated to improve the autofill data, (Hearty, [Par.0109], “However, as described above with respect to feature drifting, the behavioral aspects associated with nefarious actors will change as online application environment changes over time. For example, browser changes, website changes, improved autofill, or other suitable online application environment changes will cause “feature drift” because the behavioral aspects originally associated with nefarious actors will no longer accurately distinguish nefarious actors from good actors.”).
The claim 17 is rejected for the same reason as the claim 4, since these claims recite the same limitations.
The claim 23 is rejected for the same reason as the claim 4, since these claims recite the same limitations.
Claims 5, 18, 24 are rejected under 35 U.S.C. 103 as being unpatentable over Farchi et al. (Pub. No. US 20200342310-hereinafter, Farchi) in view of Ni et al. (Patent. No. US 12406181 -hereinafter, Ni) and further in view of OH et al. (Pub. No. US 20210034972-hereinafter, OH).
Regarding claim 5, Farchi teaches computing system of claim 1, wherein: the first neural network, it does not teach the first neural network includes a plurality of batch normalization layers; the first training parameter is to include a scale parameter and a shift parameter associated with the plurality of batch normalization layers, wherein the scale parameter and the shift parameter are generated during the process to train the first neural network; and the first population data is to include a moving mean and moving standard deviation generated during the process to train the first neural network.
On the other hand, OH teaches the first neural network includes a plurality of batch normalization layers (OH, [Par.0018], “A neural network learning apparatus according to an embodiment of the present disclosure includes an input interface for inputting training data for learning of an artificial neural network model, and a learning processor for training the artificial neural network model by using the training data, and the artificial neural network model may include a batch normalization layer between a first neural network layer and a second neural network layer,;
the first training parameter is to include a scale parameter and a shift parameter associated with the plurality of batch normalization layers (Oh, [Par.0187], “In addition, the assumption that the features have already been uncorrelated may limit what the neural network model may represent. In order to supplement these problems, the batch normalization method obtains by transforming the normalized values by using a gamma, which is a scale parameter and a beta, which is a shift parameter, respectively, and learning these variables together in a back-propagation process.” ,
wherein the scale parameter and the shift parameter are generated during the process to train the first neural network (OH, [Par.00187-0190], “[00187] the first training parameter is to include a scale parameter and a shift parameter associated with the plurality of batch normalization layers.. [0190]The Equations 1 and 2 are equations for calculating the mean and the variance of the mini-batch, the Equation 3 is an equation for calculating the normalized value of x.sub.i by using the mean and the variance calculated from the mini-batch, and the Equation 4 is an equation for performing normalization transform by multiplying the normalized value of x.sub.i by a gamma, which is a scale parameter, and adding a beta, which is a shift parameter.;
and the first population data is to include a moving mean and moving standard deviation generated during the process to train the first neural network (OH, [Par.0196-0197, 0260], “The batch normalization transform according to an embodiment of the present disclosure may transform by setting the gradient for the standard deviation and the gradient for the mean of the loss function to zero, and transform by applying the normalized statistic value obtained from the initial neural network or the previous neural network to the gradient of the loss function.[0197] The normalized statistic value may include the number of data per each set, a mean, and a standard deviation operated in each mini-batch set or the data set.” Examiner’s note, the mean and the standard deviation is generated for each of the subset of data.).
Farchi and OH are analogous in arts because they have the same field of endeavor of training data in neural network.
Accordingly, it would have been obvious to one of the ordinary skills in the art before the effective filing date of the claimed invention to have the neural network, as taught by Farchi, to include the first neural network includes a plurality of batch normalization layers; the first training parameter is to include a scale parameter and a shift parameter associated with the plurality of batch normalization layers, wherein the scale parameter and the shift parameter are generated during the process to train the first neural network; and the first population data is to include a moving mean and moving standard deviation generated during the process to train the first neural network., as taught by OH. The modification would have been obvious because one of the ordinary skills in art would be motivated to improve the accuracy of the performance of the data analysis and the machine learning model (OH, [Par.0171], “Accordingly, the processor 160, together with learning processor 140, may improve the accuracy of future performance of the data analysis and the machine learning algorithm and technique based on the updated information.”).
The claim 18 is rejected for the same reason as the claim 5, since these claims recite the same limitations.
The claim 24 is rejected for the same reason as the claim 5, since these claims recite the same limitations.
Claims 6, 19, 25 are rejected under 35 U.S.C. 103 as being unpatentable over Farchi et al. (Pub. No. US 20200342310-hereinafter, Farchi) in view of Ni et al. (Patent. No. US 12406181 -hereinafter, Ni) and further in view of Zamft et al. (Pub. No. US 20220301658-hereinafter, Zamft).
Regarding claim 6, Farchi teaches the computing system of claim 1, wherein the executable program instructions, when executed, cause the computing system to: estimate the first drift, but it does not teach estimate the first drift of the first neural network based on an input vector into a batch normalization layer of the first neural network and an output vector of the batch normalization layer,
On the other hand, Zamft teaches estimate the first drift of the first neural network based on an input vector into a batch normalization layer of the first neural network and an output vector of the batch normalization layer (Zamft ,[Par.0042-0049], “[0042]FIG. 2A shows an exemplary deep neural network 200 (in this instance the exemplary deep neural network is a feedforward neural network; however it should be understood that other types of neural networks could be implemented without departing from the spirit and scope of the present disclosure). The deep neural network 200 comprises an input layer 205, multiple hidden layers 210a-210(n), and an output layer 215. The input layer 205 is used to feed the input data or features to the deep neural network 200. In some instances, the input data or features is gene expression profiles or a transformed version thereof, e.g., log-transformed gene expression profiles. The deep neural network 200 applies a series of functions to the input data using the multiple hidden layers 210a-210(n). The number of hidden layers defines the depth of the deep neural network 200. By having multiple hidden layers 210a-210(n), the deep neural network 200 can compute complex functions by cascading simpler functions. In some instances, the depth of the deep neural network 200 is two or more hidden layers. In certain instances, the depth of the deep neural network 200 is two hidden layers, as depicted in FIG. 2A.” [Par.0048], “Each iteration of the training process for the deep neural network 200 can involve inputting data sets from the training and/or tuning set and learning a set of model parameters (configured with a defined set of hyperparameters) so that the value of the loss or error function (e.g., modified cross entropy loss) using the set of model parameters is smaller than the value of the loss or error function using a different set of model parameters in a previous iteration. The loss or error function can be constructed to measure the difference between the labels/ground truth and the inferred data such as an inferred phenotype for a set of gene expression profiles. Initial values (e.g., random values or values selected based on prior knowledge) are assigned to the set of model parameters to be used as a starting point for training the deep neural network 200, and each iteration of the training process for deep neural network 200 can further involve feeding the loss backwards through the network (i.e., backpropagation) to fine tune the set of model parameters. This process of iteratively passing batches of data through the deep neural network 200, estimating error based on the subset of the training dataset, and updating the weights, so that the error is decreased, is known as Gradient Descent.” Examiner’s note, the loss of the neural network is estimated based on the training of batch normalization from input layer to output layer of the neural network.).
Farchi and OH are analogous in arts because they have the same field of endeavor of training data in neural network.
Accordingly, it would have been obvious to one of the ordinary skills in the art before the effective filing date of the claimed invention to have the modified of estimate the first drift, as taught by Farchi, to include the estimate the first drift of the first neural network based on an input vector into a batch normalization layer of the first neural network and an output vector of the batch normalization layer, as taught by Zamft. The modification would have been obvious because one of the ordinary skills in art would be motivated to decrease the error (Zamft, [Par.0048], “[Par.0048], “Each iteration of the training process for the deep neural network 200 can involve inputting data sets from the training and/or tuning set and learning a set of model parameters (configured with a defined set of hyperparameters) so that the value of the loss or error function (e.g., modified cross entropy loss) using the set of model parameters is smaller than the value of the loss or error function using a different set of model parameters in a previous iteration. The loss or error function can be constructed to measure the difference between the labels/ground truth and the inferred data such as an inferred phenotype for a set of gene expression profiles. Initial values (e.g., random values or values selected based on prior knowledge) are assigned to the set of model parameters to be used as a starting point for training the deep neural network 200, and each iteration of the training process for deep neural network 200 can further involve feeding the loss backwards through the network (i.e., backpropagation) to fine tune the set of model parameters. This process of iteratively passing batches of data through the deep neural network 200, estimating error based on the subset of the training dataset, and updating the weights, so that the error is decreased, is known as Gradient Descent.”).
The claim 19 is rejected for the same reason as the claim 6, since these claims recite the same limitations.
The claim 25 is rejected for the same reason as the claim 6, since these claims recite the same limitations.
Claims 7, 8, 13 are rejected under 35 U.S.C. 103 as being unpatentable over Farchi et al. (Pub. No. US 20200342310-hereinafter, Farchi) further in view of Romero Aragon et al. (Pub. No. US 20200293280 -hereinafter, Romero Aragon).
Regarding claim 7, Farchi teaches identify a first neural network, wherein the first neural network is associated with a first training parameter and first population data that are generated during a process to train the first neural network (Farchi, [par.0016], “One technical solution is to compare a statistical metric of the seen dataset with the statistical metric in the unseen dataset. In some exemplary embodiments, a seen dataset may be obtained. In some exemplary embodiments, the seen dataset may comprise seen instances. Each seen instance may comprise feature values in a feature space. In some exemplary embodiments, the feature space may be an N dimensional space, where N is the number of features used to represent the instance. It is noted that some features may be obtained directly from the raw data, some features may be engineered features that are computed based on the raw data, or the like. In some exemplary embodiments, a first measurement of a statistical metric of the seen dataset may be determined.,”, [Par.0026], “As an example, the encoding decoding function may be determined using an autoencoder. The autoencoder may be an artificial neural network used to learn efficient data codings in an unsupervised manner. In some exemplary embodiments, the aim of an autoencoder may be to learn a representation (encoding) for a dataset, for dimensionality reduction, by training the network to ignore “noisy” signals. In some exemplary embodiments, along with the reduction side, a reconstructing side is learnt, where the autoencoder tries to generate from the reduced encoding a representation as close as possible to its original input, hence its name. In some exemplary embodiments, for each or some of the seen instances...” Examiner’s note, the first measurement of statistical metric is determined by autoencoder (neural network) is trained on the seen dataset, therefore, the autoencoder is considered as the first neural network. Each of an instance of the seen dataset including the feature (parameter). Using the autoencoder to train on the seen dataset is corresponding to the identifying the first neural network, and the seen dataset is considered as the first population data.);
execute a first neural network process to serve input data with the first neural network (Farchi, [Par.0026, Fig.3, Par.0062-0063], “As an example, the encoding decoding function may be determined using an autoencoder. The autoencoder may be an artificial neural network used to learn efficient data codings in an unsupervised manner. In some exemplary embodiments, the aim of an autoencoder may be to learn a representation (encoding) for a dataset, for dimensionality reduction, by training the network to ignore “noisy” signals. In some exemplary embodiments, along with the reduction side, a reconstructing side is learnt, where the autoencoder tries to generate from the reduced encoding a representation as close as possible to its original input, hence its name. In some exemplary embodiments, for each or some of the seen instances, s[i], a decode-seen instance ds[i] may be computed. In some exemplary embodiments, ds[i] may be computed as follows: ds[i]=dec(enc(s[i])). In some exemplary embodiments, for each instance in the seen dataset, a value of a delta function (Δ) may be computed. The delta function may be defined as follows: Δ(i)=i−dec(enc(i)), wherein i is the each instance, wherein enc(i) is an encoding function from the feature space to the encoded feature space, wherein dec(x) is a decoding function from the encoded feature space to the feature space, wherein dec(enc(i)) is the encoding-decoding function. In some exemplary embodiments, the values of the delta function in the seen dataset may be viewed as a random variable. Statistical metric of the random variable, such as average value, variance, or the like, may be computed with respect to the seen dataset. Similarly, the same statistical metric may be computed with respect to the unseen dataset, based on the values of the delta function in the unseen dataset. In some exemplary embodiments, the data drift may be identified based on a difference between the different measurements of the statistical metric being above a predetermined threshold.” And [Par.0062-0063], “On Step 340, similarly to Step 320, unseen instances may be encoded and decoded to obtain decoded unseen instances. [0063] On Step 350, similarly to Step 320, a second measurement may be computed. The second measurement may be based on the same statistical metric of Step 320, measuring a statistical difference between unseen instances and their corresponding unseen decoded instances.” Examiner’s note, the autoencoder is trained on the unseen dataset to computer the second measurement, therefore, the unseen dataset is considered as the input data is trained on the first neural network.);
and estimate a first drift of the first neural network based on the first neural network process, the first training parameter and the first population data to determine whether to retrain the first neural network (Farchi, [Par.0005], “Another exemplary embodiment of the disclosed subject matter is a computerized apparatus having a processor and coupled memory, the processor being adapted to perform the steps of: obtaining a seen dataset, wherein the seen dataset comprises seen instances, each of which comprising feature values in a feature space; determining a first measurement of a statistical metric of the seen dataset; obtaining an unseen dataset, wherein the unseen dataset comprises unseen instances, each of which comprising features values in the feature space; determining a second measurement of the statistical metric of the unseen dataset; identifying a data drift in the unseen dataset with respect to the seen dataset based on the first and second measurements of the statistical metric; and performing a responsive action based on the identification of the data drift.” And [Par.0022], “For example, a variance that is above 0.5 points, may be indicative of a data drift. In case that a data drift was identified, the responsive action may comprise re-training the predictive model. In some cases, the predictive model may be re-trained using at least a portion of the unseen data.” Examiner’s note, determining the drift based on the first measurement on the seen dataset (first population data), wherein, the seen dataset include the feature (parameter), if the drift identified then the retraining action is taken.) .
However Farchi does not teach a semiconductor apparatus comprising: one or more substrates; and logic coupled to the one or more substrates, wherein the logic is implemented in one or more of configurable or fixed-functionality hardware, the logic to:
On the other hand, Romero Aragon teaches a semiconductor apparatus comprising: one or more substrates; and logic coupled to the one or more substrates, wherein the logic is implemented in one or more of configurable or fixed-functionality hardware, the logic to (Romero Aragon, [Par.0057], “IG. 6 shows a semiconductor package apparatus 160. The apparatus 160 may implement one or more aspects of the method 44 (FIG. 2) and/or the method 90 (FIG. 4) and may be readily substituted for the digital processing apparatus 20 (FIG. 1), already discussed. The illustrated apparatus 160 includes one or more substrates 164 (e.g., silicon, sapphire, gallium arsenide) and logic 162 (e.g., transistor array and other integrated circuit/IC components) coupled to the substrate(s) 164. The logic 162 may be implemented at least partly in configurable logic or fixed-functionality logic hardware.”
Farchi and Romero Aragon are analogous in arts because they have the same field of endeavor of generating the computer system.
Accordingly, it would have been obvious to one of the ordinary skills in the art before the effective filing date of the claimed invention to have the modified of identify a first neural network, wherein the first neural network is associated with a first training parameter and first population data that are generated during a process to train the first neural network; execute a first neural network process to serve input data with the first neural network; and estimate a first drift of the first neural network based on the first neural network process, the first training parameter and the first population data to determine whether to retrain the first neural network., as taught by Farchi, to include the one or more substrates; and logic coupled to the one or more substrates, wherein the logic is implemented in one or more of configurable or fixed-functionality hardware, the logic to:, as taught by Romero Aragon. The modification would have been obvious because one of the ordinary skills in art would be motivated to improve the accuracy, [Romero Aragon, [Par.0022], “In order to further improve accuracy, the Mitchell approximation may be modified with bit error correction solutions. The method 44 may generally provide an enhanced method to reduce the correction solution error by computing the derivative of the logarithmic function (and antilogarithmic function, as discussed below) and obtaining a better approximation compared to conventional solutions.”).
The claim 8 is rejected for the same reason as the claim 2, since these claims recite the same limitations.
The claim 13, Farchi in view of Romero Aragon teaches the apparatus of claim 7, wherein the logic coupled to the one or more substrates includes transistor channel regions that are positioned within the one or more substrates (Romero Aragon, [Par.0057], “FIG. 6 shows a semiconductor package apparatus 160. The apparatus 160 may implement one or more aspects of the method 44 (FIG. 2) and/or the method 90 (FIG. 4) and may be readily substituted for the digital processing apparatus 20 (FIG. 1), already discussed. The illustrated apparatus 160 includes one or more substrates 164 (e.g., silicon, sapphire, gallium arsenide) and logic 162 (e.g., transistor array and other integrated circuit/IC components) coupled to the substrate(s) 164. The logic 162 may be implemented at least partly in configurable logic or fixed-functionality logic hardware. In one example, the logic 162 includes transistor channel regions that are positioned (e.g., embedded) within the substrate(s). Thus, the interface between the logic 162 and the substrate(s) 164 may not be an abrupt junction. The logic 162 may also be considered to include an epitaxial layer that is grown on an initial wafer of the substrate(s).”).
Farchi and Romero Aragon are analogous in arts because they have the same field of endeavor of generating the computer system.
Accordingly, it would have been obvious to one of the ordinary skills in the art before the effective filing date of the claimed invention to have the modified the apparatus of the claim 7, as taught by Farchi, to include the logic coupled to the one or more substrates includes transistor channel regions that are positioned within the one or more substrates, as taught by Romero Aragon. The modification would have been obvious because one of the ordinary skills in art would be motivated to improve the accuracy, [Romero Aragon, [Par.0022], “In order to further improve accuracy, the Mitchell approximation may be modified with bit error correction solutions. The method 44 may generally provide an enhanced method to reduce the correction solution error by computing the derivative of the logarithmic function (and antilogarithmic function, as discussed below) and obtaining a better approximation compared to conventional solutions.”).
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Farchi et al. (Pub. No. US 20200342310-hereinafter, Farchi) further in view of Romero Aragon et al. (Pub. No. US 20200293280 -hereinafter, Romero Aragon) and further in view of Ni et al. (Patent. No. US 12406181 -hereinafter, Ni).
Regarding claim 9, Farchi in view of Romero Aragon teaches the apparatus of claim 7, wherein the logic coupled to the one or more substrates is to:: identify a second neural network, wherein the second neural network is associated with a second training parameter and second population data that are generated during a process to train the second neural network (Farchi, [Par.0020-0022]], “In some exemplary embodiments, the responsive action may comprise obtaining a new seen dataset. Referring again to the above mentioned signs example, in case that new signs were introduced, it may be desired to retrain the predictive model with the new signs as the seen data. The new seen dataset may comprise a portion of the seen dataset, a portion of the unseen data set, a combination thereof, or the like… The statistical measurement may be the resulting variance of the PCA. A variance that is above a predetermined threshold may be indicative of a data drift. For example, a variance that is above 0.5 points, may be indicative of a data drift. In case that a data drift was identified, the responsive action may comprise re-training the predictive model. In some cases, the predictive model may be re-trained using at least a portion of the unseen data.” Examiner’s note, the predictive model is re-trained on the new training data, therefore, the feature (new sign) is considered as the second training parameter, and the new training data is considered as the second population data, determining the retraining of the predictive model is considered as the identifying the second neural network.) ;
However, Farchi does not teach the second neural network but it does not teach execute a second neural network process to serve the input data with the second neural network, estimate a second drift of the second neural network based on the second neural network process, the second training parameter and the second population data, select one of the first neural network and the second neural network for deployment based on the first drift and the second drift,
On the other hand, Ni teaches execute a second neural network process to serve the input data with the second neural network (Ni, [Col.10, lines 1-23], “In some embodiments, if computing device 120 determines that the homogeneity degree of machine learning model 112 exceeds the homogeneity threshold, that is, there is little difference between the verification feature representation and the training feature representation, this means that the feature extraction of the model input by machine learning model 112 may be accurate. The unsatisfactory performance metric of the model may be due to the output part of machine learning model 112. This situation may occur because the labels used as supervision information in training data set 102 are not accurate enough. For example, all the categories in training data set 102 are not labeled, or the labeling of the classification or regression results deviates from the actual situation. This situation can be classified as a type of concept drifts (referred to as a “second type” of concept drift P(X)). For this type of conceptual drift, machine learning model 112 can retrain the output part of machine learning model 112 without training other parts of machine learning model 112, such as the feature extraction part. Such retraining can be referred to as local retraining, and sometimes referred to as a second local update.” Examiner’s note, The model is retrained on the portion of the data cause the drift.) ;
estimate a second drift of the second neural network based on the second neural network process, the second training parameter and the second population data (Ni, [Claim 1], “executing the updated machine learning model in the machine learning system to process one or more additional data sets; and repeating the determining of the performance metric of the machine learning model at runtime, the determining of the homogeneity degree, the determining of the type of conceptual drift, the performing of the update and the executing of the updated machine learning model in each of a plurality of iterations to provide automated adaptive updating of the machine learning model in the machine learning system, with different ones of the partial update and the global update being performed in different ones of the iterations based on the type of conceptual drift determined in each such iteration; wherein determining a type of conceptual drift of the machine learning model based on the performance metric and the homogeneity degree comprises distinguishing between multiple distinct types of conceptual drift including at least conceptual drift in an output of the machine learning model, conceptual drift in an input of the machine learning model, and conceptual drift in a relationship between the input and the output of the machine learning model;” ;
and select one of the first neural network and the second neural network for deployment based on the first drift and the second drift (Ni, [Claim 1], “and wherein performing an update of the machine learning model based on the type of the conceptual drift comprises updating different portions of the machine learning model utilizing respective distinct update processes responsive to determining respective different ones of the multiple distinct types of conceptual drift, a first one of the update processes applied responsive to determining a first type of conceptual drift comprising locally updating an output layer of a neural network of the machine learning model through calibration performed in a model calibrator deployed after the output layer of the neural network and configured to automatically adjust outputs of the output layer responsive to detection of the first type of conceptual drift, a second one of the update processes applied responsive to determining a second type of conceptual drift comprising locally updating the output layer of the neural network through parameter adjustment in which one or more parameters of the output layer but no hidden layers of the neural network are automatically updated responsive to detection of the second type of conceptual drift, and a third one of the update processes applied responsive to determining a third type of conceptual drift comprising globally retraining the neural network including the output layer and at least one or more hidden layers of the neural network, the global retraining automatically updating one or more parameters of the output layer and the one or more hidden layers of the neural network responsive to detection of the third type of conceptual drift.” Examiner’s note, the machine is deployed based on the determination of the plurality of type of the concept drifts (first type of the concept drift and second type of the concept drift.).
Farchi and Ni are analogous in arts because they have the same field of endeavor of training data in neural network.
Accordingly, it would have been obvious to one of the ordinary skills in the art before the effective filing date of the claimed invention to have the modified the identify a second neural network, wherein the second neural network is associated with a second training parameter and second population data that are generated during a process to train the second neural network, as taught by Farchi, to include execute a second neural network process to serve the input data with the second neural network, estimate a second drift of the second neural network based on the second neural network process, the second training parameter and the second population data, select one of the first neural network and the second neural network for deployment based on the first drift and the second drift, as taught by Ni. The modification would have been obvious because one of the ordinary skills in art would be motivated to determine the type of the drift, (NI, [Col. 9, lines 8-30], “At block 230, computing device 120 determines a type of a conceptual drift of machine learning model 112 based on the performance metric and the homogeneity degree. In machine learning, a conceptual drift means that statistical characteristics of a target variable that the model is trying to predict change in an unforeseen way over time. This will lead to problems, because the prediction by the model becomes increasingly inaccurate over time. In the embodiments of the present disclosure, the conceptual drifts of machine learning are classified into different types. In some embodiments, the types of conceptual drifts include drifts related to the output part of machine learning model 112, denoted as P(Y).”).
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Farchi et al. (Pub. No. US 20200342310-hereinafter, Farchi) further in view of in view of Romero Aragon et al. (Pub. No. US 20200293280 -hereinafter, Romero Aragon) and further in view of Hearty et al. (Pub. No. US 20200280578 -hereinafter, Hearty).
Regarding claim 10, Farchi in view of Romero Aragon teaches the apparatus of claim 7, wherein the logic coupled to the one or more substrates is to: but it does not teach determine whether to retrain the first neural network based on a comparison of the first drift to a threshold,
On the other hand, Hearty teaches determine whether to retrain the first neural network based on a comparison of the first drift to a threshold (Hearty, [Par.0035], “Example machine learning engines include decision tree learning, association rule learning, artificial neural networks, classifiers, inductive logic programming, support vector machines, clustering, Bayesian networks, reinforcement learning, representation learning, similarity and metric learning, sparse dictionary learning, and genetic algorithms. Using one or more of the approaches described above, a computer program can ingest, parse, and understand data and progressively refine algorithms for data analytics. In some examples, the machine learning performed by the fraud prevention server 135 in executing the OAO service is an ensemble machine learning model named XGBoost (eXtreme Gradient Boosting trees), a gradient boosting algorithm implemented for speed and performance.” And [Par.0093], “These combinations may be problem specific or automatically defined (e.g., every combination of features may be assessed). The OAO drift monitoring component 1208 may also compare drift against a defined threshold and trigger an alert and retraining activity when the comparison determines that the drift exceeds a certain threshold.” .) .
Farchi and Hearty are analogous in arts because they have the same field of endeavor of training data in neural network.
Accordingly, it would have been obvious to one of the ordinary skills in the art before the effective filing date of the claimed invention to have the modified the estimating the drift, as taught by Farchi, to include determine whether to retrain the first neural network based on a comparison of the first drift to a threshold, as taught by Hearty. The modification would have been obvious because one of the ordinary skills in art would be motivated to improve the autofill data, (Hearty, [Par.0109], “However, as described above with respect to feature drifting, the behavioral aspects associated with nefarious actors will change as online application environment changes over time. For example, browser changes, website changes, improved autofill, or other suitable online application environment changes will cause “feature drift” because the behavioral aspects originally associated with nefarious actors will no longer accurately distinguish nefarious actors from good actors.”).
Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Farchi et al. (Pub. No. US 20200342310-hereinafter, Farchi) in view of Romero Aragon et al. (Pub. No. US 20200293280 -hereinafter, Romero Aragon) and further in view of OH et al. (Pub. No. US 20210034972-hereinafter, OH).
Regarding claim 11, Farch teaches the apparatus of claim 7,, wherein: the first neural network, it does not teach the first neural network includes a plurality of batch normalization layers; the first training parameter is to include a scale parameter and a shift parameter associated with the plurality of batch normalization layers, wherein the scale parameter and the shift parameter are generated during the process to train the first neural network; and the first population data is to include a moving mean and moving standard deviation generated during the process to train the first neural network.
On the other hand, OH teaches the first neural network includes a plurality of batch normalization layers (OH, [Par.0018], “A neural network learning apparatus according to an embodiment of the present disclosure includes an input interface for inputting training data for learning of an artificial neural network model, and a learning processor for training the artificial neural network model by using the training data, and the artificial neural network model may include a batch normalization layer between a first neural network layer and a second neural network layer,;
the first training parameter is to include a scale parameter and a shift parameter associated with the plurality of batch normalization layers (Oh, [Par.0187], “In addition, the assumption that the features have already been uncorrelated may limit what the neural network model may represent. In order to supplement these problems, the batch normalization method obtains by transforming the normalized values by using a gamma, which is a scale parameter and a beta, which is a shift parameter, respectively, and learning these variables together in a back-propagation process.” ,
wherein the scale parameter and the shift parameter are generated during the process to train the first neural network (OH, [Par.00187-0190], “[00187] the first training parameter is to include a scale parameter and a shift parameter associated with the plurality of batch normalization layers.. [0190]The Equations 1 and 2 are equations for calculating the mean and the variance of the mini-batch, the Equation 3 is an equation for calculating the normalized value of x.sub.i by using the mean and the variance calculated from the mini-batch, and the Equation 4 is an equation for performing normalization transform by multiplying the normalized value of x.sub.i by a gamma, which is a scale parameter, and adding a beta, which is a shift parameter.;
and the first population data is to include a moving mean and moving standard deviation generated during the process to train the first neural network (OH, [Par.0196-0197, 0260], “The batch normalization transform according to an embodiment of the present disclosure may transform by setting the gradient for the standard deviation and the gradient for the mean of the loss function to zero, and transform by applying the normalized statistic value obtained from the initial neural network or the previous neural network to the gradient of the loss function.[0197] The normalized statistic value may include the number of data per each set, a mean, and a standard deviation operated in each mini-batch set or the data set.” Examiner’s note, the mean and the standard deviation is generated for each of the subset of data.).
Farchi and OH are analogous in arts because they have the same field of endeavor of training data in neural network.
Accordingly, it would have been obvious to one of the ordinary skills in the art before the effective filing date of the claimed invention to have the neural network, as taught by Farchi, to include the first neural network includes a plurality of batch normalization layers; the first training parameter is to include a scale parameter and a shift parameter associated with the plurality of batch normalization layers, wherein the scale parameter and the shift parameter are generated during the process to train the first neural network; and the first population data is to include a moving mean and moving standard deviation generated during the process to train the first neural network., as taught by OH. The modification would have been obvious because one of the ordinary skills in art would be motivated to improve the accuracy of the performance of the data analysis and the machine learning model (OH, [Par.0171], “Accordingly, the processor 160, together with learning processor 140, may improve the accuracy of future performance of the data analysis and the machine learning algorithm and technique based on the updated information.”).
Claim12 is rejected under 35 U.S.C. 103 as being unpatentable over Farchi et al. (Pub. No. US 20200342310-hereinafter, Farchi) in view of Romero Aragon et al. (Pub. No. US 20200293280 -hereinafter, Romero Aragon) and further in view of Zamft et al. (Pub. No. US 20220301658-hereinafter, Zamft).
Regarding claim 12, Farchi in view of Romero Aragon teaches the apparatus of claim 7, wherein the logic coupled to : estimate the first drift, but it does not teach estimate the first drift of the first neural network based on an input vector into a batch normalization layer of the first neural network and an output vector of the batch normalization layer,
On the other hand, Zamft teaches estimate the first drift of the first neural network based on an input vector into a batch normalization layer of the first neural network and an output vector of the batch normalization layer (Zamft ,[Par.0042-0049], “[0042]FIG. 2A shows an exemplary deep neural network 200 (in this instance the exemplary deep neural network is a feedforward neural network; however it should be understood that other types of neural networks could be implemented without departing from the spirit and scope of the present disclosure). The deep neural network 200 comprises an input layer 205, multiple hidden layers 210a-210(n), and an output layer 215. The input layer 205 is used to feed the input data or features to the deep neural network 200. In some instances, the input data or features is gene expression profiles or a transformed version thereof, e.g., log-transformed gene expression profiles. The deep neural network 200 applies a series of functions to the input data using the multiple hidden layers 210a-210(n). The number of hidden layers defines the depth of the deep neural network 200. By having multiple hidden layers 210a-210(n), the deep neural network 200 can compute complex functions by cascading simpler functions. In some instances, the depth of the deep neural network 200 is two or more hidden layers. In certain instances, the depth of the deep neural network 200 is two hidden layers, as depicted in FIG. 2A.” [Par.0048], “Each iteration of the training process for the deep neural network 200 can involve inputting data sets from the training and/or tuning set and learning a set of model parameters (configured with a defined set of hyperparameters) so that the value of the loss or error function (e.g., modified cross entropy loss) using the set of model parameters is smaller than the value of the loss or error function using a different set of model parameters in a previous iteration. The loss or error function can be constructed to measure the difference between the labels/ground truth and the inferred data such as an inferred phenotype for a set of gene expression profiles. Initial values (e.g., random values or values selected based on prior knowledge) are assigned to the set of model parameters to be used as a starting point for training the deep neural network 200, and each iteration of the training process for deep neural network 200 can further involve feeding the loss backwards through the network (i.e., backpropagation) to fine tune the set of model parameters. This process of iteratively passing batches of data through the deep neural network 200, estimating error based on the subset of the training dataset, and updating the weights, so that the error is decreased, is known as Gradient Descent.” Examiner’s note, the loss of the neural network is estimated based on the training of batch normalization from input layer to output layer of the neural network.).
Farchi and OH are analogous in arts because they have the same field of endeavor of training data in neural network.
Accordingly, it would have been obvious to one of the ordinary skills in the art before the effective filing date of the claimed invention to have the modified of estimate the first drift, as taught by Farchi, to include the estimate the first drift of the first neural network based on an input vector into a batch normalization layer of the first neural network and an output vector of the batch normalization layer, as taught by Zamft. The modification would have been obvious because one of the ordinary skills in art would be motivated to decrease the error (Zamft, [Par.0048], “[Par.0048], “Each iteration of the training process for the deep neural network 200 can involve inputting data sets from the training and/or tuning set and learning a set of model parameters (configured with a defined set of hyperparameters) so that the value of the loss or error function (e.g., modified cross entropy loss) using the set of model parameters is smaller than the value of the loss or error function using a different set of model parameters in a previous iteration. The loss or error function can be constructed to measure the difference between the labels/ground truth and the inferred data such as an inferred phenotype for a set of gene expression profiles. Initial values (e.g., random values or values selected based on prior knowledge) are assigned to the set of model parameters to be used as a starting point for training the deep neural network 200, and each iteration of the training process for deep neural network 200 can further involve feeding the loss backwards through the network (i.e., backpropagation) to fine tune the set of model parameters. This process of iteratively passing batches of data through the deep neural network 200, estimating error based on the subset of the training dataset, and updating the weights, so that the error is decreased, is known as Gradient Descent.”).

Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to EM N TRIEU whose telephone number is [phone]. The examiner can normally be reached on Mon-Fri from 9:00-5:00.
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, Omar Fernandez Rivas can be reached on [phone]. 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.

/E.T./Examiner, Art Unit 2128

/BRIAN M SMITH/Primary Examiner, Art Unit 2122

Prosecution Timeline

• Dec 22, 2023: Application Filed
• Aug 03, 2026: Non-Final Rejection mailed — §101, §103 (current)