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 Arguments
Applicant's arguments filed September 18, 2025 have been fully considered but they are not persuasive.
On page 8 of the response, applicant argues: “The independent claims require, at minimum, defining a federated search process with N hyperparameters selected for estimation and the remaining hyperparameters selected for local computation at each of a plurality of federated systems” and “it is not practical for a person to mentally evaluate federated data in the context claimed.”
Examiner’s response: It is noted that the independent claims recite a distributed search process across a plurality of parties where each member of the party retains federated data. In the Non-Final office action, examiner analyzed the distributed search process as an additional element of transmitting the search process across a plurality of parties. This limitation was treated as an additional element and not an abstract idea. As to applicant’s assertion that it is not practical for a person to mentally evaluate federated data, examiner notes that this feature is directed to mere instruction to apply an abstract idea of data evaluation on a generic computer and federated search is a well understood and conventional search process (see US 20080319943, paragraphs [0004]-[0007]).
In addition, applicant argues that “[b]uilding an agent capable of federated automatic machine learning in the manner claimed offers on its face a technical improvement over currently available machine learning technologies.”
Examiner’s response: Examiner respectfully disagrees and notes that the recitation of an agent to define and generate pipeline definition for a machine learning is interpreted as mere instructions to apply the abstract idea of defining a machine learning pipeline definition on a generic computer. The claims do not recite any specific implementation of the agent.
On page 9, applicant argues that the teaching of regression-based extrapolation in Schmidt does not teach the amended claim feature of a search process to include a selection of the N most difficult hyperparameters having the N largest datasets for estimation.
Examiner’s response: It is noted that the amended claim limitations raised 35 USC 112(b) rejections (see below). For the purposes of claim interpretation, the limitation is interpreted as selecting the difficult hyperparameters for estimation because the value of the variable N is neither defined in the claims nor the specification. It is also noted that the criteria for determining difficult hyperparameters is not clear, see 35 USC 112(b) rejection below. Finally, new prior art, “Hyperparameter Importance Across Datasets” (Rijn et al.) is introduced to address some of the added limitations, see update prior art rejection below.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claims 1, 10, and 16 recite “wherein the model configuration comprises a plurality of hyperparameters, and wherein the search process further comprises a selection of the N most difficult hyperparameters having the N largest datasets for estimation”.
It is not clear as to the relationship between difficult hyperparameter and dataset size, and it is not clear what the largest datasets are referring to. The specification at [0022] provides an example of N most difficult hyper-parameters as largest datasets, etc. Hyperparameters are configuration values and it is not clear when a hyperparameter is the largest dataset.
Additionally, it is not clear what aspect of the hyper-parameter is being estimated in the claims. Are the values of the hyper-parameters being estimated? Based on applicant's arguments and reference to [0022] of applicant's specification, it appears the intention is to estimate the score of the hyper-parameter. However, this is not recited in the independent claims. Although, dependent claims 6 recites estimating one or more additional score, it also does not require scores to be associated with a hyper-parameter.
Dependent claims 2-9, 11-15, and 17-20 are also rejected because they fail to cure the deficiencies of their respective independent claims.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
MPEP 2106 (III) sets out steps for evaluating whether a claim is drawn to patent-eligible subject matter. The analysis of claims 1-20, in accordance with these steps, follows.
Step 1 Analysis:
Claims 1 – 9 are directed to a method, claims 10 – 15 are directed to a system comprising of hardware processor and memory, claims 16-20 are directed to computer program products. Therefore, claims 1-20 fall into one of four statutory categories (i.e., process, machine, article of manufacture).
As to claim 1,
Step 2A Prong 1: this claim recites the following abstract ideas:
defining,…, a search process comprising a model configuration for building an automatic machine learning pipeline definition (defining a search process that includes a model configuration is interpreted as a mental process that includes determining a model setting), wherein the model configuration comprises a plurality of hyperparameters, and wherein the search process further comprises a selection of the N most difficult hyperparameters having the N largest datasets for estimation (this limitation further describes the data being searched and the condition for selecting hyperparameters to estimate. These steps are also interpreted as mental processes. For example, a human can select N hyperparameters to estimate.)
aggregating,…, the received evaluation results to define aggregated parameters (this limitation is interpreted as a mental process of evaluating received results to determine aggregated parameters);
generating,…, a new pipeline definition from the aggregated parameters (using the aggregated parameter to create a pipeline definition is a mental process); and
aggregating,…, trained local models received from each member of the plurality of parties to define an aggregated model, wherein each trained local model comprises the new pipeline definition (using received pipeline definition from the trained local models to define an aggregated model is a mental process).
Step 2A Prong 2 and 2B: this claim recites the following additional elements:
the recitation of an automatic machine learning agent, search parameters aggregator, and a model aggregator are directed to mere instructions to apply the abstract idea (as discussed in Prong 1) on a generic computer. The additional limitation does not integrate the judicial exception into a practical application and does not amount to significantly more than the judicial exception. See MPEP 2106.05(f).
distributing, by a search parameters aggregator, the search process across a plurality of parties, wherein each member of the plurality of parties retains federated data comprising training data and holdout data, the search parameters aggregator configured to equally distribute search parameters and their respective values to the plurality of parties for local calculation (This limitation covers transmitting (i.e. distributing) the search process across a plurality of devices by equally transmitting the search parameters and storing federated data on the plurality of devices. See MPEP 2106.05(d)(II)(i) and (iv) which describes data transmission and storing information in memory as well-understood, routine, conventional activity. The additional limitation does not integrate the judicial exception into a practical application and does not amount to significantly more than the judicial exception. Federated search is a well understood and conventional search process (see US 20080319943, paragraphs [0004]-[0007])).
receiving, by the search parameters aggregator and from each member of the plurality of parties, an evaluation result of the model configuration against respective holdout data (This limitation covers receiving evaluation result from a plurality of devices. See MPEP 2106.05(d)(II)(i), which describes data transmission as well-understood, routine, conventional activity. The additional limitations, either individually or in an ordered combination, do not integrate the judicial exception into a practical application and does not amount to significantly more than the judicial exception).
As to claim 10, this is directed to a system claim that corresponds to method claim 1. See the rejection for claim 1 above, which also applies to claim 10. In addition, claim 10 recites the additional limitation of a system having a memory, computer readable instructions, and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations. This limitation is directed to mere instruction to apply the abstract idea on a generic computer to process the data elements, see MPEP 2106.05(f). The additional limitations, either individually or in an ordered combination, do not integrate the judicial exception into a practical application and does not amount to significantly more than the judicial exception).
As to claim 16, this is directed to a program product claim that corresponds to method claim 1. See the rejection for claim 1 above, which also applies to claim 16. In addition, claim 16 recites the additional limitation of a computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by one or more processors to cause the one or more processors to perform operations. This limitation is directed to mere instruction to apply the abstract idea on a generic computer to process the data elements, see MPEP 2106.05(f). The additional limitations, either individually or in an ordered combination, do not integrate the judicial exception into a practical application and does not amount to significantly more than the judicial exception).
As to claims 2, 11 and 17, the limitation “wherein the search process is directed to at least one of estimator selection, hyper-parameter optimization, feature engineering, and hyper-parameter optimization over new features” further limits the search process. This limitation is directed to various algorithms for performing the search process; thus, it is interpreted as a mental process. Claim 2 does not recite any additional elements.
As to claim 3, the limitation “wherein the automatic machine learning agent provides a current source code and model definition to the search parameters aggregator” further describes the data that is transmitted to the search parameter aggregator. This limitation is also directed to data transmission and does not integrate the judicial exception into a practical application and does not amount to significantly more than the judicial exception. See MPEP 2106.05(d)(II)(i), which describes data transmission as well-understood, routine, conventional activity. The additional limitations, either individually or in an ordered combination, do not integrate the judicial exception into a practical application and does not amount to significantly more than the judicial exception).
As to claims 4, 12 and 18, the limitation “wherein the evaluation result comprises a search sub-space parameter and a calculated score for the search sub-space parameter” further describes the content in the evaluation result. A person can evaluate the sub-space parameter and corresponding calculated score. Therefore, the limitation is directed to a mental process and does not recite any additional elements.
As to claims 5, 13 and 19, the limitation “wherein the calculated score comprises one or more of a machine learning score for an evaluated model, a runtime score comprising a training time and an evaluation time, a resource score describing available resources, a utilization score describing resource use, and a meta-data score describing a local data set used for model training” further defines the calculated score and is interpreted as a mental process. The claim does not recite any additional elements.
As to claims 6, 14 and 20, the limitation “estimating, by the search parameters aggregator, one or more additional scores using a regression of the calculated scores” is directed to a mental process of determining one or more additional scores. The claim does not recite any additional elements.
As to claims 7 and 15, the limitation “wherein new pipeline definitions are iteratively generated until all search spaces comprising all possible parameters are explored” is directed to performing repetitive calculation, which is a well-understood, routine, conventional activity, see MPEP 2106.05(d)(II)(ii)). The additional limitations, either individually or in an ordered combination, do not integrate the judicial exception into a practical application and does not amount to significantly more than the judicial exception).
As to claim 8, the limitation “wherein the aggregated model is distributed to each member of the plurality of parties for holdout evaluation” describes the intended result (i.e. holdout evaluation) of distributing the aggregated model; therefore, this limitation is directed to data transmission, which is a well-understood, routine, conventional activity, see MPEP 2106.05(d)(II)(i)). The additional limitations, either individually or in an ordered combination, do not integrate the judicial exception into a practical application and does not amount to significantly more than the judicial exception).
As to claim 9, the limitation “wherein holdout evaluation results are aggregated into the aggregated model” is directed to the mental process of using the evaluation results to update the aggregated model. The claim does not recite any additional elements.
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 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over US PGPUB No. 20240346326 A1 (hereinafter Tyou) in view of US PGPUB No. 20210224585 A1 (hereinafter Schmidt) and further in view of “Hyperparameter Importance Across Datasets” (published in 2018, hereinafter Rijn).
As to claim 1, Tyou teaches a computer-implemented method comprising (see [0001], “federated learning incorporating an automated machine learning (AutoML) method”):
defining, by an automatic machine learning agent, a search process for building an automatic machine learning pipeline definition (see [0049], “The search range determination unit 130 of the learning server apparatus 100 determines a NAS search range (S130)” and [0063], “determines whether or not an optimal neural network can be selected from A pieces of S.sub.r according to the user setting”);
distributing, by a search parameters aggregator, the search process across a plurality of parties (see [0049], “transmits information indicating the NAS search range (hereinafter also referred to as search range information P) to each of the n processing apparatuses 200-i via the transmission unit 110”), wherein each member of the plurality of parties retains federated data comprising training data and holdout data (see [0053], “The score calculation unit 240 of the processing apparatus 200-i extracts the local data d.sub.i stored in the local data storage unit 275. Note that the local data d.sub.i is different for each processing apparatus 200-i, and is data that is an input of the neural network” and “…the number D.sub.i of pieces of data of the local data d.sub.i”; Note. The number of D.sub.i of pieces of data of the local data corresponds to the holdout data), the search parameters aggregator configured to equally distribute search parameters and their respective values to the plurality of parties for local calculation (see [0049], “The search range determination unit 130 of the learning server apparatus 100 determines a NAS search range (S130), and transmits information indicating the NAS search range (hereinafter also referred to as search range information P) to each of the n processing apparatuses 200-i via the transmission unit 110. Note that the search range information P is information common to the n processing apparatuses 200-i.”);
receiving, by the search parameters aggregator and from each member of the plurality of parties, an evaluation result of the model configuration against respective holdout data
(see [0053], “The score calculation unit 240 transmits the combination of the index r of the neural network and the correlation score s.sub.ir thereof and the number D.sub.i of pieces of data of the local data d.sub.i to the learning server apparatus 100 via the transmission unit 210” and see [0054], “The aggregation unit 140 of the learning server apparatus 100 receives A combinations of indexes r and correlation scores s.sub.ir and one number D.sub.i of pieces of data from one processing apparatus 200-i via the reception unit 120. Since the aggregation unit 140 receives data from the n processing apparatuses 200-i, a total of A×n combinations of indexes r and correlation scores s.sub.ir and n number D.sub.i of pieces of data are received”); aggregating, by the search parameters aggregator, the received evaluation results to define aggregated parameters (see [0055], “The aggregation unit 140 aggregates A neural networks by using A×n correlation scores s.sub.ir and n number D.sub.i of pieces of data (S140), and selects an optimal neural network. The optimal neural network here is a neural network that enables highly accurate learning”);
generating, by the automatic machine learning agent, a new pipeline definition from the aggregated parameters (see [0066], “The selected network confirmation unit 150 and the selected network confirmation units 250 of the (n/Q) processing apparatuses 200-i belonging to one group cooperate to perform the federated learning using one neural network as the first global model and perform the federated learning using a total of Q neural networks as the first global model. Note that the selected or updated global model used in the federated learning is stored in the global model storage unit 180”; Note, the updated global model is the generated new pipeline definition); and
aggregating, by a model aggregator, trained local models received from each member of the plurality of parties to define an aggregated model, wherein each trained local model comprises the new pipeline definition (see [0071], “The federated learning unit 160 receives the index of the optimal neural network, and performs normal federated learning in cooperation with the federated learning unit 260 of the processing apparatus 200-i using the neural network corresponding to the index as the first global model (S160)”).
Although the Tyou reference teaches defining a search process for building an automatic machine learning pipeline definition using user setting, it does not specifically disclose “defining, by an automatic machine learning agent, a search process comprising a model configuration for building an automatic machine learning pipeline definition.”
However, Schmidt teaches a method for automatically selecting a machine learning algorithm and tuning hyperparameters of the machine learning algorithm (see Abstract) and defining, by an automatic machine learning agent, a search process comprising a model configuration for building an automatic machine learning pipeline definition (see [0056]-[0065] and [0092], “In an embodiment, different arm container images exist for different frameworks and are instantiated based on the HAMLET task specification. In this embodiment, the MasterBandit has to indicate the exact arm container type in its request to the docker server. The MasterBandit may also pass framework-specific configuration information pertaining to the HAMLET task (e.g., hyperparameter ranges to use or algorithms to consider)”), where the model configuration comprises a plurality of hyperparameters (see [0050], “automates algorithm selection and hyperparameter tuning for a very wide range of algorithms (by integration of different frameworks), for different types of ML tasks”).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the invention to modify the invention of Tyou to include model configuring for building an automatic machine learning pipeline definition as taught by Schmidt because this allows the system to target hyperparameter optimization of Deep Neural Networks (DNN) using a probabilistic model to extrapolate the performance from the first part of a learning curve for a hyperparameter configuration (see [0147] of Schmidt).
Tyou as modified by Schmidt does not teach “the search process further comprises a selection of the N most difficult hyperparameters having the N largest datasets for estimation”.
However, Rijn teaches automatically selecting a machine learning algorithm and tuning hyperparameters of the machine learning algorithm (see abstract) and
a search process further comprises a selection of the N most difficult hyperparameters (p. 2370, Section 4.2, “we used the top n configurations observed for each of the datasets”) having the N largest datasets (p. 2369, Section 4) for estimation (p. 2370, Section 4.2, “We aim to build priors based on the performance data observed across datasets…For each dataset, we propose to run two versions of this optimization procedure: one sampling uniformly from the hyperparameter space and one sampling from the obtained priors. If the priors are indeed useful and generalize across datasets, the optimizer that uses them should obtain better results on the majority of the datasets”).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the invention to further modify the invention of Tyou and Schmidt to includes the teaches of Rijn because the use of priors to perform hyper-parameter selection lead to statistically significant improvements in hyperparameter optimization.
As to claim 2, Tyou as modified by Schmidt teaches wherein the search process is directed to at least one of estimator selection, hyper-parameter optimization, feature engineering, and hyper-parameter optimization over new features (see [0147] of Schmidt, “hyperparameter parameterization”; [0050] of Schmidt, “automates algorithm selection”; [0031] of Schmidt, “parameter tuning logics”).
As to claim 3, Tyou as modified by Schmidt teaches wherein the automatic machine learning agent provides a current source code and model definition to the search parameters aggregator (see Schmidt “[0056] The dispatcher component is the point of contact for the HAMLET user to request the HAMLET service. Therefore, the dispatcher's interface A may be used by the user to: [0057] upload a dataset and define a dataset description, [0058] provide a machine learning task description (e.g. performance function, budget, possibly which AutoML frameworks and configurations to use)” and [0074] of Schmidt, “In a particular embodiment, the interface A for the user to specify the task, upload the dataset, and specify the description of the dataset is a representational state transfer (REST)-based interface”).
As to claim 4, Tyou teaches the evaluation result comprises a search sub-space parameter and a calculated score for the search sub-space parameter (see [0055], “The aggregation unit 140 aggregates A neural networks by using A×n correlation scores s.sub.ir and n number D.sub.i of pieces of data (S140), and selects an optimal neural network. The optimal neural network here is a neural network that enables highly accurate learning”).
As to claim 5, Tyou as modified by Schmidt teaches the calculated score comprises one or more of a machine learning score for an evaluated model, a runtime score comprising a training time and an evaluation time, a resource score describing available resources, a utilization score describing resource use, and a meta-data score describing a local data set used for model training (see [0146] of Schmidt, “The term “learning curve” is used to describe (1) the performance of an iterative machine learning algorithm as a function of its training time or number of iterations and (2) the performance of a machine learning algorithm as a function of the size of the dataset it has available for training” and [0130] of Schmidt, “extrapolating their performance for efficient computation resource usage by time multiplexing, which avoids the drawback of multi-armed bandit algorithms which base their calculations on statistics of past performances”).
As to claim 6, Tyou as modified by Schmidt teaches estimating, by the search parameters aggregator, one or more additional scores using a regression of the calculated scores (see [0149] of Schmidt, “A regression-based extrapolation model can be used for the extrapolation of learning curves to speed up hyperparameter optimization”).
As to claim 7, Tyou as modified by Schmidt teaches new pipeline definitions are iteratively generated until all search spaces comprising all possible parameters are explored (see [0115] of Schmidt, “In a beneficial embodiment, HAMLET also offers to build ensembles based on all or a subset of the task's trained models” and [0140] of Schmidt, “If a non-greedy action is selected, it is called exploration, as it allows to improve the estimate of the non-greedy action's value. Exploration is needed because there is always uncertainty about the accuracy of the action-value estimates. The greedy actions are those that look best at present, but some of the other actions may actually be better”).
As to claim 8, Tyou as modified by Schmidt teaches the aggregated model is distributed to each member of the plurality of parties for holdout evaluation (see [0108] of Schmidt, “In a beneficial embodiment, HAMLET tracks and stores models' wallclock execution times on a certain portion of the dataset (e.g. the test set commonly used to calculate the performance scores) in addition to the models' performance scores. This beneficially allows later to present the user not only the highest performing models, but also the fastest to execute. Also, this may inform ensemble building as indicated below”).
As to claim 9, Tyou as modified by Schmidt teaches holdout evaluation results are aggregated into the aggregated model (see [0144] of Schmidt, “Auto Tune Models (ATM) is a distributed, collaborative, scalable AutoML system, which incorporates algorithm selection and hyperparameter tuning…HAMLET uses a novel bandit algorithm which fits a simple model of the learning curve to observed rewards, but selects the action based on an extrapolation of the learning curve to find the highest possible reward given a time budget. Third, ATM and the Machine Learning Bazaar update the action value statistics based on completed function evaluations, i.e., a base learner's test performance after training it on the dataset”).
As to claim 10, this is directed to a system claim that corresponds to method claim 1. See the rejection for claim 1 above, which also applies to claim 10. In addition, Tyou teaches a system having a memory, computer readable instructions, and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations (see [0046]).
As to claim 16, this is directed to a program product claim that corresponds to method claim 1. See the rejection for claim 1 above, which also applies to claim 16. In addition, Tyou teaches a computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by one or more processors to cause the one or more processors to perform operations (see [0046] and [0090]).
As to claims 11 and 17, these are system and program product claims that correspond to method claim 2. See the rejection for claim 2 above, which also applies to claims 11 and 17.
As to claims 12 and 18, these are system and program product claims that correspond to method claim 4. See the rejection for claim 4 above, which also applies to claims 12 and 18.
As to claims 13 and 19, these are system and program product claims that correspond to method claim 5. See the rejection for claim 5 above, which also applies to claims 13 and 19.
As to claims 14 and 20, these are system and program product claims that correspond to method claim 6. See the rejection for claim 6 above, which also applies to claims 14 and 20.
As to claim 15, this is a system that corresponds to method claim 7. See the rejection for claim 7 above, which also applies to claims 15.
Conclusion
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 LI B ZHEN whose telephone number is (571)272-3768. The examiner can normally be reached M-F, 7:30a-4p.
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. 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.
/Li B. Zhen/Supervisory Patent Examiner, Art Unit 2121
Prosecution Insights