DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claims 1-20 are pending in this application.
Response to Arguments
The objections to the drawings have been withdrawn.
Applicant’s arguments regarding the rejections of claims 1-20 under 35 U.S.C. 112b have been fully considered and are persuasive. The rejections have been withdrawn.
Applicant's arguments regarding the 35 U.S.C. 103 rejections of claims 1-20 have been fully considered but they are unpersuasive.
Regarding the 35 U.S.C. 103 rejection, the applicant argues the following in the remarks:
In accordance with the claimed invention, there are a plurality of different solver types. A solver selection machine learning model is trained and configured to automatically determine an optimal solver type from among the plurality of different solver types for each of a plurality of different input problem types. Thus, when an input problem originating from a client computing entity is received, the already-trained solver selection machine learning model is applied to determine the optimal solver type for that input problem based on one or more problem features of the input problem, and then the optimal solver type is executed to provide an optimal solution for the input problem. This is not what Wang is doing. Wang is not training a solver selection machine learning model so that the trained model can select an optimal solver type for a received input problem from among a plurality of solver types.
Examiner has thoroughly considered Applicant' s arguments, but respectfully finds them unpersuasive for at least the following reasons:
As to point (a), the examiner respectfully disagrees. Wang recites on pg. 6 lines 18-20 “training the meta-knowledge training incubator; obtaining an optimal algorithm number for each meta-feature by the trained meta-knowledge training incubator”. Therefore, the meta-knowledge training incubator, which determines an optimal algorithm, has already been trained.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 2, 13, 14, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al. (CN111210023A hereinafter Wang) in view Lindawati (US 20180032955 A1).
The claim mappings of Wang are from a translation of CN111210023A.
Wang and Lindawati were cited in a prior office action.
As per claim 1, Wang teaches computer-implemented method comprising: training and configuring a solver selection machine learning model using a training set of input problems having a plurality of different input problem types and a plurality of different solver types to automatically determine an optimal solver type from among the plurality of different solver types for each of the plurality of different input problem types (pg. 2 lines 38-40 perform knowledge processing to obtain an algorithm selection training set including a one-to-one correspondence between different classification problem data sets and their corresponding learning algorithms; pg. 5 lines 35-36 train the selector, and then the selector uses a random forest algorithm to select the optimal algorithm. In the system verification, 120 data sets are used to train the selector; pg. 6 lines 56-59 The optimal algorithm corresponding to the classification problem data set includes 11 algorithms: ridge regression, perceptron, passive attack algorithm, nearest neighbor algorithm, random forest, L 2 penalty term support vector machine, L 2 penalty term logistic regression, L 1 penalty term support vector machine, L 1 penalty term logistic regression, nearest center algorithm and elastic network; pg. 6 lines 14-20 selecting each classification question data set from the UCI machine learning database and the Kagle data set, and processing each classification question data set to obtain corresponding classification meta-knowledge; at the same time, obtaining, by the knowledgebase, an optimal algorithm number corresponding to each classification question data set; selecting an effective feature from the classification meta-knowledge as a meta-feature by using a Bayesian optimization algorithm; forming an alternative training set by using all the meta-features and the corresponding optimal algorithm numbers, and training the meta-knowledge training incubator; obtaining an optimal algorithm number for each meta-feature by the trained meta-knowledge training incubator);
receiving, using one or more processors, a problem type of an input problem (pg. 2 line 44 The classification problem data set input information Records = (the data set number, 21 candidate feature values A 1; pg. 3 line 19 different classification problem data sets; pg. 2 lines 28-29 a training feature selection module, configured to select each classification question data set from the UCI machine learning database and the Kagle data set; pg. 8 lines 1-2 10 data sets selected by the person);
providing one or more problem features of the input problem to the trained and configured solver selection machine learning model (pg. 2 lines 38-40 perform knowledge processing to obtain an algorithm selection training set including a one-to-one correspondence between different classification problem data sets and their corresponding learning algorithms; pg. 5 line 8 21 features related to the data set are extracted during feature selection; pg. 3 lines 25-27 The present invention aims to solve the classification problem belonging to supervised learning in data mining, and can be used to automatically analyze a data set of a specific question, obtain a meta-knowledge feature of the data set, and predict an optimal learning algorithm by analyzing the feature; pg. 5 lines 35-36 train the selector, and then the selector uses a random forest algorithm to select the optimal algorithm. In the system verification, 120 data sets are used to train the selector; pg. 6 lines 14-20 selecting each classification question data set from the UCI machine learning database and the Kagle data set, and processing each classification question data set to obtain corresponding classification meta-knowledge; at the same time, obtaining, by the knowledgebase, an optimal algorithm number corresponding to each classification question data set; selecting an effective feature from the classification meta-knowledge as a meta-feature by using a Bayesian optimization algorithm; forming an alternative training set by using all the meta-features and the corresponding optimal algorithm numbers, and training the meta-knowledge training incubator; obtaining an optimal algorithm number for each meta-feature by the trained meta-knowledge training incubator);
executing, using the one or more processors, the trained and configured solver selection machine learning model to determine the optimal solver type for the input problem from among the plurality of different solver types based at least in part on the one or more problem features of the input problem (pg. 2 lines 38-40 perform knowledge processing to obtain an algorithm selection training set including a one-to-one correspondence between different classification problem data sets and their corresponding learning algorithms; pg. 5 line 8 21 features related to the data set are extracted during feature selection; pg. 6 lines 56-59 The optimal algorithm corresponding to the classification problem data set includes 11 algorithms: ridge regression, perceptron, passive attack algorithm, nearest neighbor algorithm, random forest, L 2 penalty term support vector machine, L 2 penalty term logistic regression, L 1 penalty term support vector machine, L 1 penalty term logistic regression, nearest center algorithm and elastic network; pg. 2 lines 29-30 obtain, by the knowledge base module, an optimal algorithm number corresponding to each classification question data set; pg. 3 lines 25-27 The present invention aims to solve the classification problem belonging to supervised learning in data mining, and can be used to automatically analyze a data set of a specific question, obtain a meta-knowledge feature of the data set, and predict an optimal learning algorithm by analyzing the feature; pg. 6 lines 14-20 selecting each classification question data set from the UCI machine learning database and the Kagle data set, and processing each classification question data set to obtain corresponding classification meta-knowledge; at the same time, obtaining, by the knowledgebase, an optimal algorithm number corresponding to each classification question data set; selecting an effective feature from the classification meta-knowledge as a meta-feature by using a Bayesian optimization algorithm; forming an alternative training set by using all the meta-features and the corresponding optimal algorithm numbers, and training the meta-knowledge training incubator; obtaining an optimal algorithm number for each meta-feature by the trained meta-knowledge training incubator; processing the to-be-processed data set to obtain a to-be-processed meta-feature; analyzing the to-be-processed meta-feature by using a meta-knowledge training incubator to obtain an optimal learning algorithm of the to-be-processed data set);
executing, using the one or more processors, the determined optimal solver type for the input problem to generate a problem output comprising an optimized solution to the input problem (pg. 6 lines 1-2 select an algorithm that is sufficiently excellent for the classification problem to solve the related problems; pg. 3 lines 25-27 The present invention aims to solve the classification problem belonging to supervised learning in data mining, and can be used to automatically analyze a data set of a specific question, obtain a meta-knowledge feature of the data set, and predict an optimal learning algorithm by analyzing the feature).
Wang fails to teach receiving, using one or more processors, a problem type of an input problem originating from a client computing entity; providing, using the one or more processors, the problem output to the client computing entity.
However, Lindawati teaches receiving, using one or more processors, a problem type of an input problem originating from a client computing entity ([0021] The data source contains data or information used by an optimizer 120. In one implementation, the data source contains information of delivery requests, for example, from a plurality of customers. Each delivery request may be associated with inter alia a time window for delivering one or more items to a customer, load size of the delivery, and destination location. Providing other types of information related to the delivery requests may also be useful. Additionally, the data source contains information of one or more depot locations as well as vehicles for servicing the delivery requests. For example, information such as number of vehicles in vehicle fleets, vehicle number, time capacity, volume, weight capacity, demand destination locations, depot locations, delivery time windows, operation hours, delivery load size, may be retrieved from the database of one or more courier companies. Depending on the application, providing other types of data in the data source may also be useful. The type of data collected from the data source may be configured by a user, for example, via the user interface of the client device; [0016] The framework may be used in a logistics problem such as, for example, a delivery problem with narrow time window constraints in supply chain management. For example, the logistics problem may be for optimizing vehicle routing and scheduling for the shipment of items in response to a plurality of delivery requests; [0023] Computer system 106 includes a processor);
providing, using the one or more processors, the problem output to the client computing entity ([0016] The framework solves the delivery problem using the genetic algorithm and outputs the optimal solution; [0026] The optimizer then reports the optimal solution across all populations via the user interface of the client device; [0023] Computer system 106 includes a processor).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang with the teachings of Lindawati so the client can have access to an optimal solution (see Lindawati [0026] The optimizer then reports the optimal solution across all populations via the user interface of the client device).
As per claim 2, Wang and Lindawati teach the computer-implemented method of claim 1. Wang teaches wherein the problem type is a domain-specific problem type and wherein the solver selection machine learning model is configured to determine an optimal domain-specific solver type from among a set of domain-specific solver types (pg. 2 lines 23-24 solving the problem that the selection mode of the learning algorithm involved in the existing data processing does not have universality; pg. 6 lines 56-59 The optimal algorithm corresponding to the classification problem data set includes 11 algorithms: ridge regression, perceptron, passive attack algorithm, nearest neighbor algorithm, random forest, L 2 penalty term support vector machine, L 2 penalty term logistic regression, L 1 penalty term support vector machine, L 1 penalty term logistic regression, nearest center algorithm and elastic network; pg. 5 lines 35-36 train the selector, and then the selector uses a random forest algorithm to select the optimal algorithm. In the system verification, 120 data sets are used to train the selector; pg. 4 lines 44-45 for the 11 alternative classification algorithms in the present invention, each algorithm has respective applicable capabilities).
As per claims 13-14, they are system claims of claims 1-2, so they are rejected for the same reasons as claims 1-2.
As per claim 20, it is a computer program product claim of claim 1, so it is rejected for the same reason as claim 1. Additionally, Wang teaches a computer program product comprising at least one non-transitory computer-readable storage medium having computer-readable program code portions stored therein, which, when executed on one or more processors, causes the one or more processors to perform processes (pg. 2 lines 27-28 The present invention provides a system for automatically selecting a data set classification learning algorithm, comprising: a training feature selection module, configured to select each classification question data set from the UCI machine learning database and the Kagle data set, process each classification question data set to obtain corresponding classification meta-knowledge, and obtain, by the knowledge base module, an optimal algorithm number corresponding to each classification question data set; pg. 3 lines 57-60 The knowledge base module 400 is configured to obtain effective information of each classification algorithm paper, and perform knowledge processing to obtain an algorithm selection training set including a one-to-one correspondence between different classification problem data sets and their corresponding learning algorithms; and store a correspondence between the classification meta-features and the classification problem data set; Since the system contains modules that are processed and store information and the system performs operations automatically, the system contains a computer program product, computer-readable program code, and one or more processors.).
Claims 3, 4, 15, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Wang and Lindawati, as applied to claims 1 and 13 above, in view of Williams et al. ( US 20190332667 A1 hereinafter Williams).
Williams was cited in a previous office action.
As per claim 3, Wang and Lindawati teach the computer-implemented method of claim 1. Wang teaches wherein the problem type of the input problem and the one or more problem features of the input problem are received (pg. 2 line 44 The classification problem data set input information Records = (the data set number, 21 candidate feature values A 1; pg. 3 line 19 different classification problem data sets; pg. 2 lines 28-29 a training feature selection module, configured to select each classification question data set from the UCI machine learning database and the Kagle data set; pg. 8 lines 1-2 10 data sets selected by the person);
Additionally, Lindawati teaches wherein the problem output is provided to the client computing entity ([0016] The framework solves the delivery problem using the genetic algorithm and outputs the optimal solution; [0026] The optimizer then reports the optimal solution across all populations via the user interface of the client device).
Wang and Lindawati fail to teach wherein the problem type of the input problem and the one or more problem features of the input problem are received via a type-agnostic problem solving application programming interface (API) request, and wherein the problem output is provided to the client computing entity via a type-agnostic problem solution API response.
However, Williams teaches wherein the problem type of the input problem and the one or more problem features of the input problem are received via a type-agnostic problem solving application programming interface (API) request, and wherein the problem output is provided to the client computing entity via a type-agnostic problem solution API response ([0050] In API documentation 401 and 402, identifiers such as “GPS_INFO”, “locCoord”, “CoordData,” or “loc” may all refer to parameters having GPS location data types. Accordingly, the NLP machine may be configured to recognize natural language features that may indicate an API-agnostic semantic entity and/or API-agnostic semantic data type associated with an API-specific input and/or output parameter; [0101] In some examples, selecting a path includes scoring a plurality of candidate paths with one or more graph theoretic measures (e.g., path length, network flow). In some examples, selecting a path includes generating one or more candidate paths with a graph theoretic algorithm (e.g., generating a shortest path that visits each of a plurality of function nodes based on an approximative solver for the Travelling Salesman Problem configured to find, with a high probability, such shortest paths); [0109] the previously learned path selection policy may be able to robustly select high-quality paths even for a new API; [0108] In some examples, feedback used to adjust the previously learned path selection policy may include visual feedback from a user, e.g., after a user issues a query, the user may be visually presented with a graphical user interface (GUI) showing one or more decisions made due to the previously learned path selection policy; [0072] For example, function node B212 may represent a function to display a map based in part on an API-agnostic semantic entity 230G associated with map info, and accordingly “getDirections” may be executed to output a map info data value for transmitting to semantic entity 230G, so that semantic entity 230G may provide an API-agnostic semantic data value representing the map info to the function to display the map.).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang and Lindawati with the teachings of Williams to promote efficiency (see Williams [0019] Accordingly, method 100 may enable computer services to be automatically configured to better perform tasks to assist users (e.g., to perform a more diverse variety of tasks, to perform tasks more efficiently, and/or to improve user satisfaction with regard to the assistance provided by the computer system).).
As per claim 4, Wang, Lindawati, and Williams teach the computer-implemented method of claim 3. Williams teaches wherein the type-agnostic problem solving API request comprises a plurality of static fields each configured to describe problem features across different problem types ([0050] In API documentation 401 and 402, identifiers such as “GPS_INFO”, “locCoord”, “CoordData,” or “loc” may all refer to parameters having GPS location data types. Accordingly, the NLP machine may be configured to recognize natural language features that may indicate an API-agnostic semantic entity and/or API-agnostic semantic data type associated with an API-specific input and/or output parameter; [0101] In some examples, selecting a path includes scoring a plurality of candidate paths with one or more graph theoretic measures (e.g., path length, network flow). In some examples, selecting a path includes generating one or more candidate paths with a graph theoretic algorithm (e.g., generating a shortest path that visits each of a plurality of function nodes based on an approximative solver for the Travelling Salesman Problem configured to find, with a high probability, such shortest paths); [0100] Various criteria may be applied to determine quality of a path, e.g., path length, a predicted degree of user satisfaction when the path is used to perform a task responsive to a query, number of disambiguating questions, monetary cost of utilizing a particular API (e.g., a cost associated with a ride service), and/or computational costs of using a particular API (e.g., a latency associated with a particular API function). The various criteria may be balanced by assigning a cost based on each criterion, e.g., in order to select a path having a minimal cost.).
As per claims 15 and 16, they are system claims of claims 3 and 4, so they are rejected for the same reasons as claims 3 and 4.
Claims 5 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Wang and Lindawati, as applied to claims 1 and 13 above, in view of John et al. (US 20210132947 A1 hereinafter John).
John was cited in a previous office action.
As per claim 5, Wang and Lindawati teach the computer-implemented method of claim 1. Wang teaches wherein the problem type of the input problem is received (pg. 2 lines 28-29 select each classification question data set from the UCI machine learning database and the Kagle data set).
Wang and Lindawati fail to teach the input problem is received at a serverless request management engine native to a server cloud infrastructure and corresponding to one of one or more availability zones.
However, John teaches input problem is received at a serverless request management engine native to a server cloud infrastructure and corresponding to one of one or more availability zones (claim 1 receiving, by the computing system and as an output of the machine-learned model, a task execution configuration for the task, wherein the task execution configuration selects for execution of the task one of a number of available serverless cloud service providers; [0073] the machine learning models may comprise those for optimization of sequential decision problems that may be solved using reinforcement learning techniques based on methods including, but not limited to, deep neural networks and random forest-based models. In some embodiments, the machine learning models may also include methods for solving parameter optimization problems such as gaussian processes; [0105] One example aspect of the present disclosure is directed to an artificial intelligence (“AI”)-enabled workflow management system (“WMS”) configured for running workflows comprising tasks and associated rules for their orchestration on serverless execution platforms of one or more cloud service providers).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang and Lindawati with the teachings of John to increase efficiency (see John [0015] The proposed systems and methods have a number of technical effects and benefits. As one example, the proposed systems and methods can enable more efficient performance of serverless workflows.).
As per claim 17, it is a system claim of claim 5, so it is rejected for the same reasons.
Claims 6 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Wang and Lindawati, as applied to claims 1 and 13 above, in view of Mueller et al. (US 20210326717 A1 hereinafter Mueller).
Mueller was cited in a previous office action.
As per claim 6, Wang and Lindawati teach the computer-implemented method of claim 1.
Wang and Lindawati fail to teach wherein the solver selection machine learning model is managed by a serverless container management engine that is native to a server cloud infrastructure.
However, Mueller teaches wherein the solver selection machine learning model is managed by a serverless container management engine that is native to a server cloud infrastructure ([0028] A serverless function itself may be executed by a compute instance, such as a virtual machine, container, etc., when triggered or invoked; [0019] use of an AutoML system to obtain a “best” ML pipeline (which may optionally be just a single ML model) based on the AutoML system exploring multiple pipeline variants and selecting a best resultant pipeline; [0001] The field of machine learning has become widely accepted as a significant driver of the future of technology. Organizations everywhere now seek to use machine learning techniques to address a wide variety of problems, such as optimizing aspects of their products, internal processes, customer experience, etc; [0025] For example, a cloud provider network (or just “cloud”) may refer to a large pool of accessible virtualized computing resources (such as compute, storage, and networking resources, applications, and services); claim 16 wherein the plurality of ML model training jobs run at least partially in parallel in that at least two of the plurality of ML model training jobs are actively trained at a same point in time by at least two different compute instances.).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang and Lindawati with the teachings of Mueller to obtain the best ML model (see Mueller [0019] use of an AutoML system to obtain a “best” ML pipeline (which may optionally be just a single ML model) based on the AutoML system exploring multiple pipeline variants and selecting a best resultant pipeline).
As per claim 18, it is a system claim of claim 6, so it is rejected for the same reasons.
Claims 7 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Wang, Lindawati, and Mueller, as applied to claims 6 and 18 above, in view of Ashrafzadeh (US 20220414547 A1 hereinafter Ashrafzadeh).
Ashrafzadeh was cited in a previous office action.
As per claim 7, Wang, Lindawati, and Mueller teach the computer-implemented method of claim 6. Mueller teaches wherein the serverless container management engine is configured to scale a total count of container instances based at least in part on solver types ([0028] A serverless function itself may be executed by a compute instance, such as a virtual machine, container, etc., when triggered or invoked; [0117] The model hosting system 140 can handle the acquisition and configuration of compute capacity (e.g., containers, instances, etc.) based on demand for the execution of trained machine learning models; [0099] the ML training containers 1030 are logical units created; claim 20 wherein the plurality of ML training jobs include at least partially training a first ML model according to a first ML algorithm type and at least partially training a second ML model according to a second ML algorithm type, wherein the first ML algorithm type is different than the second ML algorithm type.).
Wang, Lindawati, and Mueller fail to teach scale a total count of container instances based at least in part on a total count of solver types.
However, Ashrafzadeh teaches scale a total count of container instances based at least in part on a total count of solver types ([0029] The MLS infrastructure 100 can host any number of serving containers or clusters of serving containers 160A-N. Different clusters can host different versions or types of scoring services or different versions or types of combinations of scoring services 131A-V; [0027] A machine learning application is defined by a flow of operations that includes at least a scoring service; [0032] Although some examples describe the clusters of serving containers that serve one version of a scoring service 131A, one version or more version of scoring services 131B-B, and scoring services 1310-V, any clusters of any serving containers may serve any number of versions of any number of any types of any machine-learning models 175.).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang, Lindawati, and Mueller with the teachings of Ashrafzadeh for efficient deployment (see Ashrafzadeh [0029] Applications that are containerized can be quickly deployed).
As per claim 19, it is a system claim of claim 7, so it is rejected for the same reasons.
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Wang, Lindawati, and Mueller, as applied to claim 6 above, in view of Ashrafzadeh, in view of Kondou et al. (US 20200089728 A1 hereinafter Kondou) and further in view of Erickson et al. (US 20210256610 A1 hereinafter Erickson).
Kondou and Erickson were cited in a previous office action.
As per claim 8, Wang, Lindawati, and Mueller teach the computer-implemented method of claim 6. Mueller teaches the serverless container management engine ([0028] A serverless function itself may be executed by a compute instance, such as a virtual machine, container, etc., when triggered or invoked).
Wang, Lindawati, and Mueller fail to teach wherein an inbound problem queue is updated to identify the input problem, and wherein the container management engine is configured to scale a total count of container instances for the plurality of solver types based at least in part on a number of problems identified by the inbound problem queue.
However, Ashrafzadeh teaches wherein the container management engine is configured to scale a total count of container instances for the plurality of solver types ([0029] The MLS infrastructure 100 can host any number of serving containers or clusters of serving containers 160A-N. Different clusters can host different versions or types of scoring services or different versions or types of combinations of scoring services 131A-V; [0027] A machine learning application is defined by a flow of operations that includes at least a scoring service; [0032] Although some examples describe the clusters of serving containers that serve one version of a scoring service 131A, one version or more version of scoring services 131B-B, and scoring services 1310-V, any clusters of any serving containers may serve any number of versions of any number of any types of any machine-learning models 175.).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang, Lindawati, and Mueller with the teachings of Ashrafzadeh for efficient deployment (see Ashrafzadeh [0029] Applications that are containerized can be quickly deployed).
Wang, Lindawati, Mueller, and Ashrafzadeh fail to teach wherein an inbound problem queue is updated to identify the input problem, and scale a total count of container instances based at least in part on a number of problems identified by the inbound problem queue.
However, Kondou teaches wherein an inbound problem queue is updated to identify the input problem, and a number of problems identified by the inbound problem queue ([0250] the number of problems that remain in a state of processing standby queue; [0326] inputs the problem Q4′ in a queue; [0248] the management portion 1802 may hold the information on the problem during the execution in each of the optimization devices 408 ($j) or on a processing standby queue (queue)).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang, Lindawati, Mueller, and Ashrafzadeh with the teachings of Kondou to reduce waiting time (see Kondou [0431] it is possible to suppress the occurrence of a waiting time and a delay in the start of the arithmetic processing when executing the arithmetic operation of the combinatorial optimization problem.).
Wang, Lindawati, Mueller, Ashrafzadeh, and Kondou fail to teach scale a total count of container instances based at least in part on a number of problems.
However, Erickson teaches scale a total count of container instances based at least in part on a number of problems ([0033] The N sub-problems can be loaded onto N-worker nodes that perform heuristic search optimization (HSO)).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang, Lindawati, Mueller, Ashrafzadeh, and Kondou with the teachings of Erickson to create the best solution (see Erickson [0058] recombines the received best-known feasible sub-solutions (if found) sent by the N-worker nodes 510 into a best-known feasible solution).
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Wang and Lindawati, as applied to claim 1 above, in view of Khalloof et al. (A Generic Distributed Microservices and Container based Framework for Metaheuristic Optimization hereinafter Khalloof).
Khalloof was cited in a previous office action.
As per claim 9, Wang and Lindawati teach the computer-implemented method of claim 1. Wang teaches executing the optimal solver type to generate a problem output (pg. 6 lines 1-2 select an algorithm that is sufficiently excellent for the classification problem to solve the related problems; pg. 3 lines 25-27 The present invention aims to solve the classification problem belonging to supervised learning in data mining, and can be used to automatically analyze a data set of a specific question, obtain a meta-knowledge feature of the data set, and predict an optimal learning algorithm by analyzing the feature).
Wang and Lindawati fail to teach generate the problem output comprises: receiving one or more container outputs generated based at least in part on execution of one or more container instances; and generating the problem output based at least in part on the one or more container outputs.
However, Khalloof teaches generate the problem output comprises: receiving one or more container outputs generated based at least in part on execution of one or more container instances; and generating the problem output based at least in part on the one or more container outputs (pg. 1367 3.2 Execution Workflow paragraph 2 After that, the E.O.-Service asks if a specific model, which can be used for solving the optimization problem, is active. If not, the E.O.-Service requests a list of available models, which is raised to the user interface tier, so that a decision can be made. After selection of the model the E.O.-Service sends the configuration data related to the optimization task as a set of name-value pairs to the Ds.S.-Service. The Ds.S.-Service forwards them to the Co.M.-Service to be applied; pg. 1366 right column Container Management Service (Co.M.-Service). The Co.M.-Service receives the configuration parameters of the simulation model selected by the E.O.-Service to prepare, initialize and in parallel start the required number of containers defined by the Ds.S.-Service; pg. 1366 left column 3.1.1 Container Layer paragraph 2 Each microservice is deployed into one or more Docker containers to perform its task.).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang and Lindawati with the teachings of Khalloof for a parallel and scalable runtime environment (see Khalloof pg. 1364 right column paragraph 2 Highly parallel and scalable runtime environment.).
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Wang, Lindawati, and Khalloof, as applied to claim 9 above, in view of Ribalta et al. (US 20210397943 A1 hereinafter Ribalta).
Ribalta was cited in a previous office action.
As per claim 10, Wang, Lindawati, and Khalloof teach the computer-implemented method of claim 9. Khalloof teaches further comprising: monitoring execution of each container instance during each execution iteration (Fig. 1; pg. 1366 right column paragraph 3 monitors the entire process of solving the actual optimization task; pg. 1364 left column paragraph 1 EAs evaluate each individual and iteratively perform some genetic operators like selection, crossover and mutation to determine the fittest offspring that will be used for the next generation; pg. 1367 3.2 Execution Workflow paragraph 2 After that, the E.O.-Service asks if a specific model, which can be used for solving the optimization problem, is active. If not, the E.O.-Service requests a list of available models, which is raised to the user interface tier, so that a decision can be made. After selection of the model the E.O.-Service sends the configuration data related to the optimization task as a set of name-value pairs to the Ds.S.-Service. The Ds.S.-Service forwards them to the Co.M.-Service to be applied;).
Wang, Lindawati, and Khalloof fail to teach halting the execution of a container instance if a per-iteration optimization gain of the execution iteration fails to satisfy a configurable per-iteration optimization gain threshold.
However, Ribalta teaches halting the execution of a container instance if a per-iteration optimization gain of the execution iteration fails to satisfy a configurable per-iteration optimization gain threshold ([0075] In at least one embodiment, neural network training was run using a 19.12 NVIDIA Docker TensorFlow container, using TensorFlow version 1.15.0. In at least one embodiment, training with multiple GPUs was run using Horovod version 0.18.2. In at least one embodiment, neural network model was trained using an Adam optimizer employing an initial learning rate of 10.sup.−3, with β0=0.9 and β1=0.999. In at least one embodiment, convergence was detected using early stopping, terminating training after five iterations without decrease of validation loss; Ribalta teaches a per-iteration optimization gain of the execution iteration fails to satisfy a configurable per-iteration optimization gain threshold because the instant specification recites that this occurs when there is a convergence.).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang, Lindawati, and Khalloof with the teachings of Ribalta to reduce the number of iterations (see Ribalta [0075] In at least one embodiment, convergence was detected using early stopping, terminating training after five iterations without decrease of validation loss.).
Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Wang, Lindawati, and Khalloof, as applied to claim 9 above, in view of Kondou.
As per claim 11, Wang, Lindawati, and Khalloof teach the computer-implemented method of claim 9. Khalloof teaches wherein execution of a container instance is configured to generate in parallel a container output for each of one or more problems identified (Fig. 2; pg. 1366 left column paragraph 3 Combining this with container runtime automation enables full flexibility in scaling the algorithm on a cluster to fully exploit its potential. The framework can execute as many Ca.-Services instances as required, depending on the size of the input that can be processed in parallel; pg. 1367 3.2 Execution Workflow paragraph 2 After that, the E.O.-Service asks if a specific model, which can be used for solving the optimization problem, is active. If not, the E.O.-Service requests a list of available models, which is raised to the user interface tier, so that a decision can be made. After selection of the model the E.O.-Service sends the configuration data related to the optimization task as a set of name-value pairs to the Ds.S.-Service. The Ds.S.-Service forwards them to the Co.M.-Service to be applied; pg. 1366 left column 3.1.1 Container Layer paragraph 2 Each microservice is deployed into one or more Docker containers to perform its task).
Wang, Lindawati, and Khalloof fail to teach one or more problems identified by an inbound problem queue.
However, Kondou teaches one or more problems identified by an inbound problem queue ([0250] the number of problems that remain in a state of processing standby queue; [0326] inputs the problem Q4′ in a queue; [0248] the management portion 1802 may hold the information on the problem during the execution in each of the optimization devices 408 ($j) or on a processing standby queue (queue)).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang, Lindawati, and Khalloof with the teachings of Kondou to reduce waiting time (see Kondou [0431] it is possible to suppress the occurrence of a waiting time and a delay in the start of the arithmetic processing when executing the arithmetic operation of the combinatorial optimization problem.).
Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Wang and Lindawati, as applied to claim 1 above, in view of Nagase (US 20190340513 A1).
Nagase was cited in a previous office action.
As per claim 12, Wang and Lindawati teach the computer-implemented method of claim 1.
Wang and Lindawati fail to teach wherein the set of solver types comprises at least one of a brute-force solver type, a first-fit solver type, a strongest-fit solver type, a Tabu- search solver type, a simulated-annealing solver type, a late-acceptance solver type, a hill- climbing solver type, or a strategic oscillation solver type.
However, Nagase teaches wherein the set of solver types comprises at least one of a brute-force solver type, a first-fit solver type, a strongest-fit solver type, a Tabu- search solver type, a simulated-annealing solver type, a late-acceptance solver type, a hill- climbing solver type, or a strategic oscillation solver type ([0177] As the heuristic method, for example, a greedy method, a hill climbing method, a simulated annealing method, taboo search, genetic algorithms).
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to have combined Wang and Lindawati with the teachings of Nagase to promote efficiency (see Nagase [0014] The present invention has been made in consideration of the above-mentioned problems, and an object of the invention is to provide an optimal solution determination method, an optimal solution determination program, and an optimal solution determination device capable of efficiently and accurately performing determination of optimality of a solution in a combinatorial optimization problem.).
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 nonprovisional extension fee (37 CFR 1.17(a)) 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 HSING CHUN LIN whose telephone number is (571)272-8522. The examiner can normally be reached Mon - Fri 9AM-5PM.
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, Aimee Li can be reached at (571) 272-4169. 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.
/H.L./Examiner, Art Unit 2195 /Aimee Li/Supervisory Patent Examiner, Art Unit 2195
Prosecution Insights