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 .
Response to Amendment
This Final Rejection is filed in response to Applicant Arguments/Remarks Made in an Amendment filed 09/15/2025.
Claims 1-4, 6, 8, 11-12, 15-16, and 18-19 are amended.
Claims 7, 3-14, and 17 are cancelled.
New Claims 20-25 are added.
Claims 1-6, 8-12, 15--16, and 18-25 remain pending.
Response to Arguments
Argument 1, Applicant argues in Applicant Arguments/Remarks Made in an Amendment filed 09/15/2025 on pg. 9-10, that amendments to the primary claims overcomes the 35 USC 101 Rejection.
Response to Argument 1, Applicant’s arguments, with respect to the 35 USC 101 rejection have been fully considered and are persuasive. The 35 USC 101 of rejections to claims have been withdrawn.
.
Argument 2, Applicant argues in Applicant Arguments/Remarks Made in an Amendment filed 09/15/2025 on pg. 9-10, that Saxena fails to teach the Claim 1 limitations, regarding to “deploy a set of one or more models on one or more compute resources, wherein each model is trained with a respective subset of training data, wherein a first subset of training data for a first model in the set is different from a second subset of training data for a second model in the set… receive, at a unified application programming interface (API), a query… execute the selected model on at least one compute resource by applying trained parameters of the selected model to at least the information on the query to generate outputs”.
Response to Argument 1, Applicant’s arguments have been considered, however in light of the amendments a new ground of rejection (U.S. Patent Application Publication NO. 20200160229 “Atcheson”) is applied to updated rejections.
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) ) 1, 18-19, 23-25 is/are rejected under 35 U.S.C. 102(a)(2) as being anticipated by U.S. Patent Application Publication NO. 20200160229 “Atcheson”.
Claim 1:
Atcheson teaches a system, comprising: one or more processors (i.e. para. [0119], the processing system 2404 is illustrated as including hardware element 2410 that may be configured as processor) configured to: deploy a set of one or more models on one or more compute resources (i.e. para. [0081], “The user interface 702 includes a display of multiple machine learning models, represented by the propensity model 704, the classifier model 706, the image recognition model 708, and the anomaly model 71”, wherein it is noted that the models are stored on compute resources such as a database), wherein each model is trained with a respective subset of training data, wherein a first subset of training data for a first model in the set is different from a second subset of training data for a second model in the set (i.e. para. [0054], Each machine learning model 122 may be associated with a respective data source 124 that is used to train the machine learning model 122); receive, at a unified application programming interface (API), a query (i.e. para. [0076], Fig. 4, “Control 406 is selectable to select one or more models or target outcomes for use in generating the user experience”, wherein the BRI for a query encompasses a user input query for a target outcome); transmit, via the unified API, information on the query to the one or more compute resources (i.e. para. [0106], Fig. 7, User input is then received, specifying at least one target outcome to be generated for the profile information (block 2206). The user experience system 104, for instance, receives user input selecting one of the target outcomes); determine a selected model from the set of one or more models to invoke based on an analysis of the query (i.e. para. [0064], Upon receiving a selection of an available machine learning model or target outcome, the outcome selection module retrieves a corresponding machine learning model 122); execute the selected model on at least one compute resource by applying trained parameters of the selected model to at least the information on the query to generate outputs (i.e. para. [0064], the outcome selection module 114 may identify a machine learning model 122 that outputs a likely vacation destination given input data describing a user's age, geographic location, and hobbies of interest.); and receive, at the unified API, a response from the at least one compute resource generated based at least on the outputs from the selected model (i.e. para. [0065], The outcome selection module 114 communicates the indication of the one or more different machine learning models to the data translation module 116 along with their respective model metadata 202 to inform the data translation module 116 of a data type and format to be supplied to the different machine learning models.); and
a memory coupled to the one or more processors and configured to provide the one or more processors with instructions (i.e. para. [0120], The memory/storage component 2412 may include volatile media (such as random access memory (RAM)) and/or nonvolatile media (such as read only memory (ROM)).
Claim 18:
Claim 18 is the method claim reciting similar limitations to Claim 1 and is rejected for similar reasons.
Claim 19:
Claim 19 is the product claim reciting similar limitations to Claim 1 and is rejected for similar reasons
Claim 23:
Atcheson teaches the system of claim 1, wherein the selected model is determined based on one or more query parameters for the query including one or a combination of a type of prediction, a scope of the prediction, and a temporal bounding condition (i.e. para. [0054], “the outcome selection module 114 is configured to communicate with a database storing machine learning models, such as data base 120, and identify one or more machine learning models 122 that are useable to generate an output … As described herein, each of the machine learning models 122 refers to a model that utilizes algorithms to learn from, and make predictions on,”, wherein the BRI for a query parameter encompasses a user input query for a target outcome associated with a prediction).
Claim 24:
Claim 24 is the method claim reciting similar limitations to Claim 23 and is rejected for similar reasons.
Claim 25:
Claim 25 is the product claim reciting similar limitations to Claim 23 and is rejected for similar reasons.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 2-6, 9, 11, & 20-22 is/are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication NO. 20200160229 “Atcheson” and further in light of U.S. Patent Application Publication NO. 20220180146 “Saxena”.
Claim 2:
Atcheson teaches the system of claim 1.
While Atcheson teaches the concept that some of the deployed models can be optimized (i.t. para. [0072], Generally, a gradient descent model can be implemented as an optimization algorithm designed to find the minimum of a function, and in the illustrated example implementation 300, optimizes for the loss function algorithm 316), Atcheson may not explicitly teach wherein the one or more processors are further configured to:
determine a plurality of optimizer modules with which to optimize the set of the one or more models; and cause the plurality of optimizer modules to respectively perform a respective optimizing process with respect to at least one model of the set of the one or more models.
However, Saxena teaches
determine a plurality of optimizer modules with which to optimize the set of the one or more models (i.e. para. [0092], Fig. 5, “the weighted aggregated single objective 518 is transmitted to a single objective joint optimization module 526 configured to define a ML pipeline 460 to optimize each aggregated single objective”, wherein the BRI for a plurality of optimizer modules encompasses that a plurality of optimization solutions may be determined based on each objective to be optimized based things such as user-specified object weights of the collected one or more collected models); and cause the plurality of optimizer modules to respectively perform a respective optimizing process with respect to at least one model of the set of the one or more models (i.e. para. [0076], “dominance is used to determine the quality of the solutions where a first solution is said to dominate a second solution if the first solution is better than or at least equal to the second solution in all objectives, and the first solution is strictly better than the second solution in at least one objective. Those solutions which are not dominated by any of the other solutions in light of all of the objectives are referred to as a Pareto-optimal solutions through Pareto optimization, i.e., Pareto-optimal solutions are non-dominated solutions and no other solution dominates them”, wherein the respective solution module is selected based on the objective selected for the model and in part by the dominance of the solution or if the solution is determined to be the best or Pareto-optimal).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to add models; and cause the plurality of optimizer modules to respectively perform a respective optimizing process with respect to at least one model of the set of the one or more models, to Atcheson’s library of pre-trained models, with how a plurality of optimizer solutions may be applied to each respective algorithm, as taught by Saxena. One would have been motivated to combine Saxena with Atcheson and would have had a reasonable expectation of success as the combination improves robustness of computing functions, computing efficiency, time to generate a prediction as each model may have a tailor fit optimizer.
Claim 3:
Atcheson teaches the system of claim 1.
However, Atcheson may not explicitly teach
wherein the set of the one or more models are optimized according to a predetermined time interval
Saxena further teaches
wherein the set of the one or more models are optimized according to a predetermined time interval (i.e. para. [0096], “in some embodiments, the number of iterations N is bounded by the allotted processing time and/or the compute budget to generate N Pareto-optimal solutions 602 to define the Pareto-front 604”, wherein the BRI for a predetermined time interval encompasses how models may be optimized according to an allotted processing time or budget).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to add wherein the set of the one or more models are optimized according to a predetermined time interval, to Atcheson’s library of pre-trained models, with wherein the set of the one or more models are optimized according to a predetermined time interval, as taught by Saxena. One would have been motivated to combine Saxena with Atcheson and would have had a reasonable expectation of success as the combination improves robustness of computing functions, computing efficiency, time to generate a prediction as each model may have a tailor fit optimizer.
Claim 4:
Atcheson and Saxena teach the system of claim 2,
Saxena further teaches wherein optimizing the set of the one or more models comprises:
obtaining a respective starting set of parameters with which to optimize the set of the one or more models (i.e. para. [0072], “F(x) is an expression of one or more objective values extending over a domain of x, and the optimization problem Min F(x) is solved through an algorithm that is configured to determine the solutions yielding minimum values for each f(x), where each f(x) is an objective function, and each x represents an optimization parameter”, wherein the BRI for a starting set of parameters encompasses the set of objectives for the respective models that are to be optimized); and running, by each of the plurality of optimizer modules, the respective optimizing process in connection with optimizing, based at least in part on the respective starting set of parameters, a particular model of the set of one or more models to optimize (i.e. para. [0072], each Pareto-optimal solution is a decision vector which optimizes F(x), and the decision vectors x are input vectors and F(x) is the output vector. In the case of the multi-objective ML problem that is to be solved, and as described further herein, each decision vector x denotes a ML pipeline determined by the choice of data transformers and ML model together with their hyperparameters).
Claim 5:
Atcheson and Saxena teach the system of claim 4.
Saxena further teaches wherein: at least two of the plurality of optimizer modules optimize a same model selected from among the plurality of respective models to be optimized (i.e. para. [0081], the Pareto-front includes the plurality of objective values F(x), i.e., the optimized numerical values associated with each point on the Pareto-front resulting from optimizing the respective aggregated single objective. For example, for the set of x, i.e., {x.sup.1, x.sup.2, . . . x.sup.k} that denotes the set of Pareto-optimal solutions, then the Pareto-front includes the set of objective values on these Pareto-optimal solutions); and the at least two of the plurality of optimizer modules optimize the same model using a different starting set of parameters (i.e. para. [0079], “the weights facilitate transforming the multi-objective problem to a single-objective formulation by performing a weighted aggregation of multiple objectives into a combined/aggregated single objective, thereby facilitating alignment of the transformers, hyperparameters, and models in the ML pipeline”, wherein the plurality of objectives to be optimized may be combined into one single objective that may be used to optimize a model).
Claim 6:
Atcheson and Saxena teach the system of claim 5.
Saxena further teaches wherein the at least two of the plurality of optimizer modules optimize parameters of the first model and parameters of the second model in parallel (i.e. para. [0069], Fig. 5, “for example, two blocks shown in succession may, in fact, be accomplished as one step, executed concurrently, substantially concurrently, in a partially or wholly temporally overlapping manner”, wherein the combined objective solutions into a single joint objective optimization module may concurrently optimize a model”, wherein N optimization parameters may be optimized). .
Claim 9:
Atcheson and Saxena teach the system of claim 4.
Saxena further teaches
wherein running the respective optimizing process includes performing at least a number of iterations of training the particular model of the plurality of respective models to be optimized (i.e. para. [0081], A selectable number of iterations are performed such that a Pareto-front is generated… the weights on the next iterative run may be adjusted either through the uniform weights, through adaptive weighting as a function of the previous objective values, or through user-selected weights).
Claim 11:
Atcheson and Saxena teach the system of claim 2.
Saxena further teaches wherein determining the plurality of optimizer modules comprises: determining a set of available threads (i.e. para. [0027], Resource pooling: the provider's computing resources are pooled to serve multiple consumers using a multi-tenant model, with different physical and virtual resources dynamically assigned and reassigned according to demand); and creating a pool of threads with which to implement the plurality of optimizer modules (i.e. para. [0023], cloud computing is a model of service delivery for enabling convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, network bandwidth, servers, processing, memory, storage, applications, virtual machines, and services).
Claim 20:
Atcheson teaches the system of claim 1.
(i.e. para. [0054], “Each machine learning model 122 may be associated with a respective data source 124 that is used to train the machine learning model 122”, wherein different models may have different and respective data source training parameters).
While Atcheson teaches that the models may run on compute resources, Atcheson may not explicitly teach
wherein the one or more compute resources are one or more virtual machines (VM).
However, Saxena further teaches
wherein the one or more compute resources are one or more virtual machines (VM) (i.e. para. [0023], “Cloud computing is a model of service delivery for enabling convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, network bandwidth, servers, processing, memory, storage, applications, virtual machines, and services)”, wherein it is noted that the plurality of ML algorithms may be deployed on computing resources such as virtual machines).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to add wherein the one or more compute resources are one or more virtual machines (VM), to Atcheson’s stored machine learning models, with how machine learning algorithms may be enabled using virtual machines in the cloud, as taught by Saxena. One would have been motivated to combine Saxena and Atcheson because deploying resources using cloud computing that can be rapidly provisioned and released with minimal management effort or interaction with a provider of the service.
Claim 21:
Claim 21 is the method claim reciting similar limitations to Claim 20 and is rejected for similar reasons.
Claim 22:
Claim 22 is the product claim reciting similar limitations to Claim 20 and is rejected for similar reasons.
Claim(s) 8, & 12 is/are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication NO. 20200160229 “Atcheson” and further in light of U.S. Patent Application Publication NO. 20220180146 “Saxena”, as applied to Claim 2 above, and further in light of U.S. Patent NO. 9569745 “Ananthanarayanan”.
Claim 8:
Atcheson and Saxena teach the system of claim 2.
While Saxena teaches a plurality of optimizer modules, Saxena may not explicitly teach
wherein each of the plurality of optimizer modules is run on a different thread or compute node.
However, Ananthanarayanan teaches
wherein each of the plurality of optimizer modules is run on a is different thread or compute node (i.e. Col. 9, lines 45-55, multiple optimization techniques may be run in parallel, such as by separate processing devices or on separate threads of a single processing device).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to add wherein each of the plurality of optimizer modules is run on a is different thread or compute node, to Atcheson-Saxena’s machine learning optimization programs, with how different optimization techniques may be run in parallel on separate threads, as taught by Ananthanarayanan. One would have been motivated to combine Ananthanarayanan and Atcheson-Saxena as specializing threads in multithreading enhances performance and resource utilization by assigning specific tasks or functionalities to each thread, allowing for efficient execution and improved responsiveness.
Claim 12:
Atcheson and Saxena teach the system of claim 11.
Saxena further teaches wherein: (i.e. para. [0093], “hyperparameter optimization or hyperparameter tuning is associated with the problem of choosing a set of optimal hyperparameters for a learning algorithm… These measures may be the hyperparameters, and the hyperparameters have to be tuned so that the model can optimally solve the machine learning problem presented thereto. Hyperparameter optimization finds a tuple of hyperparameters that yields an optimal model which minimizes a predefined loss function on given independent data”, wherein the BRI for an optimized model being selected encompasses how a tuple of tuned hyperparameters are found and used to yield an optimal model over iterations).
While Saxena teaches a plurality of optimizer modules, Saxena may not explicitly teach that
the plurality of optimizer modules are respectively implemented on different threads from the pool of threads
However, Ananthanarayanan teaches that
the plurality of optimizer modules are respectively implemented on different threads from the pool of threads (i.e. Col. 9, lines 45-55, multiple optimization techniques may be run in parallel, such as by separate processing devices or on separate threads of a single processing device).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to add wherein the plurality of optimizer modules are respectively implemented on different threads from the pool of threads, to Atcheson-Saxena’s machine learning optimization programs, with how different optimization techniques may be run in parallel on separate threads, as taught by Ananthanarayanan. One would have been motivated to combine Ananthanarayanan and Atcheson-Saxena as specializing threads in multithreading enhances performance and resource utilization by assigning specific tasks or functionalities to each thread, allowing for efficient execution and improved responsiveness.
Claim(s) 10, 15-16 is/are rejected under 35 U.S.C. 103 as being unpatentable over .S. Patent Application Publication NO. 20200160229 “Atcheson” and further in light of U.S. Patent Application Publication NO. 20220180146 “Saxena”, as applied to Claim 4 above, and further in light of U.S. Patent Application Publication NO. 20210232920 “Parangi”.
Claim 10:
Atcheson and Saxena teach the system of claim 4.
Saxena may not explicitly teach
wherein a training dataset used in connection with optimizing the particular model is stored in cache while a particular optimizer module runs the respective optimization process.
However, Parangi teaches
wherein a training dataset used in connection with optimizing the particular model is stored in cache while a particular optimizer module runs the respective optimization process (i.e. para. [0031, 0075], Figs 1, 4C, “the system not only optimizes a specific model but also optimizes how that model is generated, and potentially that feedback process as well, and so on); training, by the machine learning engine, the machine learning model using the second training data set … a cache memory 440 in communication with the central processing unit 421”, wherein it is noted that the optimized machine learning model and a training data set may reside in the cache memory).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to add wherein a training dataset used in connection with optimizing the particular model is stored in cache while a particular optimizer module runs the respective optimization process, to Atcheson-Saxena’s machine learning plurality of optimization programs that optimize a plurality of models, with how an optimized model and its training data may be stored in cache memory, as taught by Parangi. One would have been motivated to combine Parangi and Atcheson-Saxena because caching model data during training significantly speeds up the process by storing frequently accessed data in a temporary location, reducing the need for repeated computations and improving overall efficiency.
Claim 15:
Atcheson and Saxena teach the system of claim 2.
Saxena may not explicitly teach
wherein the at least one of the one or more models is stored in cache during optimization with respect to at least one model.
However, Parangi teaches
wherein the at least one of the one or more models is stored in cache during optimization with respect to at least one model (i.e. para. [0031, 0075], Figs 1, 4C, “the system not only optimizes a specific model but also optimizes how that model is generated, and potentially that feedback process as well, and so on); training, by the machine learning engine, the machine learning model using the second training data set … a cache memory 440 in communication with the central processing unit 421”, wherein it is noted that the optimized machine learning model and a training data set may reside in the cache memory).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to add wherein the at least one of the one or more models is stored in cache during optimization with respect to at least one model, to Atcheson-Saxena’s machine learning plurality of optimization programs that optimize a plurality of models, with how an optimized model and its training data may be stored in cache memory, as taught by Parangi. One would have been motivated to combine Parangi and Atcheson-Saxena because caching model data during training significantly speeds up the process by storing frequently accessed data in a temporary location, reducing the need for repeated computations and improving overall efficiency.
Claim 16:
Atcheson and Saxena teach the system of claim 2.
Saxena may not explicitly teach
wherein at least a subset of the set of the one or more models are obtained and stored in a cache in response to determining to optimize the subset of the set of the one or more models.
However, Parangi teaches
wherein at least a subset of the set of the one or more models are obtained and stored in a cache in response to determining to optimize the subset of the set of the one or more models (i.e. para. [0017, 0031, 0075], Figs 1, 4C, “The system 100 may include a plurality of machine learning models 107a-n… the system not only optimizes a specific model but also optimizes how that model is generated, and potentially that feedback process as well, and so on); training, by the machine learning engine, the machine learning model using the second training data set … a cache memory 440 in communication with the central processing unit 421”, wherein it is noted that the optimized machine learning model and a training data set may reside in the cache memory).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to add wherein at least a subset of the set of the one or more models are obtained and stored in a cache in response to determining to optimize the subset of the set of the one or more models, to Atcheson-Saxena’s machine learning plurality of optimization programs that optimize a plurality of models, with how a subset of machine learning models and its training data may be stored in cache memory in response to be specified for optimization, as taught by Parangi. One would have been motivated to combine Parangi and Atcheson-Saxena because caching model data during training significantly speeds up the process by storing frequently accessed data in a temporary location, reducing the need for repeated computations and improving overall efficiency.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
U.S. Patent Application Publication NO. 20220156642 “Schmidt” teaches in para. [0003], selecting the appropriate machine learning algorithm from a set of algorithms (e.g. Support Vector Machine, SVM), configuring the algorithm's hyperparameters (e.g. the SVM kernel to use), possibly also selecting and configuring input data preprocessing steps (e.g. k-means clustering, selecting k) and selecting and configuring optimizers that “train” the machine learning algorithm.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 DAVID H TAN whose telephone number is (571)272-7433. 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, Cesar Paula can be reached at (571) 272-4128. 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.
/D.T./Examiner, Art Unit 2145
/CESAR B PAULA/Supervisory Patent Examiner, Art Unit 2145