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 .
This action is responsive to the correspondence on 09/18/2025. Claims 1-3, 5, 9-14, 18-19 and 21-25 are amended. Claims 4, 6-8, 15-17, and 20 have been cancelled. Claims 1-3, 5, 9-14, 18, 19 and 21-25 are now pending. Claims 1, 12 and 20 are independent claims.
Response to Arguments
Applicant's arguments filed 09/18/2025 have been fully considered but they are not persuasive.
With respect to 35 U.S.C. 101
Applicant argues the claims do not recite judicial exceptions. Applicant seems to suggest that because the claims do not describe mathematical relationships, calculations, formulas or equations, it does not recite a judicial exception.
Examiner disagrees. Examiner notes that such an analysis is relevant as to whether or not a claim recites mathematical concept. Mathematical concept is only one of the three enumerated types of judicial exceptions in the MPEP. Further, the rejection set forth does not argue that the claim recites mathematical concept, rather that the claim recites certain identified mental processes. An argument that the claim does not recite mathematical concept is not responsive to prior rejection which explicitly identifies that the claim recites activity which can be performed in the mind.
With respect to 35 U.S.C. 103
With respect to independent claim 1 and 12
Applicant argues the cited art alone or in combination does not describe the amended features, noting that at most Xiao describes a disk failure prediction model and not systems used to collect data for the disk failure prediction model.
Examiner notes that collecting data is not claimed. Nevertheless, the rejections have been updated to address the amendments.
With respect to the dependent claims
Applicant argues that none of the cited art overcome the deficiencies in the independent claim.
Examiner disagrees for the reasons stated above.
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.
Claim 1-3, 5, 9-14, and 18-19, 21-25 are rejected under 35 U.S.C. 101 for the following reasons
Regarding Claim 1
Under step 1, the claim is directed to a system for generating a machine learning pipeline for determining maintenance action to be performed…, which is directed to a machine, one of the statutory categories. The claim recites limitations which are considered abstract ideas.
Under Step 2A Prong 1,the following limitations correspond to an evaluation performed in the human mind. “to monitor health and operations of the hardware systems in the data center…to monitor the security of a network of the data center…to detect fraud attempts against the data center…to manage user accounts associated with the data center…”generate, from the plurality of ML pipeline primitives in the ML pipeline primitive database, a plurality of ML pipelines each associated with a respective ML pipeline configuration”, “select a sub-set of ML pipelines from the plurality of ML pipelines, the selecting being based on training each ML pipeline of the plurality of ML pipelines for a first amount of time, a number of ML pipelines from the sub-set of ML pipelines being less than a number of ML pipelines from the plurality of ML pipelines”, “d) evolve the sub-set of ML pipelines to generate evolved ML pipelines, wherein the evolving comprises cloning at least one of the sub-set of the ML pipelines to generate a corresponding at least one evolved ML pipeline, wherein the evolving further comprises modifying parameters of each non-cloned ML pipeline of the sub-set of ML pipelines using a mutation operation or a crossover operation to generate additional evolved ML pipelines;”, “select a sub-set of evolved ML pipelines from the evolved ML pipelines… for a second amount of amount of time being larger than the first amount of time a number of ML pipelines from the sub-set of evolved ML pipelines being less than a number of ML pipelines from the evolved ML pipelines”, “and (f) iterate (d) to (e) until determination is made that iterating (d) to (e) is to be stopped, wherein at each iteration of (d) to (e) a number of ML pipelines from the sub-set of evolved ML pipelines is less than a number of ML pipelines selected in a previous iteration of (d) to (e) while the second volume of data is more than a volume of data in the previous iteration of (d) to (e).”, “wherein selecting the sub-set of evolved ML pipelines from the evolved ML pipelines comprises scoring each one of the ML pipelines of the evolved ML pipelines and sorting the ML pipelines of the evolved ML pipelines, wherein the scoring comprises scoring each ML pipeline based on an accuracy of the respective ML pipeline and a complexity of the respective ML pipeline, and wherein the scoring prioritizes ML pipelines having a higher accuracy and lower complexity”, determine, based on output from the ML pipeline, predictive maintenance actions to be performed on the hardware systems in the data center”
In the context of the claims generating, accessing, evolving and selecting are all processes which can be performed in the mind. For example, a set of pipelines can be initialized or generated by a mere determination of parameters associated with the pipeline. Further they can be evaluated according to certain data properties and can be modified or evolved by making a determination in the human mind for their new configuration, this determination can also be refined via additional selection or determination. Further this process can be repeated and still be considered an activity performed in the mind.
Step 2A Prong Two Analysis: The judicial exception in not integrated into a practical application. In particular, the claim recites additional element(s) that are mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. (“a computer implemented method…a health and operations monitoring system configured…a network security monitoring system configured…a fraud detection system configured… a user identification and customized content creation system configured…and an ML generation system comprising at least one processor and memory comprising executable instructions that, when executed by the at least one processor, cause the ML generation system to:”,) See MPEP 2106.05(f). The generic computer components in these steps are recited at a high-level of generality (i.e., as a generic computer component performing a generic computer function) such that it amounts no more than mere instructions to apply the exception using a generic computer component. In addition, the claim recites additional element(s) “[the selecting] including training each one of the ML pipelines of the evolved ML pipelines”, “execute the predictive maintenance on the hardware systems in the data center” that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP 2106.05(g). Similarly, the additional element “access a dataset from the operation data database, the network data database, the usage data database, or the content data database, comprising data suitable for evaluating respective performances of the plurality of ML pipelines;” describes mere data gathering which amounts to adding insignificant extra solution activity to the judicial exception. See MPEP 2106.05(g). In addition the limitations “an operation data database comprising data from the health and operations monitoring system; a network data database comprising data from the network security monitoring system; a usage data database comprising data from the fraud detection system; a content data database comprising data from the user identification and customized content creation system: an ML pipeline primitive database comprising a plurality of ML pipeline primitives;” is generally linking the use of the judicial exception to a particular technological environment or field of use, see MPEP 2106.05(h). Accordingly, the additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claim is directed to an abstract idea.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using generic computer components to perform the abstract idea amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements, “access a dataset from the operation data database, the network data database, the usage data database, or the content data database, comprising data suitable for evaluating respective performances of the plurality of ML pipelines;” are insignificant extra-solution activities that are considered well-understood, routine, conventional activities, for the following reasons. Examiner notes that accessing data amounts to receiving or transmitting data over a network (MPEP 2106.05(d)(II)(i). According to MPEP 2106.05(d)(II)(i). Further, the additional elements of “the selecting including training each one of the ML pipelines of the evolved ML pipelines;” “executing the predictive maintenance actions on the hardware systems in the data center” are insignificant extra-solution activities that are considered well-understood, routine, conventional activities. According to MPEP 2106.05(d)(I), “A factual determination is required to support a conclusion that an additional element (or combination of additional elements) is well-understood, routine, conventional activity. Berkheimer v. HP, Inc., 881 F.3d 1360, 1368, 125 USPQ2d 1649, 1654 (Fed. Cir. 2018)...The required factual determination must be expressly supported in writing, as discussed in MPEP § 2106.07(a). Appropriate forms of support include one or more of the following: ...(c) A citation to a publication that demonstrates the well-understood, routine, conventional nature of the additional element(s).” In accordance with the MPEP, the following factual determination is based on the technical publication: Lapedes, A., & Farber, R. (1987). How Neural Nets Work. In Neural Information Processing Systems. American Institute of Physics. (PTO-892 NPL Doc. U, copy attached). Lapedes et al. in pg. 443-444 “Backpropagation is a learning algorithm for neural networks that seeks to find weights… such that given an input pattern from a training set of pairs of Input/Output patterns… A popular configuration for backpropagation is a totally feedforward net… For example, the commonly used steepest descents procedure is implemented” discloses that 2 layer feedforward neural networks are popular and commonly trained with steepest descent procedure using training data of input patterns, a supervised learning technique (corresponding to routine and conventional). This procedure explains how two sets of weights each from a different hidden layer is adjusted by backpropagation/steepest decent based on input and output. Evolved pipelines include feedforward neural networks. (corresponds the selecting including training each one of the ML pipelines of the evolved ML pipelines). Further, Xiao, J., Xiong, Z., Wu, S., Yi, Y., Jin, H., & Hu, K. (2018). Disk Failure Prediction in Data Centers via Online Learning. In Proceedings of the 47th International Conference on Parallel Processing. Association for Computing Machinery. Xiao discusses abstract pg 1 “many researchers derive disk failure prediction models using machine learning techniques” introduction “proactive fault tolerance technique performs prediction before the failure actually occurs, leveraging the past behavior of a disk. Self-Monitoring, Analysis and Reporting Technology (SMART) is one such method… SMART-attribute-based proactive disk failure prediction models have been continuously proposed and become increasingly popular in recent years” discloses a system for detecting and monitoring disk failure which is popular and used by many researchers (corresponding to routine and conventional). Disk monitoring and detection as described in the specification 0046 as a type of predictive maintenance (corresponding to executing the predictive maintenance actions on the hardware systems in the data center). As such, the insignificant extra-solution activities are considered well-understood, routine, conventional activities. Therefore, the claim is not patent eligible.
Regarding Claim 2
The claim is directed to a machine. The claim recites the following limitations “iterating (d) to (e) is to be stopped comprise instruction that cause the ML generation system to determine that iterating (d) to € is to be stopped is based on at least one of the number of ML pipelines from the sub-set of evolved ML pipelines being equal to one (1), performances of the ML pipelines from the sub-set of evolved ML pipelines being equal or superior to a performance threshold required for operations of the data center, an amount of time being exceeded or an amount of processing resources being used..” Under Step 2A Prong 1, these limitations only serve to describe the abstract idea addressed in the independent claim.
Furthermore under step 2A Prong 2 and 2B, the claim does not recite additional elements to consider other than those considered in the independent claim. Accordingly, the recited additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea, nor do they amount to significantly more than the judicial exception because they do not impose any meaningful limits on practicing the abstract idea.
Regarding Claim 3
The claim is directed to a machine. The claim recites the following limitations “wherein the number of ML pipelines from the sub-set of evolved ML pipelines is half the number of ML pipelines from the evolved ML pipelines.” Under Step 2A Prong 1, these limitations only serve to describe the abstract idea addressed in the independent claim.
Furthermore under step 2A Prong 2 and 2B, the claim does not recite additional elements to consider other than those considered in the independent claim. Accordingly, the recited additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea, nor do they amount to significantly more than the judicial exception because they do not impose any meaningful limits on practicing the abstract idea.
Regarding Claim 5
The claim is directed to a machine. The claim recites the following limitations “wherein a probability that the mutation operation is applied to each of the non-cloned ML pipelines is 90% and a probability that the crossover operation is applied to each of the non-cloned ML pipelines is 10%.” Under Step 2A Prong 1, these limitations only serve to describe the abstract idea addressed in the independent claim.
Furthermore, under step 2A Prong 2 and 2B, the claim does not recite additional elements to consider other than those considered in the independent claim. Accordingly, the recited additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea, nor do they amount to significantly more than the judicial exception because they do not impose any meaningful limits on practicing the abstract idea.
Regarding Claim 9
The claim is directed to a machine. The claim recites the following limitations “wherein the sorting is based on one of non-dominated sorting or crowding distance sorting” Under Step 2A Prong 1, these limitations only serve to describe the abstract idea addressed in the independent claim.
Furthermore under step 2A Prong 2 and 2B, the claim does not recite additional elements to consider other than those considered in the independent claim. Accordingly, the recited additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea, nor do they amount to significantly more than the judicial exception because they do not impose any meaningful limits on practicing the abstract idea.
Regarding Claim 10
The claim is directed to a machine. The claim recites the following limitations “wherein the ML pipeline primitives comprise one of parameters relating to principal component analysis (PCA), parameters relating to polynomial features, parameters relating to combine features and parameters relating to a decision tree.” Under Step 2A Prong 1, these limitations only serve to describe the abstract idea addressed in the independent claim.
Furthermore under step 2A Prong 2 and 2B, the claim does not recite additional elements to consider other than those considered in the independent claim. Accordingly, the recited additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea, nor do they amount to significantly more than the judicial exception because they do not impose any meaningful limits on practicing the abstract idea.
Regarding Claim 11
The claim is directed to a machine. The claim recites the following limitations “wherein the ML pipeline comprises one or more of a pre- processing routine, a selection of an algorithm, configuration parameters associated with the algorithm, a training routine of the algorithm on a dataset and/or a trained ML model.” Under Step 2A Prong 1, these limitations only serve to describe the abstract idea addressed in the independent claim.
Furthermore under step 2A Prong 2 and 2B, the claim does not recite additional elements to consider other than those considered in the independent claim. Accordingly, the recited additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea, nor do they amount to significantly more than the judicial exception because they do not impose any meaningful limits on practicing the abstract idea.
Regarding Claim 12
Under step 1, the claim is directed to a system for operating a data center based on a generated machine learning pipeline, which is directed to a machine, one of the statutory categories. The claim recites limitations which are considered abstract ideas.
Under Step 2A Prong 1. The claims recites similar abstract ideas identified in the rejection of claim 1 as reciting mental evaluations. Additionally, the claim recites “to filter network traffic received at the data center.” Which is a mental evaluation about data, which can be performed in the mind. Filtering describes generally an abstract step of selecting which from available data. Network traffic is merely a measure of the amount of data in a network, such data can be filtered via an evaluation in the human mind.
Step 2A Prong Two Analysis: The judicial exception in not integrated into a practical application. In particular, the claim recites additional element(s) that are mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. (a computer implemented method… a health and operations monitoring system configured…a network security monitoring system configured…a fraud detection system configured… a user identification and customized content creation system configured…and an ML generation system comprising at least one processor and memory comprising executable instructions that, when executed by the at least one processor, cause the ML generation system to:) See MPEP 2106.05(f). The generic computer components in these steps are recited at a high-level of generality (i.e., as a generic computer component performing a generic computer function) such that it amounts no more than mere instructions to apply the exception using a generic computer component. Additionally, the additional element “operate, by the health and operations monitoring system, at least one of the ML pipelines from the sub-set of evolved ML pipelines [to filter network traffic received at the data center.]” Amounts to adding the words “apply it” with the judicial exception, or mere instructions to implement an abstract idea on a computer (See MPEP 2106.05(f)). In addition, the claim recites additional element(s) “the selecting including training each one of the ML pipelines of the evolved ML pipelines” that amounts to adding insignificant extra-solution activity to the judicial exception. See MPEP 2106.05(g). Similarly, the additional element “access, from the network data database, data relating to previously received data packets and a label indicating whether each data packet is legitimate or illegitimate, the data being suitable for evaluating respective performances of a plurality of ML pipelines” describes mere data gathering which amounts to adding insignificant extra solution activity to the judicial exception. See MPEP 2106.05(g). In addition the limitations “an operation data database comprising data from the health and operations monitoring system; a network data database comprising data from the network security monitoring system; a usage data database comprising data from the fraud detection system; a content data database comprising data from the user identification and customized content creation system: an ML pipeline primitive database comprising a plurality of ML pipeline primitives;” is generally linking the use of the judicial exception to a particular technological environment or field of use, see MPEP 2106.05(h). Accordingly, the additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claim is directed to an abstract idea.
Step 2B Analysis: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using generic computer components to perform the abstract idea amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements, “access, from the network data database, data relating to previously received data packets and a label indicating whether each data packet is legitimate or illegitimate, the data being suitable for evaluating respective performances of a plurality of ML pipelines” are insignificant extra-solution activities that are considered well-understood, routine, conventional activities, for the following reasons. Examiner notes that accessing data amounts to receiving or transmitting data over a network (MPEP 2106.05(d)(II)(i). According to MPEP 2106.05(d)(II)(i). Further, the additional elements of “the selecting including training each one of the ML pipelines of the evolved ML pipelines;” are insignificant extra-solution activities that are considered well-understood, routine, conventional activities. According to MPEP 2106.05(d)(I), “A factual determination is required to support a conclusion that an additional element (or combination of additional elements) is well-understood, routine, conventional activity. Berkheimer v. HP, Inc., 881 F.3d 1360, 1368, 125 USPQ2d 1649, 1654 (Fed. Cir. 2018)...The required factual determination must be expressly supported in writing, as discussed in MPEP § 2106.07(a). Appropriate forms of support include one or more of the following: ...(c) A citation to a publication that demonstrates the well-understood, routine, conventional nature of the additional element(s).” In accordance with the MPEP, the following factual determination is based on the technical publication: Lapedes, A., & Farber, R. (1987). How Neural Nets Work. In Neural Information Processing Systems. American Institute of Physics. (PTO-892 NPL Doc. U, copy attached). Lapedes et al. in pg. 443-444 “Backpropagation is a learning algorithm for neural networks that seeks to find weights… such that given an input pattern from a training set of pairs of Input/Output patterns… A popular configuration for backpropagation is a totally feedforward net… For example, the commonly used steepest descents procedure is implemented” discloses that 2 layer feedforward neural networks are popular and commonly trained with steepest descent procedure using training data of input patterns, a supervised learning technique (corresponding to routine and conventional). This procedure explains how two sets of weights each from a different hidden layer is adjusted by backpropagation/steepest decent based on input and output. Evolved pipelines include feedforward neural networks. (corresponds the selecting including training each one of the ML pipelines of the evolved ML pipelines). As such, the insignificant extra-solution activities are considered well-understood, routine, conventional activities. Therefore, the claim is not patent eligible.
Claim 13 is rejected for the reasons set forth in claim 3 in connection with claim 12
Claim 14 is rejected for the reasons set forth in claim 5 in connection with claim 12
Claim 18 is rejected for the reasons set forth in claim 9 in connection with claim 12
Claim 19 is rejected for the reasons set forth in claim 10 in connection with claim 12
Regarding Claim 21
The claim is directed to a machine. The claim recites the following limitations “to modify parameters of each ML pipeline relating to principal component analysis (PCA) used in the ML pipeline, modifying parameters relating to polynomial features of the ML pipeline, modifying parameters relating to combine features of the ML pipeline, or modifying parameters relating to a decision tree of the ML pipeline.” Under Step 2A Prong 1, these limitations correspond to a mental evaluation.
The claim recites the following additional element(s), in addition to those already identified in the parent claim: (“instructions that cause the ML generation system”) that are mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See MPEP 2106.05(f). As such, these additional elements do not integrate the abstract idea into a practical application nor provide significantly more.
Regarding Claim 22
The claim is directed to a machine. The claim recites the following limitations “to modify hyper parameters of each ML pipeline.” Under Step 2A Prong 1, these limitations correspond to a mental evaluation.
The claim recites the following additional element(s), in addition to those already identified in the parent claim: (“instructions that cause the ML generation system to…”) that are mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See MPEP 2106.05(f). As such, these additional elements do not integrate the abstract idea into a practical application nor provide significantly more.
Regarding Claim 23
The claim is directed to a machine. The claim recites the following limitations “wherein the scoring is further based on precision, recall, false positive rate, and minimizing fitting time.” Under Step 2A Prong 1, these limitations only serve to describe the abstract idea addressed in the independent claim.
Furthermore under step 2A Prong 2 and 2B, the claim does not recite additional elements to consider other than those considered in the independent claim. Accordingly, the recited additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea, nor do they amount to significantly more than the judicial exception because they do not impose any meaningful limits on practicing the abstract idea.
Regarding Claim 24
The claim is directed to a machine. The claim recites the following limitations “generate, for each pipeline of the plurality of ML pipelines, hyper parameters for an ML algorithm of the respective ML pipeline.” Under Step 2A Prong 1, these limitations only serve to describe the abstract idea addressed in the independent claim.
The claim recites the following additional element(s), in addition to those already identified in the parent claim: (“instructions that cause the ML generation system to…”) that are mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See MPEP 2106.05(f). As such, these additional elements do not integrate the abstract idea into a practical application nor provide significantly more.
Regarding Claim 25
The claim is directed to a process. The claim recites the following limitations “select a type of ML algorithm to include in the respective ML pipeline; and determine, based on the type of ML algorithm, hyper parameters for the ML algorithm, wherein the hyper parameters comprise polynomial features stored in a feature matrix.” Under Step 2A Prong 1, these limitations only serve to describe the abstract idea addressed in the independent claim.
The claim recites the following additional element(s), in addition to those already identified in the parent claim: (“instructions that cause the ML generation system to…”) that are mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. See MPEP 2106.05(f). As such, these additional elements do not integrate the abstract idea into a practical application nor provide significantly more.
Claim Rejections - 35 U.S.C. § 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.
Claim(s) 1-3, 9, 10, 11, 21-22, 24-25 are rejected under 35 U.S.C. § 103 as being unpatentable over Qi et al “DarwinML: A Graph-based Evolutionary Algorithm for Automated Machine Learning” hereinafter Qi, further in view of Gijsbers, P. “Automatic construction of machine learning pipelines” hereinafter Gijsbers, further in view of Lisha et al. “Hyperband: A Novel Bandit-Based Approach to Hyperparameter Optimization” hereinafter Lisha, further in view of Xiao “Disk Failure Prediction in Data Centers via Online Learning” hereinafter Xiao, further in view of Al-Naymat “Dynamics-Based Approach for Accurate User Identification and Authentication using Machine Learning Techniques”
Regarding claim 1
Qi teaches, A computer-implemented method for generating a machine learning (ML) pipeline, the method comprising: (pg 3 Section IV “ Our goal is to find the architecture that performs best on a given ML task”) an ML pipeline primitive database comprising a plurality of ML pipeline primitives; and an ML generation system comprising at least one processor and memory comprising executable instructions that, when executed by the at least one processor, cause the ML generation system to: (pg 3 Section 4A “The evolutionary search algorithm based on tournament selection is employed to explore for architectures in the complex configuration space” candidate architectures comprise the database pg 6 Section B “Results were listed in Table III. Random forest is chosen to be the baseline. Being EA based AutoML models, TPOT and Autostacker evolved for 100 and 3 generations, respectively” such generation is understood by Phosita to be performed with a computer. Further, TPOT and autostacker are software systems requiring a computer with memory and processor and instructions.)(a) generate from the plurality of ML pipeline primitives int eh ML pipeline primitive data base, a plurality of ML pipelines each associated with a respective ML pipeline configuration (pg 3 Section IVA ¶02 “in initialization the first generation is build up by individuals according to the convention in EA, generated by the random operation” Section IVB ¶01 “on: The random operation is designed to randomly sample individuals in the configuration or searching space. The pseudo code of the random operation is shown in Algorithm 1. In the algorithm, K and D are the numbers of vertices and layers of the DAG to be generated as an individual” individuals corresponding to ML pipelines are generated based on randomly assigned primitives, these being the number of vertices, layers, and size of layers. A plurality of individuals are generated in the initialization stage.) (b) access a dataset from the operation data database, the network data database, the usage data database, or the content data database comprising data suitable for evaluating respective performances of the plurality of ML pipelines ( pg 6 Section A ¶01-03 “DarwinML was tested on the same datasets used in Autostacker [15], where 15 datasets were selected from PMLB… On each dataset, DarwinML were repeated 10 times with random initialization” Figure 6 Caption” Fig. 6: The scatter plot shows the performance evolution when architectures is searched” the performance is measured using a variety of datasets comprising instances features and classes for evaluating performance. As shown in Figure 6 the performance is evaluated for each generation.) (c) select a sub-set of ML pipelines from the plurality of ML pipelines, …, a number of ML pipelines from the sub-set of ML pipelines being less than a number of ML pipelines from the plurality of ML pipelines (d) evolve the sub-set of ML pipelines to generate evolved ML pipelines, wherein the evolving comprises cloning at least one of the sub-set of the ML pipelines to generate a corresponding at least one evolved ML pipeline, wherein the evolving further comprises modifying parameters of each non-cloned ML pipeline of the sub-set of ML pipelines using a mutation operation or a crossover operation to generate additional evolved ML pipelines (pg 3 Section IVA ¶02, Also Algorithm 5 and Figure 2 “In In the second generations, individuals evolve by applying all four operations. Firstly, the keep best operator is applied to get top 15% best individual set from first generation… Then, the top 15% best individuals are inherent back in the population to ensure models with good performance are reserved… on. Finally, fitness of all individuals, except the ones inherent from keep best, are evaluated after a training procedure” before evolving tournament selection is applied to the first generation to select the top performing individuals, The selected “keep best” individuals are cloned into the next iteration because they are selected without modification. As shown in the figure the remaining Individuals which are not cloned are modified using at least a mutation or random operation which corresponds to mutation operation claimed as it modifying parameters of the pipeline.
PNG
media_image1.png
186
757
media_image1.png
Greyscale
) (e) selecting a sub-set of evolved ML pipelines from the evolved ML pipelines, …, a number of ML pipelines from the sub-set of evolved ML pipelines being less than a number of ML pipelines from the evolved ML pipelines and (f) iterate (d) to (e) until determination is made that iterating (d) to (e) is to be stopped. (pg 3 Section IVA ¶02-¶05 “In the second generations, individuals evolve by applying all four operations. Firstly, the keep best operator is applied to get top 15% best individual set from first generation. Then, tournament selection[35] picks the top model from a randomly chosen sub-group in previous generation and keep best…The evolution stops when a predefined duration or number of population has been reached. The final output is the individual with the highest fit” again for each subsequent generation the keep best operator is applied to select a smaller subset of models, this process is iterated until a predetermined population is reached or a predetermined duration is reached.) the selecting including training each one of the ML pipelines of the evolved ML pipelines (pg 3 column 2 para. 01 “t. A number of top models are selected by repeating random grouping and tournament selection. Secondly, the mutation and heredity operations are applied to these promising individuals, and new individuals are created for the current generation… Finally, fitness of all individuals, except the ones inherent from keep best, are evaluated after a training procedure” the evolved models of each generation are trained in order to evaluate fitness for keeping the best models for the next generation. This is also described in Figure 2.)
Qi does not explicitly teach, determine, based on output from the ML pipeline, predictive maintenance actions to be performed on hardware systems in a data center … the selecting including training each one of the ML pipelines of the evolved ML pipelines for a second amount of amount of time being larger than the first amount of time a number of ML pipelines from the sub-set of evolved ML pipelines being less than a number of ML pipelines from the evolved ML pipelines; wherein at each iteration of (d) to (e) a number of ML pipelines from the sub-set of evolved ML pipelines is less than a number of ML pipelines selected in a previous iteration of (d) to (e) … determine, based on output from the ML pipeline, predictive maintenance actions to be performed on the hardware systems in the data center; and (h) execute the predictive maintenance actions to be performed on the hardware systems in the data center …sorting the ML pipelines of the evolved ML pipelines, wherein the scoring comprises scoring each ML pipeline based on an accuracy of the respective ML pipeline and a complexity of the respective ML pipeline, and wherein the scoring prioritizes ML pipelines having a higher accuracy and lower complexity; … a health and operations monitoring system configured to monitor health and operations of the hardware systems in the data center; a network security monitoring system configured to monitor the security of a network of the data center; a fraud detection system configured to detect fraud attempts against the data center: a user identification and customized content creation system configured to manage user accounts associated with the data center; an operation data database comprising data from the health and operations monitoring system; a network data database comprising data from the network security monitoring system; a usage data database comprising data from the fraud detection system;
a content data database comprising data from the user identification and customized content creation system:
Gijsbers however when addressing the use of genetic algorithms for model selection teaches, sorting the ML pipelines of the evolved ML pipelines, wherein the scoring comprises scoring each ML pipeline based on an accuracy of the respective ML pipeline and a complexity of the respective ML pipeline, and wherein the scoring prioritizes ML pipelines having a higher accuracy and lower complexity; (pg 21 “The goal of TPOT is to not only find a pipeline which has good accuracy, but also to keep the total size of the pipeline as small as possible… To achieve this, the selection procedure of the multiobjective genetic algorithm Non-dominated Sorting Genetic Algorithm II (NSGA-II) is used [22]. It finds the Pareto fronts of the trade-off between performance and pipeline length” the selection and scoring is according to a pareto front between complexity, or pipeline length, and accuracy, or performance. Such selection/scoring prioritizes pipelines having a higher accuracy and lower complexity as claimed.)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the model selection system of Qi with the improved model selection system which increases the data volume size for each iterative selection and which selects based on accuracy and complexity disclosed by Gijsbers . One would have been motivated to make such a combination to “to find pipelines of similar quality in much less time, making the tool more accessible and practical by requiring less computational time” and to select low complexity models because “One reason is the assumption that less-complex pipelines will also generalize better” (pg 21 and 23 Gijsbers)
Qi/Gijsbers does not explicitly teach, for determining maintenance actions to be performed on hardware systems in a data center… wherein the dataset comprises status reports generated by devices in the data center, the status reports indicating when a component of the devices has failed…the selecting including training each one of the ML pipelines of the evolved ML pipelines for a second amount of amount of time being larger than the first amount of time a number of ML pipelines from the sub-set of evolved ML pipelines being less than a number of ML pipelines from the evolved ML pipelines; wherein at each iteration of (d) to (e) a number of ML pipelines from the sub-set of evolved ML pipelines is less than a number of ML pipelines selected in a previous iteration of (d) to (e) … determining, based on output from the ML pipeline, predictive maintenance actions to be performed on the hardware systems in the data center; and (h) executing the predictive maintenance actions to be performed on the hardware systems in the data center; … a health and operations monitoring system configured to monitor health and operations of the hardware systems in the data center; a network security monitoring system configured to monitor the security of a network of the data center; a fraud detection system configured to detect fraud attempts against the data center: a user identification and customized content creation system configured to manage user accounts associated with the data center; an operation data database comprising data from the health and operations monitoring system; a network data database comprising data from the network security monitoring system; a usage data database comprising data from the fraud detection system;
a content data database comprising data from the user identification and customized content creation system:
Lisha however when addressing successive having of child pipelines in hyper parameter optimization teaches, the selecting including training each one of the ML pipelines of the evolved ML pipelines for a second amount of amount of time being larger than the first amount of time a number of ML pipelines from the sub-set of evolved ML pipelines being less than a number of ML pipelines from the evolved ML pipelines; wherein at each iteration of (d) to (e) a number of ML pipelines from the sub-set of evolved ML pipelines is less than a number of ML pipelines selected in a previous iteration of (d) to (e) (pg 6 Section 3.1 “Hyperband extends the SuccessiveHalving algorithm… and calls it as a subroutine… The idea behind the original SuccessiveHalving algorithm follows directly from its name: uniformly allocate a budget to a set of hyperparameter configurations, evaluate the performance of all configurations, throw out the worst half, and repeat until one configuration remains…. SuccessiveHalving requires the number of configurations n as an input to the algorithm. Given some finite budget B (e.g., an hour of training time to choose a hyperparameter configuration), B/n resources are allocated on average across the configurations.” further pg 7-8 algorithm 1 “There are two components to Hyperband; (1) the inner loop invokes SuccessiveHalving for fixed values of n and r (lines 3–9)” as described by successive halving algorithm the budget or time spent per configuration or “pipeline” is divided by the number of remaining configurations, each iteration half are removed thus the total budget, or time spent training, for the next iteration increases by a factor of two for each pipeline. Further each iteration is the total remaining configurations is halved, thus the number of pipelines removed in each iteration is decreasing.)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the model selection system of Qi/Gijsbers with the improved model selection system which is optimized for a given amount of compute resources described by Lisha . One would have been motivated to make such a combination because the Neural architecture search described by Lisha “Hyperband is able to exploit situations in which adaptive allocation works well, while protecting itself in situations where more conservative allocations are required.” (Lisha pg 8)
Qi/Gijsbers/Lisha does not explicitly teach, for determining maintenance actions to be performed on hardware systems in a data center … determine, based on output from the ML pipeline, predictive maintenance actions to be performed on the hardware systems in the data center; and (h) executing the predictive maintenance actions to be performed on the hardware systems in the data center…; … a health and operations monitoring system configured to monitor health and operations of the hardware systems in the data center; a network security monitoring system configured to monitor the security of a network of the data center; a fraud detection system configured to detect fraud attempts against the data center: a user identification and customized content creation system configured to manage user accounts associated with the data center; an operation data database comprising data from the health and operations monitoring system; a network data database comprising data from the network security monitoring system; a usage data database comprising data from the fraud detection system; a content data database comprising data from the user identification and customized content creation system:
Xiao however when the use of machine learning algorithms to detect component failure teaches, for determining maintenance actions to be performed on hardware systems in a data center … determinine, based on output from the ML pipeline, predictive maintenance actions to be performed on the hardware systems in the data center; and (h) executing the predictive maintenance actions to be performed on the hardware systems in the data center. (abstract pg 1 “In this paper, we introduce a novel disk failure prediction model using Online Random Forests (ORFs)” introduction pg 1 and 2 “More importantly, ORF generates random trees on-the-fly using the gradually gathered SMART samples.… We simulate the long-term use of ORF-based prediction models and demonstrate the effectiveness and adaptivity of our method in real-world data centers” smart samples correspond to maintenance actions, SMART is a test run on hard drives to access remaining lifetime. Pg 3 “In this paper, we attempt to adopt online learning method for disk failure prediction and present an ORF-based method for practical usage. ORF can evolve with sequential arrival of data on-the-fly and forget old information by controlled discarding outdated trees” because the model is learning online SMART data is continually retrieved as the models are evolved/updated.) a health and operations monitoring system configured to monitor health and operations of the hardware systems in the data center;… a network security monitoring system configured to monitor the security of a network of the data center… an operation data database comprising data from the health and operations monitoring system;… a network data database comprising data from the network security monitoring system; (abstract pg 1 “Disk failure has become a major concern with the rapid expansion of storage systems in data centers. Based on SMART (Self-Monitoring, Analysis and Reporting Technology)… a novel disk failure prediction model” disk failure prediction models monitor health or failure. Similarly, the health of the disk is a measure of how the secure the data is in the network from being lost. pg 1 “The experiments on dataset collected from 34,535 disks, monitored over three years” the location of the dataset is the claimed database)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the model selection system of Qi/Gijsbers/Lisha to be applied to online learning in drive disk failure detection as described in Xiao . One would have been motivated to make such a combination because “[The system] achieves stable failure detection rates of 93-99% with low false alarm rates. Furthermore, we demonstrate the ability of our approach on maintaining stable prediction performance for the long-term usage in data centers.” (Abstract Xiao)
Qi/Gijsbers/Lisha/Xiao does not explicitly teach, a fraud detection system configured to detect fraud attempts against the data center: a user identification and customized content creation system configured to manage user accounts associated with the data center; a usage data database comprising data from the fraud detection system; a content data database comprising data from the user identification and customized content creation system:
Al-Naymat when addressing machine learning based user identification teaches, a fraud detection system configured to detect fraud attempts against the data center: a user identification and customized content creation system configured to manage user accounts associated with the data center; a content data database comprising data from the user identification and customized content creation system: a usage data database comprising data from the fraud detection system; (abstract pg 1 “we applied three different classification techniques namely: J48, Random Forest and Multi-layer Perceptron (MLP), to accurately identify the user behavior (legitimate or illegitimate) and its authority” pg 5 Section 2 “Due to the lack of available datasets in the KSD field as well as accompanying documentation, also because there is no standard dataset in KSD field, so we decided to build our own dataset…. The website was installed on an Ubuntu OS desktop as server, and could be accessed from other desktops with Windows 7 OS for the users… If the user did not have an account then they would be asked to sign up. All users could choose their own user name, email and password, and the chosen data were used only for login and to avoid duplicates in the users…. The website was developed in order to record the timestamp for two actions while the user was typing the chosen password by pressing the key action and releasing the key action… Finally, when the users entered the words three times on three different days (sessions), they would have a complete profile, containing 9 records of data.” user account data is collected for customized content the recorded data comprising user identification and customized content created. The key actions recorded is usage data. The system for user identification amounts to fraud detection, as detection of non-authenticated users is fraud detection)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the model selection system of Qi/Gijsbers/Lisha/Xiao to be applied to user authentication using ML models as described in Al-Naymat. One would have been motivated to make such a combination because “In our experiments, we obtained the highest result using Random Forest classifier with an accuracy rate equal to 93.13%. We similarly ran experiments on the authentication process (verification mode) and obtained an accuracy result of 94.9% using the MLP classifier” (Al-Naymat conclusion). Further Qi describes the motivations to use more complex model architectures discovered via model search “According to a comparison between tree-based and graph-based representations for symbolic regression a graph-based representation is much flexible and general than the tree-based one” (Qi introduction) further noting that “use GP to evolve a better function to approximate target value” (pg 2 Qi)
Regarding claim 2
Qi/Gijsbers/Lisha/Xiao/Al-Naymat teaches claim 1
Further Qi teaches, instructions that cause the ML generation system to determine that that iterating (d) to (e) is to be stopped is based on at least one of the number of ML pipelines from the sub-set of evolved ML pipelines being equal to one (1), performances of the ML pipelines from the sub-set of evolved ML pipelines being equal or superior to a performance threshold required for operations of the datacenter, an amount of time being exceeded or an amount of processing resources being used.(pg 3 Section IVA ¶05 “The evolution stops when a predefined duration or number of population has been reached. The final output is the individual with the highest fit” The algorithm stops when a final output pipeline is discovered or when an amount of time is exceeded.)
Regarding claim 3
Qi/Gijsbers/Lisha/Xiao/Al-Naymat teaches claim 1
Further Gijsbers teaches, wherein the number of ML pipelines from the sub-set of evolved ML pipelines is half the number of ML pipelines from the evolved ML pipelines (pg 25 Section 3.1.2 ¶02 “Let the dataset contain N instances, then the final layer will always train on the entire dataset, and each subsequent layer will use half of the data the layer above did. In this study, the number of layers used is 4, for every dataset. The respective sample sizes used at each layer are thus N/8, N/4, N/2 and N. We use the term higher layer loosely to denote layers which sample more of the entire dataset” pg 27 and pg 28 LayeredEA “First, a new population is created from the individuals evaluated during the last generation in the same layer, by performing mutation and cross-over… Then, every g generations, the best individuals from each layer get passed to the next one. In our configuration we chose to transfer half of the layer’s population” examiner notes that after each evolution stage, half of the pipelines are selected. This is also shown in Figure 3.1, where the top k, or top half, of the evolved pipelines are transferred to the next layer.)
Regarding claim 9
Qi/Gijsbers/Lisha/Xiao/Al-Naymat claim 1
Further Gijsbers teaches, wherein the sorting is based on one of non-dominated sorting or crowding distance sorting. (Pg 37 “the selection procedure of the multi objective genetic algorithm Non-dominated Sorting Genetic Algorithm II…It finds the Pareto fronts of the trade-off between performance and pipeline length,… Moreover, it also considers a crowding distance for individuals in the Pareto front, so that it will aim to select individuals spaced out along the Pareto front”)
Qi/Gijsbers/Lisha/Xiao are combined for the reasons set forth in claim 1
Regarding claim 10
Qi/Gijsbers/Lisha/Xiao/Al-Naymat teaches claim 1
Further Gijsbers teaches, wherein the ML pipeline primitives comprise one of parameters relating to principal component analysis (PCA), parameters relating to polynomial features, parameters relating to combine features and parameters relating to a decision tree (Section 2.4.2 “To make this possible, individuals are represented as tree-like structures, where each internal node in the tree is a primitive (also called function), and each leaf is a terminal…In the case of TPOT, the primitives are various algorithms, from preprocessing to machine learning algorithms, the terminals are then variables such as (preprocessed) datasets or hyperparameters…At the root of the tree is the machine learning algorithm which is able to do the final prediction to a classification or regression problem”)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the primitive initialization of ML models discussed by Qi with the tree based primitives disclosed by Gijsbers . One would have been motivated to make such a combination because tree based methods allow an algorithm to determine the complexity of the pipeline by accessing the length of the pipeline. Gijsbers notes that “less-complex pipelines will also generalize better” (Gijsbers pg 21) and further Qi mentions that generalized models would be valuable for large scale data sets. “In the future, we plan to generalize the approach to neural architecture search by extending the ML models with building blocks of neural networks to get better performance on large scale datasets” (pg 7 Qi)
Regarding claim 11
Qi/Gijsbers/Lisha/Xiao/Al-Naymat teaches claim 1
Further Gijsbers teaches, wherein the ML pipeline comprises one or more of a pre- processing routine, a selection of an algorithm, configuration parameters associated with the algorithm, a training routine of the algorithm on a dataset and/or a trained ML model ( Figure 2.6 pg 21 depicts an example pipeline contains at least per-processing routine and selection of algorithm
PNG
media_image2.png
474
925
media_image2.png
Greyscale
)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the hyperparameters selected to optimize as discussed by Qi with the broader set of hyperparameters including a the algorithm selection disclosed by Gijsbers . One would have been motivated to make such a combination because it allows for the algorithm to search a broader set of possible configurations. “In the future, we plan to generalize the approach to neural architecture search by extending the ML models with building blocks of neural networks to get better performance on large scale datasets” (pg 7 Qi)
Regarding claim 21
Qi/Gijsbers/Lisha/Xiao/Al-Naymat teaches claim 1
Gijsbers further teaches, instructions that cause the ML generation system to modify parameters of each ML pipeline comprises modifying parameters of each ML pipeline relating to principal component analysis (PCA) used in the ML pipeline, modify parameters relating to polynomial features of the ML pipeline, modify parameters relating to combine features of the ML pipeline, or modify parameters relating to a decision tree of the ML pipeline. ( pg 20 “TPOT uses genetic programming and thus represents machine learning pipelines as a tree-like structure. In the case of TPOT, the primitives are various algorithms, from preprocessing to machine learning algorithms, the terminals are then variables such as (preprocessed) datasets or hyperparameters. An example of such a tree-based pipeline is shown in Figure 2.6
PNG
media_image3.png
316
600
media_image3.png
Greyscale
” as shown in the figure the pipeline operators to be modified include PCA and polynomial features, parameters are combined and related to a decision tree.)
Regarding claim 22
Qi/Gijsbers/Lisha/Xiao/Al-Naymat teaches claim 1
Gijsbers further teaches, instructions that cause the ML generation system to modify parameters of each ML pipeline comprises modifying hyper parameters of each ML pipeline. ( Section 2.2 pg 8 “Meta-learning is the process about learning how machine learning algorithms perform across a range of tasks. It aims to learn which algorithm will work well for a dataset with certain characteristics, or which hyperparameters will give a good performance.” pg 35 “All hyperparameter combinations for each baselearner considered can be found in Table 4.1, and form a subset of the algorithms and hyperparameter combinations considered by default in TPOT.” The entire disclosure of Gijsbers is directed towards different types of “meta-learning” approaches for evolving ML pipeline values including hyper parameters. TPOT as discussed above is one type of meta-learner that implements the claim function.)
Regarding claim 24
Qi/Gijsbers/Lisha/Xiao/Al-Naymat teaches claim 1
Gijsbers teaches, instructions that cause the ML generation system to generate the plurality of ML pipelines comprises generating, for each pipeline of the plurality of ML pipelines, hyper parameters for an ML algorithm of the respective ML pipeline. ( pg 4 abstract “we construct models which can recommend algorithms and hyperparameter settings to begin TPOT's pipeline optimization procedure with”)
Regarding claim 25
Qi/Gijsbers/Lisha/Xiao/Al-Naymat teaches claim 1
Gijsbers teaches, instructions that cause the ML generation system to, for each ML pipeline of the plurality of pipelines select a type of ML algorithm to include in the respective ML pipeline; and determine, based on the type of ML algorithm, hyper parameters for the ML algorithm, ( pg 4 abstract “we construct models which can recommend algorithms and hyperparameter settings to begin TPOT's pipeline optimization procedure with”) wherein the hyper parameters comprise polynomial features stored in a feature matrix. ( pg 20 “TPOT uses genetic programming and thus represents machine learning pipelines as a tree-like structure. In the case of TPOT, the primitives are various algorithms, from preprocessing to machine learning algorithms, the terminals are then variables such as (preprocessed) datasets or hyperparameters. An example of such a tree-based pipeline is shown in Figure 2.6
PNG
media_image3.png
316
600
media_image3.png
Greyscale
” the hyperparameters include for example polynomial features which are stored in computer memory, nominally a feature matrix. Examiner encourages further elaborating on the structure of the matrix itself if the manner in which the features are stored is the inventive concept.)
Claim(s) 5 are rejected under 35 U.S.C. § 103 as being unpatentable over Qi/Gijsbers/Lisha/Xiao/Al-Naymat, further in view of Tuson et al. “Cost Based Operator Rate Adaptation: An Investigation” hereinafter Tuson, further in view of Mohammed et al US Document ID US 20150170052 A1 hereinafter Mohammed.
Regarding claim 5
Qi/Gijsbers/Lisha/Xiao/Al-Naymat teaches claim 1
Qi/Gijsbers/Lisha/Xiao/Al-Naymat does not explicitly teach, wherein a probability that the mutation operation is applied to each of the non-cloned ML pipelines is 90% and a probability that the crossover operation is applied to each of the non-cloned ML pipelines is 10%.
Tuson however when discussing determination of crossover and mutation probabilities in a genetic algorithm teaches, a probability that the crossover operation is applied to each of the non-cloned ML pipelines is 10%. (pg 463 Section 4.1 “The effect of varying crossover probability on a genetic algorithm with fixed operator probabilities was investigated. An exhaustive search was made of the operator probabilities: a GA was run for crossover probabilities 0.05 to 0.95 with steps of 0.05 (mutation was applied otherwise).” The GA algorithm is able to be run with crossover probability in the range of .05-.95.)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the model selection system of Qi/Gijsbers/Lisha/Xiao/Al-Naymat to implement a fixed crossover rate of during evolution as disclosed by Tuson. One would have been motivated to make such a combination to discover “The effect of varying crossover probability on a genetic algorithm with fixed operator probabilities” and to “give some indication of how hard the GA is to tune” (Section 4.1 Tuson pg 463)
Qi/Gijsbers/Lisha/Xiao/Al-Naymat/Tuson does not explicitly teach, wherein a probability that the mutation operation is applied to each of the non-cloned ML pipelines is 90%
Mohammed however when discussing determination of crossover and mutation probabilities in a genetic algorithm teaches, wherein a probability that the mutation operation is applied to each of the non-cloned ML pipelines is 90% (¶0073 “FIG. 6 (b) shows the resource profile of a conventional early-start schedule and the proposed GA based schedule… and the mutation percentage is 90%.”)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify genetic algorithm used in Qi/Gijsbers/Lisha/Xiao/Al-Naymat/Tuson to implement a high mutation evolution percentage as disclosed in Mohammed. One would have been motivated to make such a combination to discover to enforce greater randomization to avoid local minima as noted in Mohammed “The purpose of mutation in GAs is to allow the algorithm to avoid local minima by preventing the population of chromosomes from becoming too similar to each other, thus slowing or even stopping evolution” (Mohammed ¶0020)
Claim(s) 12-13, 18 and 19 are rejected under 35 U.S.C. § 103 as being unpatentable over Qi/Gijsbers/Lisha/Xiao/Al-Naymat, further in view of Roosmalen “Applying Deep Learning on Packet Flows for Botnet Detection” hereinafter Roosmalen.
Regarding claim 12
Qi/Gijsbers/Lisha/Xiao/Al-Naymat teaches the limitations of claim 1 commonly recited in claim 12
Qi/Gijsbers/Lisha/Xiao/Al-Naymat does not explicitly teach, generating and executing a machine learning (ML) pipeline to prevent illegitimate network traffic from accessing the data center access, from the network data database, data relating to previously received data packets and a label indicating whether each data packet is legitimate or illegitimate, the data being suitable for evaluating respective performances of a plurality of ML pipelines…ML pipelines is configured to receive network packets and discriminate between legitimate and illegitimate network packets in the received network packets;… operate, by the health and operations monitoring system at least one of the ML pipelines from the sub-set of evolved ML pipelines to filter network traffic received at the data center.
Qi/Gijsbers/Lisha/Xiao/Al-Naymat is combined for the reasons set forth in the rejection of claim 1
Roosmalen however when addressing using a machine learning model for filtering illegitimate network traffic teaches, access, from the network data database, data relating to previously received data packets and a label indicating whether each data packet is legitimate or illegitimate, the data being suitable for evaluating respective performances of a plurality of ML pipelines (pg 4 “We performed our experiments on a 83 GB dataset that we assembled from the following 5 datasets:… UNB ISCX IDS dataset… is a labeled network dataset with complete payload data. It covers a wide range of normal traffic [legitimate] as well as non-P2P-botnet [illegitimate] attack traffic”) generating and executing a machine learning (ML) pipeline to prevent illegitimate network traffic from accessing the data center …ML pipelines is configured to receive network packets and discriminate between legitimate and illegitimate network packets in the received network packets;… operate, by the health and operations monitoring system at least one of the ML pipelines from the sub-set of evolved ML pipelines to filter network traffic received at the data center. (abstract “We present a novel approach to botnet detection that applies deep learning on flows of TCP/UDP/IP-packets.” Pg 6 “Table 2 lists our results obtained with a network architecture having three hidden-layers… The table shows results for both the unbalanced test set (2% botnet traffic, 98% non-botnet traffic)” a network with a given configuration, pipeline, is applied to distinguish/filter between network traffic at a data center.)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to use the determined machine learning pipeline of Qi/Gijsbers/Lisha/Xiao/Al-Naymat for use on monitoring data center operations as disclosed in Roosmalen . One would have been motivated to make such a combination to because “, we obtained 99.7% accuracy for classifying P2P-botnet traffic. This is comparable to or better than conventional botnet detection approaches, while reducing efforts for feature engineering and feature selection to a minimum.” (Roosmalen Abstract)
Claim 13 is rejected for the reasons set forth in claim 3 in connection with claim 12
Claim 18 is rejected for the reasons set forth in claim 9 in connection with claim 12
Claim 19 is rejected for the reasons set forth in claim 10 in connection with claim 12
Claim(s) 14 are rejected under 35 U.S.C. § 103 as being unpatentable over Qi/Gijsbers/Lisha/Xiao/Al-Naymat/Roosmalen, further in view of Tuson et al. “Cost Based Operator Rate Adaptation: An Investigation” hereinafter Tuson, further in view of Mohammed et al US Document ID US 20150170052 A1 hereinafter Mohammed.
Claim 14
Qi/Gijsbers/Lisha/Xiao/Al-Naymat/Roosmalen teaches claim 12
The limitations of Claim 14 are rejected for the reasons set forth in claim 5 in connection with claim 12
Qi/Gijsbers/Lisha/Xiao/Al-Naymat/Roosmalen are combined for the reasons set forth in the rejection of claim 12.
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the model selection system of Qi/Gijsbers/Lisha/Xiao/Al-Naymat/Roosmalen to implement a fixed crossover rate of during evolution as disclosed by Tuson. One would have been motivated to make such a combination to discover “The effect of varying crossover probability on a genetic algorithm with fixed operator probabilities” and to “give some indication of how hard the GA is to tune” (Section 4.1 Tuson pg 463)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify genetic algorithm used in Qi/Gijsbers/Lisha/Xiao/Al-Naymat/Roosmalen/Tuson to implement a high mutation evolution percentage as disclosed in Mohammed. One would have been motivated to make such a combination to discover to enforce greater randomization to avoid local minima as noted in Mohammed “The purpose of mutation in GAs is to allow the algorithm to avoid local minima by preventing the population of chromosomes from becoming too similar to each other, thus slowing or even stopping evolution” (Mohammed ¶0020)
Claim(s) 23 are rejected under 35 U.S.C. § 103 as being unpatentable over Qi/Gijsbers/Lisha/Xiao/Al-Naymat, further in view of Zhen et al “Performance assessment of the deep learning technologies in grading glaucoma severity” hereinafter Zhen.
Regarding claim 23
Qi/Gijsbers/Lisha/Xiao/Al-Naymat teaches the claim 1
Gijsbers teaches, scoring is further based …and minimizing fitting time. (pg 18-19 Section 2.4.1 “In genetic algorithms, the aim is to find an optimal solution to a defined function…. Once the representation of individuals is defined, as well as the mutation, crossover and selection functions, the algorithm executes the following steps:… Repeat steps 2-5 until a specified termination condition is reached. These can include a time limit being met….” Pg 20 Section 2.4.2 “TPOT uses genetic programming and thus represents machine learning pipelines as a tree-like structure” TPOT is a type of genetic evolving model whose scoring is limited or based on meeting a time limit being set, thus scoring is based on minimizing a fitting time according to the limit set.)
Qi/Gijsbers/Lisha/Xiao/Al-Naymat does not explicitly teach, scoring… based on precision, recall, false positive rate
Zhen when addressing selecting an optimal model from a set of multiple similar machine learning models teaches, scoring… based on precision, recall, false positive rate ( Section 2.5 pg 5 “Both the original and fine-tuned CNN architectures mentioned above were constructed through the Keras… Classification experiments were performed on a PC… The classification performance was assessed quantitatively using the precision, recall, F1 score…” definitionally, and as shown in equation 1 of the art, precision is based on the false positive rate.)
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the model performance assessment methods described by Qi/Gijsbers/Lisha/Xiao to include common measure statistics for accessing model performance as described by Zhen. One would have been motivated to make such a combination in order to accurate describe the performance difference between sets of models or pipelines. As noted in Zhen “Comprehensive performance comparison of these typical CNN architectures for glaucoma classification is not only helpful for our understanding of their unique characteristics, but also may aid in developing novel CNN frameworks” (Zhen pg 9).
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOHNATHAN R GERMICK whose telephone number is (571)272-8363. The examiner can normally be reached M-F 7:30-4:30.
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, Kakali Chaki can be reached on 571-272-3719. 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.
/J.R.G./
Examiner, Art Unit 2122
/KAKALI CHAKI/Supervisory Patent Examiner, Art Unit 2122