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 .
Status of the Claims
The status of the claims as of the response filed 6/15/2026 is as follows: Claims 2, 6, 9, 12-21, 23, and 25 remain cancelled. Claims 1, 5-7, 10-11, 22, 24, and 26 are as previously presented. Claims 3-4 and 8 are original. Claims 1, 3-5, 7-8, 10-11, 22, 24, and 26 are currently pending in the application and have been considered below.
Response to Arguments
Rejection Under 35 USC 101
On pages 7-10 of the response filed 6/15/2026 Applicant argues that the claims are patent eligible because they “recite a specific technical implementation involving iteratively receiving parameter gradients from computing nodes, aggregating those gradients to determine parameter updates, and transmitting those updates back to the computing nodes until a convergence condition is reached” and provide “a specific improvement to the federated learning process itself,” e.g. via use of a unified red flag rule transmitted from the master node to the computing nodes and performance of a specific iterative gradient aggregation process and final distortion of the final model, which “ensure[s] data quality across distributed computing nodes in a federated learning system while maintaining data privacy.” Applicant’s arguments are fully considered, and are found persuasive. Accordingly, the 35 USC 101 rejections are withdrawn.
Rejection Under 35 USC 103
On pages 10-11 Applicant argues that “The data filtration taught by Tuor is directed to selecting relevant data for the desired modeling task,” which they submit is “fundamentally different from filtering to remove medical data with ‘significant or obvious errors’” because “the former selects data that is topically appropriate for the modeling objective, whereas the latter removes data that contains errors or defects regardless of its topical relevance.” Applicant further argues that “Patil’s own filtering mechanism… is designed to ‘increase the heterogeneity of the training dataset’ and to create highly tailored training data by selecting specific examples based on metadata about variation features,” which “is distinct from the purpose of the unified red flag rule, which is to remove data with obvious errors that could lead to incorrect model outputs.” Applicant’s arguments are fully considered, but are not persuasive. Examiner notes that the only positively recited limitations in claim 1 related to the red flag rule are that the definition information indicates a unified red flag rule (with the intended result of preventing medical data having significant or obvious errors from being input to the initial medical validation model) and processing the respective local training datasets by filtering out historical medical data satisfying the unified red flag rule. Examiner further notes that the claims do not define what “medical data having significant or obvious errors” actually entails, leaving one of ordinary skill in the art to utilize the broadest reasonable interpretation of such data types. Examiner maintains that filtering out data that is found to be unnecessary or irrelevant to the model being trained as in Patil and Tuor satisfies the broadest reasonable interpretation of “medical data having significant or obvious errors” because including such unnecessary or irrelevant data in a training dataset would clearly be problematic or erroneous with respect to the training of that model. Accordingly, Examiner maintains that the cited combination of prior art references sufficiently teaches claim 1.
On page 11 Applicant argues that the cited portions of Calcutt do “not teach ‘predetermined validation actions’ related to the medical data’” as in claim 4. Applicant’s arguments are fully considered, but are not persuasive. Examiner notes that the first limitation merely describes the content of the definition information; the recited specific types of unified data (e.g. unified validation categories indicating a plurality of predetermined validation actions to be performed on the medical data) do not positively impact or affect the structure or function of the invention such that they amount to non-functional descriptive language. As explained in para. 14 of the non-final rejection mailed 5/13/2026, the limitations of claim 4 will be considered to be satisfied by a prior art reference disclosing definition information that indicates unified or standard features/names for any type of inputs and outputs represented in the local training datasets that are then used to map local input and output features to the unified input and output features. Examiner maintains that the cited combination of references sufficiently suggests this BRI; Calcutt teaches a federated learning method that includes defining data formatting and nomenclature standardization constraints for training data input to the model and harmonizing local training datasets in accordance with the defined constraints (Calcutt [0069], [0077], [0103]), which would be applied to trained models used in a variety of medical contexts when considered in the context of the combination with Patil. Accordingly, Examiner maintains that the positively recited functions of claim 4 are taught by the prior art.
On page 12 Applicant argues that cited combination of references for claim 26 does not “teach or suggest that both the training datasets and the validation dataset are unfiltered,” instead merely suggesting a rationale for not filtering training data. Applicant’s arguments are fully considered, but are not persuasive. Examiner agrees that Patil fails to explicitly disclose that the training datasets are unfiltered, because it specifically mentions removing unnecessary/irrelevant data in [0089]; Zhu is utilized to remedy this deficiency for the training data, with the motivation of expanding model generalizability (as Applicant acknowledges). However, Examiner respectfully disagrees that Patil does not teach use of an unfiltered validation dataset; though it only discusses use of a single validation dataset at a central server in [0095], there is no disclosure of that dataset being filtered in any way, leaving one of ordinary skill in the art to conclude that it can be left unfiltered. Though Patil does not explicitly disclose each local node having its own local validation dataset, Anwar remedies this deficiency by showing a method of validating federated learning models in a distributed fashion with each node having its own local validation dataset (see [0059] & [0072]). Examiner notes that there is again no mention of filtering the local validation datasets of Anwar, leaving one of ordinary skill in the art to conclude that they can be left unfiltered. Accordingly, Examiner maintains that the combination of Patil, Zhu, and Anwar does sufficiently teach all of claim 26, including both the training datasets and validation datasets being unfiltered.
Claim Interpretation
Claim 1 and its dependent claims as well as claim 26 describe the various iterations of the model with the descriptor “medical validation,” but do not positively claim any iterations of the model as performing any medical validation functions or resulting in any specific medical validation outputs. The only positively recited aspects of claims 1 and 26 are directed to exchanging various information about the various iterations of the model between master and computing nodes to perform a federated learning process until a convergence condition is reached. That is, the type of model being specified as a “medical validation” model has no functional bearing on the positively recited claim limitations, because the structure and function of the invention would remain unchanged if the model was described as a “medical validation” model, a “diagnostic” model, a “treatment recommendation” model, or any other adjective. Because the model being a “medical validation” model has no specific functional or structural impact on the claims, this language is considered non-functional descriptive language and is not considered patentably limiting in this case. See MPEP 2111.05. Accordingly, a prior art reference disclosing the federated training of any type of model in the manner recited by claims 1 or 26 will be considered to meet the respective claim language for purposes of analyzing patentability over the prior art.
Claims 1 and 26 further recite “distributing, by the master node, the final medical validation model to at least one of the plurality of computing nodes or at least one further computing node for use in medical validation.” The words “for use in medical validation” do not confer patentable weight because they merely reflect an intended use or intended result of the model distribution step, and do not positively affect the structure or function of the invention, because no step for actually executing the model to achieve a medical validation function is actually claimed. The only positively recited function of this limitation is “distributing, by the master node, the final medical validation model to at least one of the plurality of computing nodes or at least one further computing node,” so the claim language will be considered to be met if the prior art teaches that the final model is distributed to a computing node for any purpose. See MPEP 2111.04.
Claim 3 recites “wherein the respective local training datasets comprise historical medical data generated in medical tests and labeling information indicating local validation categories of the historical data.” This limitation merely further describes the content of the training data, and does not positively impact or affect the structure or function of the invention such that it also amounts to non-functional descriptive language, as explained for claim 1 above. Accordingly, a prior art reference disclosing local training datasets comprising historical data and labeling information of any type will be considered to meet the claim language for purposes of analyzing patentability over the prior art.
Claim 4 recites “wherein the definition information indicates unified item names in medical data input to the initial medical validation model, and unified validation categories output from the initial medical validation model, the unified validation categories indicating a plurality of predetermined validation actions to be performed on the medical data; and wherein the respective local training datasets are processed by mapping local item names used in the historical medical data to the unified item names, and mapping the local validation categories to the unified validation categories.” This first limitation merely further describes the content of the definition information as being unified “item names in medical data” and unified “validation categories”; these specific types of unified data do not positively impact or affect the structure or function of the invention such that they also amount to non-functional descriptive language, as explained for claim 1 above. Accordingly, a prior art reference disclosing definition information indicating unified or standard features/names for any type of inputs and outputs represented in the local training datasets that are then used to map local input and output features to the unified input and output features will be considered to meet the claim language for purposes of analyzing patentability over the prior art.
Claim Rejections - 35 USC § 103
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
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.
Claims 1, 3, 10-11, 22, and 24 are rejected under 35 U.S.C. 103 as being unpatentable over Patil et al. (US 20230351204 A1) in view of Anwar et al. (US 20220292392 A1) and Tuor et al. (Reference U on the accompanying PTO-892).
Claim 1
Patil teaches a computer-implemented method (Patil abstract), comprising:
transmitting, by a master node to a plurality of computing nodes, definition information about an initial medical validation model (Patil [0004], [0082], [0085], noting central server 102 (i.e. a master node) sends information about a global model such as parameters, selected training data, and/or clinical requirements (i.e. definition information) to each clinical site (i.e. a plurality of computing nodes) so that a local copy of the model can be created at each clinical site; see also [0035], noting the model may be trained to perform any type of task that may be performed on medical data by a model, considered sufficient to meet the interpretation of the “medical validation” model outlined in para. 6 above);
performing, by the master node, a federated learning process together with the plurality of computing nodes, to jointly train the initial medical validation model using respective processed local training datasets available at the plurality of computing nodes (Patil [0004], [0082], [0085], [0090]-[0093], noting central server 102 (i.e. a master node) sends information about a global model such as parameters, selected training data, and/or clinical requirements (i.e. definition information) to each clinical site (i.e. a plurality of computing nodes) so that a local copy of the model can be created at each clinical site via a federated learning process with the selected training data meeting the clinical requirements (i.e. with local training datasets processed by the plurality of computing nodes based on the definition information));
determining, by the master node, a final medical validation model based on a result of the federated learning process (Patil [0004], [0083], noting the central server receives results of the federated learning process and creates or updates a global (i.e. final) copy of the model),
wherein (Patil [0089], noting each clinical site may use a sample eliminator-augmenter to remove unnecessary/irrelevant samples from the training data; the sample eliminator-augmenter is considered equivalent to use of a red flag rule to filter out medical data having significant or obvious errors (such as being irrelevant or unnecessary)), and
wherein the federated learning process comprises: obtaining, by the master node, a trained medical validation model from the result of the federated learning process; distributing the trained medical validation model to the plurality of computing nodes; receiving feedback (Patil [0092]-[0095], noting the central server receives the trained models from each clinical site as a result of the federated training process, generates or updates the global model, and evaluates performance metrics (e.g. accuracy) of the global model using validation data to determine when training should end. [0095] further notes that the trained model may be distributed back to the plurality of computing nodes for retraining); and
wherein determining the trained medical validation model from the results of the federated learning process by the master node comprises iteratively performing steps of: receiving, from the plurality of computing nodes, parameter gradients generated by the plurality of computing nodes based on the respective processed local training datasets; aggregating the received parameter gradients to determine parameter updates; and transmitting the parameter updates to the plurality of computing nodes to update intermediate initial medical validation models of the plurality of computing nodes until a convergence condition for the federated learning process is reached to obtain the trained medical validation model (Patil Fig. 6, [0093]-[0095], noting the central server receives the local parameter gradients trained generated from each clinical site as a result of the federated training process, considers them together (i.e. aggregates them) to generate or update weights of the global model, and evaluates a validation condition to determine whether training should end or whether the updated weights should be distributed back to the plurality of computing nodes for another iteration of retraining. Because Applicant’s specification provides no definition of a “convergence condition” beyond that it is a condition triggering the end of an iterative federated learning process as in [0088], the validation condition of Patil is considered functionally equivalent to the convergence condition of the instant claim); and
distributing, by the master node, the final medical validation model to at least one of the plurality of computing nodes or at least one further computing node for use in medical validation (Patil [0095], noting the global model may be distributed from the central server back to the clinical sites (i.e. computing nodes) for further training; this meets the interpretation of the claim language as outlined in para. 7 above).
In summary, Patil teaches a method for performing federated learning on local datasets which may be locally processed to remove irrelevant/unnecessary data (i.e. data meeting a red flag rule). However, Patil fails to explicitly disclose that the definition information sent out from the master node to each of the computing nodes contains a unified red flag rule against which each of the local training datasets are evaluated for data filtering purposes. However, Tuor teaches an analogous federated learning method in which a unified training data selection/filtration criterion is centrally determined and disseminated to each local node so that irrelevant/noisy data is filtered out or removed from the training dataset prior to performing the federated learning process (Tuor Introduction on Pg 5020, section III on Pg 5021, section IV on Pgs 5022-5023). It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the local training data filtration method of Patil such that the filtration rules/criteria are specified in a unified manner via definition information sent by the master node to all client nodes as in Tuor in order to allow the model requestor to centrally indicate what types of data are relevant so that the local model instances are only trained on data meeting unified standards for relevancy to the desired modeling task without noisy information that would have a negative impact on the learned model (as suggested by Tuor abstract & Introduction on Pg 5020).
The combination of Patil and Tuor thus teaches a method for performing federated learning on local datasets filtered to meet unified criteria, combining the results of the federated learning into a final model at the central server, and using validation data at the central server to determine performance metric feedback of the model and determine whether a condition is met or whether the weights of the global model should be distributed back to the clinical sites for additional epochs of training. This disclosed validation method relies on a single validation operation that takes place at the central server, rather than a distributed validation operation taking place with local validation datasets at each computing node as required by the claim. Accordingly, the present combination fails to explicitly disclose receiving feedback from the plurality of computing nodes, the feedback indicating respective performance metrics of the trained medical validation model determined by the computing nodes using respective local validation datasets.
However, Anwar teaches an analogous federated learning architecture wherein the local nodes each receive a copy of the global model, evaluate the global model’s performance with respect to a local data set, and send the performance feedback to the central server as an indicator of the global model’s bias for or against the local node’s dataset (Anwar [0059], [0072]). It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the model validation operation of the combination such that it takes place in a distributed manner as in Anwar in order to determine how much the global model is biased for or against the local data of each computing node (as suggested by Anwar [0059]), thereby allowing for improved global model evaluation and selection.
Claim 3
Patil in view of Anwar and Tuor teaches the method of claim 1, and the combination further teaches wherein the respective local training datasets comprise historical medical data generated in medical tests and labeling information indicating local validation categories of the historical medical data (Patil [0037], [0058], noting the models learn a mapping between example inputs (i.e. historical data) and ground truth labels (i.e. labeling information) in the local training data; this meets the interpretation of the claim language as outlined in para. 8 above).
Claim 10
Patil in view of Anwar and Tuor teaches the method of claim 1, and the combination further teaches wherein determining the final medical validation model based on the received feedback comprises: in response to the respective performance metrics meeting a model release criterion, determining the trained medical validation model as the final medical validation model; and in response to the respective performance metrics failing to meet the model release criterion, adjusting the trained medical validation model to generate the final medical validation model (Patil [0095], noting that if the model meets a predetermined accuracy requirement (i.e. a model release criterion) the training ends (i.e. the model is determined to be the final model), but if the criterion is not met further retraining occurs; when considered in the context of the combination with Anwar, the performance/accuracy metric being evaluated as in [0095] of Patil would be based on the respective performance metrics provided by each local computing node’s distributed validation procedure as in Anwar).
Claim 11
Patil in view of Anwar and Tuor teaches the method of claim 1, and the combination further teaches wherein the master node is communicatively connected with the plurality of computing nodes in a star topology network (Patil Fig. 1, showing central server 102 as a central hub communicating with clinical sites 104 to 112 in a star topology).
Claim 22
Patil in view of Anwar and Tuor teaches an electronic device comprising: at least one processor; and at least one memory comprising computer readable instructions which, when executed by the at least one processor of the electronic device, cause the electronic device to perform the steps of claim 1 (Patil [0098]-[0101]).
Claim 24
Patil in view of Anwar and Tuor teaches a non-transitory computer readable medium having stored thereon a computer program product comprising instructions which, when executed by a processor of an apparatus, cause the apparatus to perform the steps of claim 1 (Patil [0098]-[0101]).
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Patil, Anwar, and Tuor as applied to claims 1 and 3 above, and further in view of Calcutt et al. (US 20200311300 A1).
Claim 4
Patil in view of Anwar and Tuor teaches the method of claim 3, but the present combination fails to explicitly disclose wherein the definition information indicates unified item names in medical data input to the initial medical validation model, and unified validation categories output from the initial medical validation model, the unified validation categories indicating a plurality of predetermined validation actions to be performed on the medical data; and wherein the respective local training datasets are processed by mapping local item names used in the historical medical data to the unified item names, and mapping the local validation categories to the unified validation categories.
However, Calcutt teaches an analogous federated learning method that includes defining data formatting and nomenclature standardization constraints for training data input to the model and harmonizing local training datasets in accordance with the defined constraints (Calcutt [0069], [0077], [0103], considered to meet the claim interpretation outlined in para. 9 above). It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the model definition information and training process of the combination to include defining unified feature names/formats for the local training datasets and performing mapping of the local training datasets to the unified feature names/formats as in Calcutt in order to improve the training process by ensuring standardization and cohesion of the local training datasets while still maintaining privacy of the local datasets (as suggested by Calcutt [0077] & [0103]).
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Patil and Anwar and Tuor as applied to claims 1 and 3 above, and further in view of Fang et al. (Reference U on the PTO-892 mailed 7/15/2025).
Claim 5
Patil in view of Anwar and Tuor teaches the method of claim 3, but the present combination fails to explicitly disclose wherein the definition information further indicates a scaled value range for an item in medical data input to the initial medical validation model, and wherein the respective local training datasets are processed by mapping values of the item in the historical medical data into values within the scaled value range.
However, Fang teaches an analogous federated learning method that includes defining data value scaling ranges for training data input to the model and transforming local training datasets in accordance with the defined scaling ranges (Fang Fig. 5, third paragraph of “5 Experiment” on Pg 8, noting features are transformed via normalization scaling to values between [-k, k] or [-1, 1] in federated learning training datasets in accordance with defined transformations). It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the model definition information and training process of the combination to include defining scaled value ranges for items of the local training datasets and performing normalization transformations to scale the features as in Fang because appropriate transformations help scale the features and significantly improve the performance of learning models (as suggested by Fang first paragraph of “1 Introduction” on Pg 1).
Claims 7-8 are rejected under 35 U.S.C. 103 as being unpatentable over Patil and Anwar and Tuor as applied to claim 1 above, and further in view of Krishnapuram et al. (US 20140088989 A1).
Claim 7
Patil in view of Anwar and Tuor teaches the method of claim 1, and the combination further teaches wherein the definition information indicates an item in medical data input to the initial medical validation model, a value of the indicated item being unavailable from historical medical data in a local training dataset, and wherein the local training datasets are processed by filling in a (Patil [0089], noting each clinical site checks for local training data meeting the clinical requirements and may use a sample eliminator-augmenter to add additional data, considered equivalent to the clinical requirements (i.e. definition information) indicating an item in medical data for filling in).
In summary, the combination teaches a method for performing federated learning on local datasets based on clinical requirements, as well as augmenting the training data to add additional data. However, it fails to explicitly disclose that the additional filled in value is a predetermined value for an indicated item of medical data. However, Krishnapuram teaches an analogous federated learning method that includes site-specific data imputation methods to augment the training data when certain medical item values are missing by filling in a predetermined data type (Krishnapuram [0091]). It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the training data augmentation methods of the combination to include data imputation of a predetermined value for a certain type of missing data as in Krishnapuram in order to leverage the localized data distributions of medical items specific to each clinical site when filling in missing data (as suggested by Krishnapuram [0091]).
Claim 8
Patil in view of Anwar, Tuor, and Krishnapuram teaches the method of claim 7, and the combination further teaches wherein the predetermined value comprises either one of an average value of a reference value range of the indicated item and a median value of available values of the indicated item in historical medical data generated in other medical tests (Krishnapuram [0091], noting the imputed value may be an average or median value known from the site-specific localized dataset).
Claim 26 is rejected under 35 U.S.C. 103 as being unpatentable over Patil in view of Anwar and Zhu et al. (US 20220114475 A1).
Claim 26
Patil teaches a computer-implemented method (Patil abstract), comprising:
transmitting, by a master node to a plurality of computing nodes, definition information about an initial medical validation model (Patil [0004], [0082], [0085], noting central server 102 (i.e. a master node) sends information about a global model such as parameters, selected training data, and/or clinical requirements (i.e. definition information) to each clinical site (i.e. a plurality of computing nodes) so that a local copy of the model can be created at each clinical site; see also [0035], noting the model may be trained to perform any type of task that may be performed on medical data by a model, considered sufficient to meet the interpretation of the “medical validation” model outlined in para. 6 above);
performing, by the master node, a federated learning process together with the plurality of computing nodes, to jointly train the initial medical validation model using respective processed local training datasets available at the plurality of computing nodes (Patil [0004], [0082], [0085], [0090]-[0093], noting central server 102 (i.e. a master node) sends information about a global model such as parameters, selected training data, and/or clinical requirements (i.e. definition information) to each clinical site (i.e. a plurality of computing nodes) so that a local copy of the model can be created at each clinical site via a federated learning process with the selected training data meeting the clinical requirements (i.e. with local training datasets processed by the plurality of computing nodes based on the definition information));
determining, by the master node, a final medical validation model based on a result of the federated learning process (Patil [0004], [0083], noting the central server receives results of the federated learning process and creates or updates a global (i.e. final) copy of the model),
wherein the federated learning process comprises: obtaining, by the master node, a trained medical validation model from the result of the federated learning process; distributing the trained medical validation model to the plurality of computing nodes; receiving feedback (Patil [0092]-[0095], noting the central server receives the trained models from each clinical site as a result of the federated training process, generates or updates the global model, and evaluates performance metrics (e.g. accuracy) of the global model using validation data (which there is no mention of filtering) to determine when training should end. [0095] further notes that the trained model may be distributed back to the plurality of computing nodes for retraining); and
wherein determining the trained medical validation model from the results of the federated learning process by the master node comprises iteratively performing steps of: receiving, from the plurality of computing nodes, parameter gradients generated by the plurality of computing nodes based on the respective processed local training datasets; aggregating the received parameter gradients to determine parameter updates; and transmitting the parameter updates to the plurality of computing nodes to update intermediate initial medical validation models of the plurality of computing nodes until a convergence condition for the federated learning process is reached to obtain the trained medical validation model (Patil Fig. 6, [0093]-[0095], noting the central server receives the local parameter gradients trained generated from each clinical site as a result of the federated training process, considers them together (i.e. aggregates them) to generate or update weights of the global model, and evaluates a validation condition to determine whether training should end or whether the updated weights should be distributed back to the plurality of computing nodes for another iteration of retraining. Because Applicant’s specification provides no definition of a “convergence condition” beyond that it is a condition triggering the end of an iterative federated learning process as in [0088], the validation condition of Patil is considered functionally equivalent to the convergence condition of the instant claim); and
distributing, by the master node, the final medical validation model to at least one of the plurality of computing nodes or at least one further computing node for use in medical validation (Patil [0095], noting the global model may be distributed from the central server back to the clinical sites (i.e. computing nodes) for further training; this meets the interpretation of the claim language as outlined in para. 7 above).
In summary, Patil teaches a method for performing federated learning on local datasets which may be locally processed to remove irrelevant/unnecessary data, such that this reference fails to explicitly disclose wherein the local training datasets are not filtered to remove medical data from the local training datasets during the federated learning process. However, Zhu teaches an analogous federated learning process that includes the ability to train a localized model on an entire (i.e. unfiltered) local dataset as an alternative embodiment to training the localized model on a subset of the local dataset (Zhu [0048], [0092]). It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the localized learning on a filtered training dataset as in Patil to include training on an entire local dataset as in Zhu because Zhu shows that these are equally desirable alternative embodiments for a federated learning framework (as suggested by Zhu [0048] & [0092]) and doing so would permit expanded model generalizability.
The combination of Patil and Zhu thus teaches a method for performing federated learning on entire local datasets, combining the results of the federated learning into a final model at the central server, and using validation data at the central server to determine performance metric feedback of the model and determine whether a condition is met or whether the weights of the global model should be distributed back to the clinical sites for additional epochs of training. This disclosed validation method relies on a single validation operation that takes place at the central server, rather than a distributed validation operation taking place with local validation datasets at each computing node as required by the claim. Accordingly, the present combination fails to explicitly disclose receiving feedback from the plurality of computing nodes, the feedback indicating respective performance metrics of the trained medical validation model determined by the computing nodes using respective local validation datasets.
However, Anwar teaches an analogous federated learning architecture wherein the local nodes each receive a copy of the global model, evaluate the global model’s performance with respect to a local data set (which there is no mention of filtering), and send the performance feedback to the central server as an indicator of the global model’s bias for or against the local node’s dataset (Anwar [0059], [0072]). It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the model validation operation of the combination such that it takes place in a distributed manner as in Anwar in order to determine how much the global model is biased for or against the local data of each computing node (as suggested by Anwar [0059]), thereby allowing for improved global model evaluation and selection.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KAREN A HRANEK whose telephone number is (571)272-1679. The examiner can normally be reached M-F 8:00-4:00 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, Shahid Merchant can be reached at 571-270-1360. 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.
/KAREN A HRANEK/ Primary Examiner, Art Unit 3684
Prosecution Insights