Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Detailed Action
This action is in response to the claims filed 4/12/2023:
Claims 1 – 10 are pending.
Claims 1, 5, 6, and 10 are independent.
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 5 and 10 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.
Regarding claims 5 and 10, “the training including” lacks antecedent basis. Claim 5 introduces “machine learning algorithm trained to classify image data” and “the machine learning algorithm having been trained taking into account at least one inequality constraint” such that it’s unclear if “the training” simultaneously trains a machine learning algorithm to classify image data and take into account at least one inequality constraint, or if there are multiple training procedures, in which case “the training” is ambiguous. As these interpretations are contradictory the scope of the claim cannot be reasonably determined. In the interest of further examination the claim limitation is interpreted as “training a machine learning algorithm to classify image data taking into account at least one inequality constraint”.
Claim Rejections - 35 USC § 101
101 Rejection
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 6-8 and 10 are rejected under 35 USC § 101 because the claimed invention is directed to non-statutory subject matter.
Regarding claims 6-8 and 10, claims 6-8 and 10 are directed towards non-statutory subject matter, “software per se”. Claim 6 for example recites “A controller for training a machine learning algorithm taking into account at least one inequality constraint, wherein each of the at least one inequality constraint represents a secondary constraint, the controller comprising: an optimization unit” however, there is no indication that the controller cannot be entirely software. In fact, the instant specification appears to support the interpretation of the controller being software ([0074] “the controller 10 is in particular configured to perform an above-described method for training a machine learning algorithm taking into account at least one inequality constraint. Furthermore, code implementing the optimization unit, code implementing the training unit, and code implementing the ascertaining unit can also be combined in a computer program product.”). Claim 10 recites analogous elements and claims 7 and 8 depend on claim 6. Therefore, claims 6-8 and 10 are rejected as software-per-se.
Therefore, when considering the elements separately and in combination, they do not add significantly more to the inventive concept. Accordingly, claims 6-8 and 10 are rejected under 35 U.S.C. § 101.
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 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)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1, 2, 6, and 7 are rejected under U.S.C. §102(a)(1) as being anticipated by Neutatz (“Enforcing constraints for machine learning systems via declarative feature selection: an experimental study”, 2021).
Regarding claim 1, Neutatz teaches A method for training a machine learning algorithm taking into account at least one inequality constraint, wherein each of the at least one inequality constraint represents a secondary constraint, the method comprising the following steps: ([p. 5] "The user can specify a single constraint, such as accuracy, or multiple constraints, e.g. EO > 0.9 and accuracy > 0.8" [p. 6] "To enable the single-objective strategies to find feature sets that satisfy multiple constraints, we aggregate the distance to each constraint threshold in a single objective function. Our goal is to minimize the overall distance across all constraints [...] instead of optimizing for classification accuracy, we optimize for minimal distance to satisfy the constraints")
optimizing hyperparameters for the machine learning algorithm by applying a tree-structured Parzen estimator, ([p. 1] “model hyperparameter optimization (HPO)” [p. 5] "Another way to leverage rankings for FS is to pick the top-k features. Approaches based on the second category randomized search, better known as HPO, find the optimal value for k. We choose the well-known tree-structured Parzen estimator approach (Tpe) [7] for this task" [p. 6] "Tree-Structured ParzenEstimator-Tpe(NR). We can solve the same optimization problem of SA(NR) also with the well-known HPO tree-structured Parzen estimator approach")
wherein the tree-structured Parzen estimator is based on an acquisition function adapted based on the at least one inequality constraint; and ([p. 5] "The user can specify a single constraint, such as accuracy, or multiple constraints, e.g. EO > 0.9 and accuracy > 0.8 […] Sequential selection and single-objective randomized search can be further divided into strategies that use a ranking and those that do not (NR) […] We can leverage all the described FS strategies for DFS because they all follow the wrapper approach" [p. 6] "To enable the single-objective strategies to find feature sets that satisfy multiple constraints, we aggregate the distance to each constraint threshold in a single objective function. Our goal is to minimize the overall distance across all constraints [...] instead of optimizing for classification accuracy, we optimize for minimal distance to satisfy the constraints" [p. 6] "Simulated Annealing- SA(NR) […] the optimization goal is to find the binary vector 𝑏 that optimizes a given objective […] Tree-Structured ParzenEstimator-Tpe (NR). We can solve the same optimization problem of SA(NR) also with the well-known HPO tree-structured Parzen estimator approach […] To enable the single-objective strategies to find feature sets that satisfy multiple constraints, we aggregate the distance to each con straint threshold in a single objective function" Eqn. 1 interpreted as acquisition function. Cm interpreted as inequality constraint)
training the machine learning algorithm based on the optimized hyperparameters.([p. 6] "We formulate the meta-learning problem as a multi-label binary classification task where we predict for each strategy whether it can satisfy a given ML scenario or not. Algorithm 1 describes how Dfs Optimizer is trained and deployed." See Algorithm 1).
Regarding claim 2, Neutatz teaches The method as recited in claim 1, further comprising: ascertaining the acquisition function adapted based on the at least one inequality constraint, and wherein the ascertaining of the acquisition function adapted based on the at least one inequality constraint includes factorizing each of the at least one inequality constraint.(Neutatz [p. 6 §4.3] "Our goal is to minimize the overall distance across all constraints. Therefore, we sum up the squared distance of each achieved validation score dm to its corresponding constraint threshold:" The piecewise partial sum distance calculation in Eqn. 1 for each constraint interpreted as synonymous with factorizing each of the at least one inequality constraint to ascertain the acquisition function).
Regarding claim 6, Neutatz teaches A controller for training a machine learning algorithm taking into account at least one inequality constraint, wherein each of the at least one inequality constraint represents a secondary constraint, the controller comprising:([p. 5] "The user can specify a single constraint, such as accuracy, or multiple constraints, e.g. EO > 0.9 and accuracy > 0.8" [p. 6] "To enable the single-objective strategies to find feature sets that satisfy multiple constraints, we aggregate the distance to each constraint threshold in a single objective function. Our goal is to minimize the overall distance across all constraints [...] instead of optimizing for classification accuracy, we optimize for minimal distance to satisfy the constraints")
an optimization unit configured to optimize hyperparameters for the machine learning algorithm by applying a tree-structured Parzen estimator, wherein the tree-structured Parzen estimator is based on an acquisition function adapted based on the at least one inequality constraint; and([p. 5] "Another way to leverage rankings for FS is to pick the top-k features. Approaches based on the second category randomized search, better known as HPO, find the optimal value for k. We choose the well-known tree-structured Parzen estimator approach (Tpe)[7] for this task" [p. 6] "Tree-Structured ParzenEstimator-Tpe(NR). We can solve the same optimization problem of SA(NR) also with the well-known HPO tree-structured Parzen estimator approach")
a training unit configured to train the machine learning algorithm based on the optimized hyperparameters.([p. 6] "We formulate the meta-learning problem as a multi-label binary classification task where we predict for each strategy whether it can satisfy a given ML scenario or not. Algorithm 1 describes how Dfs Optimizer is trained and deployed." See Algorithm 1).
Regarding claim 7, Neutatz teaches The controller as recited in claim 6, further comprising: an ascertaining unit configured to ascertain the acquisition function adapted based on the at least one inequality constraint, the ascertaining of the acquisition function adapted based on the at least one inequality constraint includes factorizing each of the at least one inequality constraint.([p. 6 §4.3] "Our goal is to minimize the overall distance across all constraints. Therefore, we sum up the squared distance of each achieved validation score dm to its corresponding constraint threshold:" The piecewise partial sum distance calculation in Eqn. 1 for each constraint interpreted as synonymous with factorizing each of the at least one inequality constraint to ascertain the acquisition function).
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim 3 and 8 are rejected under U.S.C. §103 as being unpatentable over the combination of Neutatz and Poojari (“Genetic Algorithm based technique for solving Chance Constrained Problems”, 2005).
Regarding claim 3, Neutatz teaches The method as recited in claim 2.
However, Neutatz doesn't explicitly teach, wherein the ascertaining of the acquisition function adapted based on the at least one inequality constraint includes multiplying an acquisition function for an objective function by an acquisition function for each of the at least one inequality constraint.
Poojari, in the same field of endeavor, teaches the ascertaining of the acquisition function adapted based on the at least one inequality constraint includes multiplying an acquisition function for an objective function by an acquisition function for each of the at least one inequality constraint. ([p. 9] "In order to construct the feasibility scoring function we first define the degree of constraint satisfaction. The degree of constraint satisfaction, d(v,i), for an individual measures the relative magnitude of the violation of the constraints with respect to the other individuals in the same generation. The non-negative function, d(v,i), is constructed using the penalty function as follows [See Eqn. 15] […] and the multiplicative penalty is defined as” See Eqn. 17 which is explicitly defined as the product of d(v,i)).
Neutatz as well as Poojari are directed towards machine learning parameter optimization. Therefore, Neutatz as well as Poojari are reasonably pertinent analogous art. It would be obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to adapt the piecewise additive objective function in Neutatz to a piecewise multiplicative objective function as outlined in Eqn. 15-17 of Poojari. Poojari provides as additional motivation for combination ([p. 11] “Our intention is to explore wider regions of the search space at the beginning. By increasing the penalty proportionally to the generation number, it increases the pressure for the population to generate more feasible solutions later on and hopefully converges”).
Regarding claim 8, Neutatz teaches The controller as recited in claim 7.
However, Neutatz doesn't explicitly teach, wherein the ascertaining unit is further configured to ascertain the acquisition function adapted based on the at least one inequality constraint by multiplying an acquisition function for an objective function by an acquisition function for each of the at least one inequality constraint.
Poojari, in the same field of endeavor, teaches The controller as recited in claim 7, wherein the ascertaining unit is further configured to ascertain the acquisition function adapted based on the at least one inequality constraint by multiplying an acquisition function for an objective function by an acquisition function for each of the at least one inequality constraint.([p. 9] "In order to construct the feasibility scoring function we first define the degree of constraint satisfaction. The degree of constraint satisfaction, d(v,i), for an individual measures the relative magnitude of the violation of the constraints with respect to the other individuals in the same generation. The non-negative function, d(v,i), is constructed using the penalty function as follows" See Eqn. 17).
Neutatz as well as Poojari are directed towards machine learning parameter optimization. Therefore, Neutatz as well as Poojari are reasonably pertinent analogous art. It would be obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to adapt the piecewise additive objective function in Neutatz to a piecewise multiplicative objective function as outlined in Eqn. 15-17 of Poojari. Poojari provides as additional motivation for combination ([p. 11] “Our intention is to explore wider regions of the search space at the beginning. By increasing the penalty proportionally to the generation number, it increases the pressure for the population to generate more feasible solutions later on and hopefully converges”).
Claims 4 and 9 are rejected under U.S.C. §103 as being unpatentable over the combination of Neutatz and Marculescu (“Hardware-Aware MachineLearning: Modeling and Optimization”, 2018).
Regarding claim 4, Neutatz teaches The method as recited in claim 1.
However, Neutatz doesn't explicitly teach, wherein the at least one inequality constraint is at least one specification relating to available computing resources.
Marculescu, in the same field of endeavor, teaches the at least one inequality constraint is at least one specification relating to available computing resources. ([p. 1] "ML practitioners are currently challenged with the task of designing the DNN model, i.e., of tuning the hyper-parameters of the DNN architecture, while optimizing for both accuracy of the DL model and its hardware efficiency. Therefore, state-of-the-art methodologies have proposed hardware aware hyper-parameter optimization techniques" [p. 2] "there is an ever-increasing interest in works that co-optimize for both the hardware efficiency and the accuracy of the DL model. In this paper, we investigate the main tools for employing hardware-aware hyper-parameter optimization, such as methodologies based on hardware-aware Bayesian optimization [34, 35], multi-level co-optimization [30] and Neural Architecture Search (NAS)" [p. 4] "Different formulations have been used for SMBO probabilistic models, such as Gaussian Processes (GP) [32] or tree-structured Parzen estimators (TPE) [2]. Regardless of the choice of the SMBO formulation, intuitively the probabilistic model encapsulates the belief about the shape of functions that are more likely to fit the data observed so far, providing us with a cheap approximation for the mean and the uncertainty of the objective function").
Neutatz as well as Marculescu are directed towards using TPE for machine learning model optimization. Therefore, Neutatz as well as Marculescu are analogous art in the same field of endeavor. It would have been obvious before the effective filing date of the claimed invention to combine the teachings of Neutatz with the teachings of Marculescu by using hardware-aware constraints in the multi-constraint TPE objective of Neutatz. Marculescu explicitly anticipates using TPE as a drop-in SMBO for a multi-constraint hardware aware objective and provides as further motivation for combination ([p. 7 §5.2] "while the use of a simpler DNN design can improve the overall runtime, it could significantly degrade the utilization or throughput achieved given a fixed underlying hardware platform. Hence, we believe that several novel approaches that focus on hardware-aware hyper-parameter optimization would be extending current SMBO models to multiple design objectives.").
Regarding claim 9, Neutatz teaches The controller as recited in claim 6.
However, Neutatz doesn't explicitly teach, wherein the at least one inequality constraint is at least one specification relating to available computing resources.
Marculescu, in the same field of endeavor, teaches The controller as recited in claim 6, wherein the at least one inequality constraint is at least one specification relating to available computing resources.([p. 1] "ML practitioners are currently challenged with the task of designing the DNN model, i.e., of tuning the hyper-parameters of the DNN architecture, while optimizing for both accuracy of the DL model and its hardware efficiency. Therefore, state-of-the-art methodologies have proposed hardware aware hyper-parameter optimization techniques" [p. 2] "there is an ever-increasing interest in works that co-optimize for both the hardware efficiency and the accuracy of the DL model. In this paper, we investigate the main tools for employing hardware-aware hyper-parameter optimization, such as methodologies based on hardware-aware Bayesian optimization [34, 35], multi-level co-optimization [30] and Neural Architecture Search (NAS)" [p. 4] "Different formulations have been used for SMBO probabilistic models, such as Gaussian Processes (GP) [32] or tree-structured Parzen estimators (TPE) [2]. Regardless of the choice of the SMBO formulation, intuitively the probabilistic model encapsulates the belief about the shape of functions that are more likely to fit the data observed so far, providing us with a cheap approximation for the mean and the uncertainty of the objective function").
Neutatz as well as Marculescu are directed towards using TPE for machine learning model optimization. Therefore, Neutatz as well as Marculescu are analogous art in the same field of endeavor. It would have been obvious before the effective filing date of the claimed invention to combine the teachings of Neutatz with the teachings of Marculescu by using hardware-aware constraints in the multi-constraint TPE objective of Neutatz. Marculescu explicitly anticipates using TPE as a drop-in SMBO for a multi-constraint hardware aware objective and provides as further motivation for combination ([p. 7 §5.2] "while the use of a simpler DNN design can improve the overall runtime, it could significantly degrade the utilization or throughput achieved given a fixed underlying hardware platform. Hence, we believe that several novel approaches that focus on hardware-aware hyper-parameter optimization would be extending current SMBO models to multiple design objectives.").
Claims 5 and 10 are rejected under U.S.C. §103 as being unpatentable over the combination of Neutatz and Thiagarajan (“Explanation and Use of Uncertainty Quantified by Bayesian Neural Network Classifiers for Breast Histopathology Images”, 2021).
Regarding claim 5, Neutatz teaches the machine learning algorithm having been trained taking into account at least one inequality constraint, wherein each of the at least one inequality constraint represents a secondary constraint, the training including:(Neutatz [p. 5] "The user can specify a single constraint, such as accuracy, or multiple constraints, e.g. EO > 0.9 and accuracy > 0.8" [p. 6] "To enable the single-objective strategies to find feature sets that satisfy multiple constraints, we aggregate the distance to each constraint threshold in a single objective function. Our goal is to minimize the overall distance across all constraints [...] instead of optimizing for classification accuracy, we optimize for minimal distance to satisfy the constraints")
optimizing hyperparameters for the machine learning algorithm by applying a tree-structured Parzen estimator, wherein the tree-structured Parzen estimator is based on an acquisition function adapted based on the at least one inequality constraint; and(Neutatz [p. 5] "The user can specify a single constraint, such as accuracy, or multiple constraints, e.g. EO > 0.9 and accuracy > 0.8" [p. 6] "To enable the single-objective strategies to find feature sets that satisfy multiple constraints, we aggregate the distance to each constraint threshold in a single objective function. Our goal is to minimize the overall distance across all constraints [...] instead of optimizing for classification accuracy, we optimize for minimal distance to satisfy the constraints" Eqn. 1 interpreted as acquisition function. Cm interpreted as inequality constraint)
training the machine learning algorithm based on the optimized hyperparameters.(Neutatz [p. 6] "We formulate the meta-learning problem as a multi-label binary classification task where we predict for each strategy whether it can satisfy a given ML scenario or not. Algorithm 1 describes how Dfs Optimizer is trained and deployed." See Algorithm 1).
However, Neutatz doesn't explicitly teach A method for classifying image data, comprising: classifying image data using a machine learning algorithm trained to classify image data, .
Thiagarajan, in the same field of endeavor, teaches A method for classifying image data, comprising: classifying image data using a machine learning algorithm trained to classify image data, ([p. 2] "The proposed Bayesian-CNN architecture was implemented for image classification" [p. 5] "A Tree-structured Parzen Estimator (TPE) algorithm is used which is a sequential model-based optimization approach [55]. In this approach, models are constructed to approximate the performance of hyperparameters based on historical measurements [...] The architecture of TL-CNN follows VGG-16 [40]. Architectures for Bayesian–CNN and modified Bayesian–CNN are shown in Fig. 2.").
Neutatz as well as Thiagarajan are directed towards using TPE for machine learning model optimization. Therefore, Neutatz as well as Thiagarajan are analogous art in the same field of endeavor. It would have been obvious before the effective filing date of the claimed invention to combine the teachings of Neutatz with the teachings of Thiagarajan by using the model agnostic TPE system in Neutatz as the TPE system in the image classification system of Thiagarajan. Thiagarajan provides as additional motivation for combination ([p. 2] "We found that the Bayesian–CNN improves accuracy and reduces overfitting in comparison to CNN in addition to quantifying uncertainties").
Regarding claim 10, claim 10 is substantially similar to claim 5. Therefore, the rejection applied to claim 5 also applies to claim 10.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Ozaki (“Multiobjective tree-structured parzen estimator”, 2021) is directed towards using TPE for machine learning hyperparameter optimization over multiple constraints.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SIDNEY VINCENT BOSTWICK whose telephone number is (571)272-4720. The examiner can normally be reached M-F 7:30am-5:00pm EST.
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, Miranda Huang can be reached on (571)270-7092. 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.
/SIDNEY VINCENT BOSTWICK/Examiner, Art Unit 2124
/VINCENT GONZALES/Primary Examiner, Art Unit 2124