DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 07/24/2026 has been entered.
Response to Amendment
Claims 1-20 remain pending within the application.
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.
Claims 1-2, 4-6, 8, 9, 11-13, 15, 16, 18, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Goyal et al. (Pub. No.: US 2021/0374566 A1), hereafter Goyal, in view of Gulluk et al. ("Sample Efficient Subspace-Based Representations For Nonlinear Meta-Learning"), hereafter Gulluk.
Regarding claim 1, Goyal discloses:
A method for machine learning of multiple tasks by iteratively building a model zoo, the method comprising:… a multi-task learner implemented using at least one processor (Goyal, Fig. 2 and ¶[0034-0035] teaches the multi-task learner in Fig. 2 implemented using at least one processor),
initializing each task of a plurality of potentially competing tasks with a task- specific weight (Goyal, Fig. 3 and ¶[0020] teaches initializing the weights of competing tasks),
wherein each task comprises a supervised learning task in which a model is provided access to a plurality of input-label samples and learns to classify inputs associated with the task (Goyal, ¶[0002] teaches labeled training of machine learning models for each new task as each task comprising a supervised learning task where model learns to classify labeled inputs associated with the task),
in a first round, automatically selecting, using the task-specific weights, by a multi-task learner implemented using at least one processor, a set of the potentially competing tasks (Goyal, ¶[0020], ¶[0039], Fig. 3 step 1, and Fig. 8 teaches automatically selecting previous tasks and a new task as a set of potentially competing tasks from a plurality, where the tasks are shown to be potentially competing in Fig. 8 and the selection occurs in a first round following the initialization and by using the task specific weights),
training, by the multi-task learner, a first machine learning model using the input-label samples for each task of the selected set of potentially competing tasks (Goyal, Fig. 3 step 1 and step 2, and ¶[0076] teaches training and learning task classifiers as training a first machine learning model on each task of the selected first set of potentially competing tasks, where the machine learning model is a supervised model using input label samples for each task),
calculating, by the multi-task learner and for each task in the set, a metric of empirical risk of the first model on each task of the set (Goyal, Fig. 3 step 1, step 2, and ¶[0048] teaches a weighed error as a metric of empirical risk of the first model on each task of the set of potentially competing tasks, where the error is computed for each task in the set as new tasks are added),
updating the task-specific weights of the tasks in the set using the metrics of empirical risk, wherein the tasks for which the first model has lower metrics of empirical risk are assigned … task-specific weights than the tasks for which the first model has higher metrics of empirical risk (Goyal, ¶[0048] teaches updating the task-specific weights of the tasks in the set using the metrics of empirical risk, i.e. error, where the tasks for which the model has lower metrics of error are assigned higher task specific weights than tasks for which the first model has higher error),
in a second round after the first round, automatically selecting, by the multi-task learner from the potentially competing tasks, and using the task- specific weights, a subset of potentially competing tasks (Goyal, Fig. 3 line 4 and step 1 teaches a subsequent iteration from the previous round, where now the second subset comprises a new task different from the previous round’s new task and selecting a subset of potentially competing tasks using the task specific weights),
training, by the multi-task learner, a second machine learning model using input-label samples for each task of the subset of potentially competing tasks (Goyal, Fig. 3 step 1 and step 2, ¶[0076] teaches training and learning task classifiers for this subsequent round as training a second machine learning model using input label samples, on each task of the selected second subset of potentially competing tasks),
forming, by the multi-task learner, an ensemble machine learning model that includes the first and second machine learning models (Goyal, Fig. 3, step “return” and step 2 teaches combining the task specific classifiers to form an ensemble model by majority vote and boosting),
generating predictions for inputs associated with the set of tasks and the subset of tasks using the ensemble machine learning model (Goyal, Fig. 3, step “return”, step 2, ¶[0076] and ¶[0066] teaches generating predictions for inputs to the model associated with the set of tasks and the subset of tasks using the ensemble model).
While Goyal teaches updating the task-specific weights of the tasks in the set using the metrics of empirical risk, wherein the tasks for which the first model has lower metrics of empirical risk are assigned … task-specific weights than the tasks for which the first model has higher metrics of empirical risk, they do not teach assigning lower weights with lower metrics of empirical risk.
Gulluk teaches:
tasks … which … has lower metrics of empirical risk are assigned lower … weights than the tasks …which … has higher metrics of empirical risk (Gulluk, page 3 assumption 2 and definition 2.6, page 3, right column, first paragraph, lines 3-6 “ Theorem 3.4 upper bounds the population cross-entropy risk of f in the few-shot learning stage. Theorem 3.5 upper bounds the population risk of the ERM estimator fe, which is learned from finite data.”, and page 4, right column, paragraph 2, final 3 lines “The weight matrices Wj1, Wj2, and Wj3 are different for each task” teaches assigning lower weights to task with lower metrics of risk for risk minimization).
Goyal and Gulluk are analogous art because they are from the same field of endeavor, multi-task learning and machine learning models.
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Goyal to include tasks … which … has lower metrics of empirical risk are assigned lower … weights than the tasks …which … has higher metrics of empirical risk, based on the teachings of Gulluk. One of ordinary skill in the art would have been motivated to make this modification in order to improve sample efficiency of future tasks, as suggested by Gulluk (page 2, left column, second bullet point).
Regarding claim 2, Goyal, in view of Gulluk, discloses the method of claim 1 (and thus the rejection of claim 1 is incorporated). Goyal further discloses:
wherein selecting the set and the (Goyal, Fig. 2, Fig. 3 and ¶[0046] teaches maintaining, per round, a vector of task specific weights and selecting the subsequent sets of competing tasks based on the task-specific weights in the knowledge base).
Regarding claim 4, Goyal, in view of Gulluk, discloses the method of claim 1 (and thus the rejection of claim 1 is incorporated). Goyal further discloses:
further comprising revisiting at least a first task from the plurality of potentially competing tasks by adding one or more new models (Goyal, Fig. 3 and ¶[0051-0052] teaches revisiting competing tasks from previous tasks by adding newly learned classifiers).
Regarding claim 5, Goyal, in view of Gulluk, discloses the method of claim 4 (and thus the rejection of claim 4 is incorporated). Goyal further discloses:
further comprising maintaining one or more models from previous training rounds not updated in successive training rounds (Goyal, Fig. 2 and ¶[0037] teaches maintaining one or more models from previous training rounds not updated in successive training rounds in a knowledge base).
Regarding claim 6, Goyal, in view of Gulluk, discloses the method of claim 1 (and thus the rejection of claim 1 is incorporated). Goyal further discloses:
wherein each of the plurality of(Goyal, ¶[0086-0088] teaches each of the plurality of competing tasks shares a common input domain from the same dataset).
Claims 8 and 15 are substantially similar to claim 1, and thus are rejected on the same basis as claim 1.
Claims 9 and 16 are substantially similar to claim 2, thus are rejected on the same basis as claim 2.
Claims 11-13 are substantially similar to claims 4-6, thus are rejected on the same basis as claims 4-6.
Claim 18 is substantially similar to claim 4, thus are rejected on the same basis as claim 4.
Claims 19 is substantially similar to claim 6, thus are rejected on the same basis as claim 6.
Claims 3, 10, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Goyal et al. (Pub. No.: US 2021/0374566 A1), hereafter Goyal, in view of Gulluk et al. ("Sample Efficient Subspace-Based Representations For Nonlinear Meta-Learning"), hereafter Gulluk, in further view of Chai et al. ("Multi-task Learning with Gaussian Processes"), hereafter Chai.
Regarding claim 3, Goyal, in view of Gulluk, discloses the method of claim 2 (and thus the rejection of claim 2 is incorporated). Goyal further discloses:
wherein selecting the second subset of potentially competing tasks based on the task-specific weights comprises drawing the second subset of potentially competing tasks from a … distribution of the task-specific weights (Goyal, Fig 3 and ¶[0021] teaches drawing the second subset of potentially competing tasks from a distribution of the task-specific weights).
wherein selecting the set and the subset of potentially competing tasks using the vector of task-specific weights comprises drawing the set and the subset of potentially competing tasks from a … distribution of the potentially competing tasks weighted with the task-specific weights (Goyal, Fig 3 and ¶[0021] teaches drawing the sets of potentially competing tasks from a distribution of potentially competing tasks weighted with the task-specific weights).
While Goyal teaches drawing the set and the subset of potentially competing tasks from a … distribution of the potentially competing tasks weighted with the task-specific weights, they do not teach this distribution to be a multinomial distribution.
Chai discloses:
drawing … set.. of … tasks from a multinomial distribution of the task-specific weights (Chai, page 18, section 2.2.3, paragraph 1 “
PNG
media_image1.png
137
656
media_image1.png
Greyscale
” teaches drawing sets of tasks from a multinomial distribution of task specific weights using task clustering).
Goyal, Gulluk, and Chai are analogous art because they are from the same field of endeavor, task learning and machine learning models.
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Goyal, in view of Gulluk, to include drawing … set.. of … tasks from a multinomial distribution of the task-specific weights, based on the teachings of Chai. One of ordinary skill in the art would have been motivated to make this modification in order to improve predictive performance, as suggested by Chai (Chai, page 129, paragraph 1, line 2).
Claims 10 and 17 are substantially similar to claim 3, thus are rejected on the same basis as claim 3.
Claims 7, 14, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Goyal et al. (Pub. No.: US 2021/0374566 A1), hereafter Goyal, in view of Gulluk et al. ("Sample Efficient Subspace-Based Representations For Nonlinear Meta-Learning"), hereafter Gulluk, in further view of Moura et al. (Pub. No.: US 2020/0118423 A1), hereafter Moura.
Regarding claim 7, Goyal, in view of Gulluk, discloses the method of claim 1 (and thus the rejection of claim 1 is incorporated).
While Goyal discloses potentially competing tasks (Fig. 8), they do not disclose:
learning at least one task-specific adapter for at least one task … having a different input domain from at least one other task …
Moura discloses:
learning at least one task-specific adapter for at least one task … having a different input domain from at least one other task …(Moura, ¶[0052] teaches adapting the model to different data domains as learning at least one task-specific adapter for at least one competing task having a different input domain from at least one other competing task).
Goyal, Gulluk, and Moura are analogous art because they are from the same field of endeavor, multi-task learning and machine learning models.
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Goyal, in view of Gulluk, to include learning at least one task-specific adapter for at least one task … having a different input domain from at least one other task …, based on the teachings of Moura. One of ordinary skill in the art would have been motivated to make this modification in order to achieve a high convergence rate and improve accuracy, as suggested by Moura (Moura, ¶[0063]).
Claims 14 and 20 are substantially similar to claim 7, thus are rejected on the same basis as claim 7.
Response to Arguments
Applicant's arguments filed 07/24/2026 have been fully considered with regards to the 35 U.S.C. 102/103 rejection, but they are not persuasive.
The applicant asserts on page 8 of the remarks that Goyal does not teach the limitation “wherein each task comprises a supervised learning task in which a model is provided access to a plurality of input-label samples and learns to classify inputs associated with the task”. The examiner respectfully disagrees, as Fig. 2, Fig. 3, ¶[0002] and ¶[0039] teaches labeled training of machine learning models for each new task as each task comprising a supervised learning task where the model learns to classify labeled inputs associated with the task. The BRI of the claims at present to do not show a “task level pool sampling of competing tasks” that differs from the sets of tasks recited in the claims and taught by Goyal.
The applicant asserts on page 10 of the remarks “Goyal describes weighting of "each example within a learning sample", which refers to updating of individual samples, versus tasks and does not describe using the weights for task selection in subsequent boosting rounds.”. The examiner respectfully disagrees, as Fig. 3 explicitly recites weights for each task classifier as weights for task selection, and these weights are used for subsequent task selection during the iterative process.
Regarding further arguments presented in pages 9-11 of the remarks, these arguments have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. The examiner refers to the rejection under 35 USC § 103 in the current office action for more details.
Claims 8 and 15 are substantially similar to claim 1 and are rejected on the same basis.
Claims dependent on the independent claims do not overcome the deficiencies of the rejected independent claims.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's
disclosure.
- U.S. Pub No. 20230267307 A1: Wang et al. teaches multitask learning weight adjustment of tasks.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HUMAIRA ZAHIN MAUNI whose telephone number is (703)756-5654. The examiner can normally be reached Monday - Friday, 9 am - 5 pm (ET).
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, MATT ELL can be reached at (571) 270-3264. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/H.Z.M./Examiner, Art Unit 2141
/MATTHEW ELL/Supervisory Patent Examiner, Art Unit 2141
