Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment
This Final Rejection is Filed in response to Applicant Arguments/Remarks Made in an Amendment filed 01/15/2026.
Claims 3, 5, 7-10, 12, 16, 19, & 23 are amended.
Claims 1-20 & 22-23 remain pending.
Response to Arguments
Argument 1, Applicant argues in Applicant Arguments/Remarks Made in an Amendment filed 01/15/2026, pg. 10-11 that prior art fails to teach the primary limitations of,
“receiving, from the one or more of the plurality of client units, one or more readily trained machine learned models”.
Response to Argument 1, the examiner respectfully disagrees. Applicant’s specification para. [0194] describes, “evaluating a plethora of different local data sets in a heterogeneous environment of many different local client units in the form of readily trained (incremental) machine learned models. This has the benefit that a plurality (incremental) machine learned models are held available for deployment at the local sites if necessary”, wherein the at most the specification describes “readily trained machine learning models” as incremental models. It is unclear if applicant wishes to describe “readily trained” models as models that have been pre-trained or models that undergo incremental learning by a user, as the two represent different strategies for maintaining machine learning models.
Even applicants arguments which characterize Burger as, “one model and various parameters that may be updated to retrain the model”, would read on the BRI of the above limitations as using an updated model would be a using a readily trained model that has been incrementally trained by a user. For the reasons listed above, Burger teaches a central server that receive readily trained machine learning models from client devices.
The following paragraphs of Burger support this interpretation
para. [0075-0085], FIG. 5 is a system diagram of an example client computing device 500 for performing incremental training of a machine learning tool 510…. The server computer can complete the incremental training by aggregating the gradients from various client devices and updating the operational parameters of the machine learning model. Thus, the incremental training can be partially performed at the client computing device 500.
Argument 2, Applicant argues in Applicant Arguments/Remarks Made in an Amendment filed 01/15/2026, pg. 11-14, that applicant respectfully disagrees with the combination of Burger, Grzybowski, and Rainero.
Response to Argument 2, the examiner respectfully disagrees with applicant’s characterization that Burger is not concerned with a particular user experience, when Burger teaches that is it a user on a client device performing incremental training on a previously trained model in order to produce a more accurate classification model. Applicant’s arguments that a toolset to create a plurality of machine learning models is not taught by Grzybowski which only teaches a toolset to create a virtual environment is seen as arguments against the references individually. In response to applicant's arguments against the references individually, one cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986).
In response to applicant’s argument that there is no teaching, suggestion, or motivation to combine the references, the examiner recognizes that obviousness may be established by combining or modifying the teachings of the prior art to produce the claimed invention where there is some teaching, suggestion, or motivation to do so found either in the references themselves or in the knowledge generally available to one of ordinary skill in the art. See In re Fine, 837 F.2d 1071, 5 USPQ2d 1596 (Fed. Cir. 1988), In re Jones, 958 F.2d 347, 21 USPQ2d 1941 (Fed. Cir. 1992), and KSR International Co. v. Teleflex, Inc., 550 U.S. 398, 82 USPQ2d 1385 (2007). In this case, both Burger and Grzybowski deal with the transmission of a software package from a central server to a remote client device. One would have been motivated to combine how Grzybowski teaches tailoring and sending a user specific software toolset to a user’s client device with how Burger’s capabilities to send a tailored machine learning toolset to a user’s client device based on the client’s power capabilities in order save user’s time by having virtual tools made specific to the device that the individual is using and even specific to the individual's personal requirements or settings.
Argument 3, Applicant argues in Applicant Arguments/Remarks Made in an Amendment filed 01/15/2026, pg. 14, that “no portion of paragraph [0094] of Burger describes downloading an entire machine learned model that has been updated toa client unit that used a previous version of that particular model.
Response to Argument 3, the examiner respectfully disagrees as Burger teaches that an incrementally trained model executing on a client device may have its operating parameter updated by a central server computer. Burger further teaches in para. [0019], that “any of the software-based examples (comprising, for example, computer-executable instructions for causing a computer to perform any of the disclosed methods) can be uploaded, downloaded, or remotely accessed through a suitable communication means”. Thus the BRI of the claim 6 limitation, “downloading the one or more updated machine learned models to all of the plurality of client units that use previous versions of the respective one or more updated machine learned models”, would encompass how each client unit may download additional updates to a previous version of a model in order to bring the local model in line with a most recently incrementally trained model on a server.
Argument 4, Applicant argues in Applicant Arguments/Remarks Made in an Amendment filed 01/15/2026, pg. 14-15, that Bjork fails to teach the newly amended dependent Claim 23 limitation, “wherein the toolset includes one or more algorithms selected based on the configuration data, the one or more algorithms including untrained and trainable machine learned algorithms”.
Response to Argument 4, Applicant arguments have been considered, however in light of the amendments a newly found combination of prior art (U.S. Patent Application Publication NO. 20200265301 “Burger” in light of U.S. Patent Application Publication NO. 20130173509 “Bjork”, in light of U.S. Patent Application Publication NO. 20110145272 “Grzybowski”, and further in light of U.S. Patent Application Publication NO. 20060026511 “Rainero”) is applied to updated rejections.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitation(s) is/are:
“an interface unit configured to communicate with the plurality of client units”, and
“the computing unit being configured to receive, from the one or more of the plurality of client units, configuration data, the configuration data including information about a field of application of a machine learning method at the respective client unit, computational or cost limitations of the respective client unit, and local metadata of the respective client unit…” in claim 12.
Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
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.
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.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1-9 & 11-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication NO. 20200265301 “Burger” in light of U.S. Patent Application Publication NO. 20130173509 “Bjork”, in light of U.S. Patent Application Publication NO. 20110145272 “Grzybowski”, and further in light of U.S. Patent Application Publication NO. 20060026511 “Rainero”.
Claim 1:
Burger teaches a computer-implemented method for client-specific federated learning in a system including a central server unit (i.e. para. [0029], Fig. 1, “the server computer(s) 110 can be located in a datacenter as part of a cloud service that is offered for use by customers of the cloud service provider”, wherein it is noted that tools such as, a modelling framework 431 can be used to define and use a neural network model, that can be seen in Fig. 5, that can be offered to data scientist customers to create different neural network models) and a plurality of client units (i.e. para. [0029], including one or more server computer(s) 110), the plurality of client units being respectively located at different respective local sites (i.e. para. [0117], The disclosed technology may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network) and including respectively local data (i.e. para. [0033], Fig. 1, “machine learning tool 170 can receive raw and/or processed input data from the input device 124”, wherein it is noted that the input data is received at a local client device, thus each local client device site include its respectively local input data), the computer-implemented method comprising:
receiving, from one or more of the plurality of client units, configuration data , the configuration data including information about a field of application of a machine learning method at the respective client unit, computational or cost limitations of the respective client unit (i.e. para. [0077], “the machine learning tool 510 can be a stripped-down or reduced functionality version of the machine learning tool 410 from FIG. 4. For example, the client computing device 500 may have reduced computing power, reduced memory or storage, or a reduced energy budget”, wherein the BRI for computational or cost limitations of the respective client unit encompasses the configuration data received that indicates that the server should send a stripped down toolset to the client because it has reduced power, memory, storage, or energy budget), and local metadata of the respective client unit; transmitting a toolset from the central server unit, to the one or more of the plurality of client units, the toolset being configured based on the configuration data (i.e. para. [0077], “Accordingly, the modeling framework 531 and the compile 532 can be optional components on the client computing device 500”, wherein upon receiving configuration data that shows a client device does not have enough computing power, memory, or storage, the framework toolset may be configured as optional and even be removed entirely from that client device. The examiner notes that the BRI for the toolset being configured based on the configuration data encompasses where the toolset may be configured on any of the included configuration data information instead requiring a consideration of all parts of the configuration data. Thus configuring the modeling framework to be stripped down based solely on specific configuration data indicating that a client has reduced power, memory, storage, or energy budget reads on the above limitation) such that a plurality of different types of machine learned models are creatable from the toolset at the one or more of the plurality of client units (i.e. para. [0065], Fig. 4-5, “The modelling framework 431 can be used to define and use a neural network model. As one example, the modelling framework 431 can include pre-defined APIs… The pre-defined APIs can include both lower-level APIs (e.g., activation functions, cost or error functions, nodes, edges, and tensors) and higher-level APIs (e.g., layers, convolutional neural networks, recurrent neural networks, linear classifiers, and so forth) … A data scientist can create different neural network models by using different APIs, different numbers of APIs, and interconnecting the APIs in different ways”, wherein it is noted in Fig. 5 that the modeling framework 531 and the compile 532 can be optional components on the client computing device 500. Thus the BRI for a toolset encompasses a framework transferred from the cloud service provider used by a data scientist to create a plurality of different types of neural network models to suit their various needs, that may be configured to be stripped-down due to a client units hardware limitations); receiving, from the one or more of the plurality of client units, one or more readily trained machine learned models, the one or more machine learned models being respectively created from the toolset and trained based and the respective local data by the respective one or more of the plurality of client units (i.e. para. [0075-0085], “FIG. 5 is a system diagram of an example client computing device 500 for performing incremental training of a machine learning tool 510…. The server computer can complete the incremental training by aggregating the gradients from various client devices and updating the operational parameters of the machine learning model. Thus, the incremental training can be partially performed at the client computing device 500”, wherein the BRI for local data encompasses how the client device may select their own local parameters as training data for the ML model and the server may receive the incrementally trained model as readily trained by a first user of one of the client units); and storing the one or more readily trained machine learned models received, in the central server unit (i.e. para. [0094], “application executing on the client device can update operational parameters for the server computer … such as by performing the incremental training of the neural network model and distributing the updated operational parameters to local server computer memory and/or storage”, wherein it is noted that the BRI for the central server unit encompasses the local server storage for the server computers that communicate with the plurality of client devices over the network),
While Burger teaches that a central server may receive configuration data detailing the computational limitations of a client device and subsequently send transmit a configured machine learning toolset that is stripped down based on the received configuration data, Burger may not explicitly teach that the configuration data includes
information about a field of application of a machine learning method at the respective client unit
and local metadata of the respective client unit
wherein the toolset is configured by querying at least one repository and matching the configuration data with at least one of machine learning models, tools, and toolsets in the server unit.
However, Bjork teaches
configuration data, the configuration data including information about a field of application of a machine learning method at the respective client unit (i.e. para. [0036], “Similar to any conventional machine learning operations, the model training process requires multiple iterations over the data records in the dataset, i.e. the raw data 400c, when training the data model using some appropriate computation algorithm that has been selected depending on the field of use”, wherein the raw data may indicate a field of use and thus an appropriate machine learning algorithm tool for training may be selected),
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to add the configuration data including information about a field of application of a machine learning method at the respective client unit, to Burger’s server that receives client data and transmits a tailored framework tool for training machine learning models, with how user selected configuration data includes information about a field of use for a machine learning application and then selects an appropriate algorithm in response, as taught by Bjork. One would have been motivated to combine Bjork and Burger in order save user’s time by requiring fewer steps in the process of creating a machine learning model.
While Burger-Bjork teach that a central server receives configuration data of a client, wherein the configuration data includes information about a field of application of a machine learning method at the respective client unit and computational or cost limitations of the respective client unit and sending a tailored toolkit based obtained configuration data, Burger-Bjork may not explicitly teach configuration data including information about
local metadata of the respective client unit,
wherein the toolset is configured by querying at least one repository and matching the configuration data with at least one of machine learning models, tools, and toolsets in the server unit.
However, Grzybowski teaches configuration data including information about
local metadata of the respective client unit (i.e. para. [0101-0103], “A channel reception module 908 receives selection of a channel from the user, as based on the channel display module 906. A profile generation module 910 generates a profile for the user and device based upon the selected channel and the particular capabilities of the device…the profile generation module 910 can generate a user profile based on the information included in a user key including selected portions of a manifest, claims, and user authentication information… A profile transmission module 912 transmits the profile generated at the profile generation module 910 to a remote system, such as a server configured to return a virtual user environment”, wherein a central environment server may receive client configuration data including local metadata of the client unit. Wherein the BRI for local metadata encompasses the server receiving information about user selected channel and particular capabilities of the device).
Grzybowski further teaches
transmitting a toolset from the central server unit, to the one or more of the plurality of client units, the toolset being configured based on the configuration data (i.e. para. [0103], An environment receipt module 914 receives a virtual user environment for operation at the computing device, such that the virtual user environment is customized to be user-specific and device-specific based on the information provided in the profile”, wherein the BRI for a toolset encompasses a virtual environment that selected by the remote sever and returned to the user device according to received user client data).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to add the configuration data including information about local metadata of the respective client unit, to Burger-Bjork’s server that receives client data and transmits a tailored framework tool for training machine learning models, with how a remote server receives local client metadata and transmits a tailored software package back to the local client, as taught by Grzybowski. One would have been motivated to combine Grzybowski and Burger-Bjork in order save user’s time by having virtual tools made specific to the device that the individual is using and even specific to the individual's personal requirements or settings.
While Burger-Bjork-Grzybowski teach a remote server that transmits a tailored machine learning software toolset from the server based on a received client’s configuration data that includes hardware specifications, Burger-Bjork-Grzybowski may not explicitly teach
wherein the toolset is configured by querying at least one repository and matching the configuration data with at least one of machine learning models, tools, and toolsets in the server unit.
However, Rainero teaches
wherein the toolset is configured by querying at least one repository and matching the configuration data (i.e. para. [0034-0035], Fig. 4, “In various exemplary embodiments, performing one or more server based image processing operations may include determining one or more parameters associated with at least one of the client device… the one or more server based image processing operations may include processing of the image information based on one or more of at least a client device make or model”, wherein the BRI for configuration data encompasses a client’s parameter data encompassing supported hardware. The examiner notes in para. [0026], that the sever may “receives the selection of parameters sent from the client 120”, wherein the BRI for a toolset encompasses how data is configured at the server into a matching format. It is noted that the BRI for configuration data encompasses the parameters that may be the client hardware and software capabilities and thus matching the client parameter data with an appropriate and supported data format, as configured by an application in the server’s software application library, results in a data tool being configured in a format that matches the client’s parameters) with at least one of machine learning models, tools, and toolsets in the server unit (i.e. para. [0026], Fig. 2, “the server 110 processes the image data based on client device/handset display hardware and/or software capabilities, image formats supported by the client device/handset, network bandwidth, document file format, document content, user hints and the like”, wherein the BRI for a tool or a toolset encompasses a requested document located on a server that is configured by an application that matches the client’s parameters into a matching format that is transmitted from the server).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to add wherein the toolset is configured by querying at least one repository and matching the configuration data with at least one of machine learning models, tools, and toolsets in the server unit, to Burger-Bjork- Grzybowski’s server that receives client data and transmits a tailored framework tool for training machine learning models, with a server receives client hardware parameters, finds a matching tool for the hardware parameters, and configures a data package to a format matching the clients hardware parameters, as taught by Rainero. One would have been motivated to combine server that receives and compares client configuration to a repository of matching formats and sends the appropriate match back to the client of Rainero and the server based machine learning tools personalization of Burger-Bjork-Grzybowski in order save time and transfer costs for toolsets as a server will not have to send all potential toolsets, which might exceed the client capability and transferring entire application files when only pieces need to be viewed will in general exceed reasonable transfer times and costs.
Claim 2:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 1.
Burger further teaches
wherein the one or more readily trained machine learned models received from the one or more of the plurality of client units comprise one or more incremental machine learned models, learning of the one or more incremental machine learned models being resumable by incremental machine learning (i.e. para. [0085], “The server computer can complete the incremental training by aggregating the gradients from various client devices and updating the operational parameters of the machine learning model. Thus, the incremental training can be partially performed at the client computing device 500”, wherein it is noted a model updated aver every instance of incremental training).
Claim 3:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 1.
Burger further teaches wherein the toolset includes a plurality of different untrained and trainable machine learning algorithms, trainable with the respective local data using methods of machine learning (i.e. para. [0065], “The pre-defined APIs can include… and higher-level APIs (e.g., layers, convolutional neural networks, recurrent neural networks, linear classifiers, and so forth)”, wherein it is noted that higher level APIs may include convolutional neural networks, which are a series of algorithm layers that may be trained using the data provided by the client machine).
Claim 4:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 1.
Burger teaches further comprising: downloading, to one or more of the plurality of client units, one or more cross-site machine learned models from the central server unit, the one or more cross-site machine learned models being created at client units different from the one or more of the plurality of client units that the one or more cross-site machine learned models are downloaded to (i.e. para. [0094], “the application executing on the client device can update operational parameters for the server computer and all client devices communicating with the server, such as by performing the incremental training of the neural network model and distributing the updated operational parameters to local server computer memory and/or storage and to the individual client devices”, wherein the BRI for cross-site models encompasses a first client device may update a machine learning model to server and subsequently the updated model may be distributed to all the remaining client devices part of the system).
Claim 5:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 1.
Burger teaches further comprising: receiving, from one or more of the plurality of client units, one or more updated machine learned models, the one or more updated machine learned models being machine learned models updated locally by the one or more of the plurality of client units based on the respective local data of the one or more of the plurality of client units (i.e. para. [0094], “the incremental training can be partially performed at the client device”, wherein it is noted in para. [0076], that the incremental training may be respectively based on a client’s local data, such as the retraining logic 570 which generates the updated model parameters to be stored in the model parameters 534); and storing the one or more updated machine learned models in the central server unit (i.e. para. [0094], application executing on the client device can update operational parameters for the server computer … such as by performing the incremental training of the neural network model and distributing the updated operational parameters to local server computer memory and/or storage).
Claim 6:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 5.
Burger teaches further comprising: downloading the one or more updated machine learned models to all of the plurality of client units that use previous versions of the respective one or more updated machine learned models (i.e. para. [0094], “application executing on the client device can update operational parameters for the server computer … such as by performing the incremental training of the neural network model and distributing the updated operational parameters to local server computer memory and/or storage”, wherein each local server may be updated).
Claim 7:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 1.
Burger teaches further comprising: evaluating, at the central server unit, performance of one or more of the one or more readily trained machine learned models in processing the respective local data of the respective one or more of the plurality of client units (i.e. para. [0103-106, 110], “FIG. 8 illustrates a method 800 of performing incremental training of a machine learning tool using a server computer… additional training data can be received from the edge device. The additional training data can be selected based on a measure of quality applied to an output of the machine learning tool executing at the edge device”, wherein the BRI for evaluating performance encompasses how the system can determine a predicted quality of the classification of the input data measured by a local client or edge device. Thus a server computer may perform additional incremental training based on the evaluation that a predicted quality level of the local client or edge device’s trained operation parameters are below a threshold quality level).
Claim 8:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 1.
Burger teaches further comprising: receiving, from at least one of the plurality of client units, one or more performance logs, wherein
the one or more performance logs are generated locally at a respective client unit among the at least one of the plurality of client units; each performance log of the one or more performance logs corresponds to a machine learned model available at the respective client unit (i.e. para. [0082], Fig. 5, “The quality analyzer 535 can also determine a quality of the results from the machine learning tool 510… . The quality analyzer 535 can mark the input data as data that was misclassified and the misclassified input data can be uploaded (with or without a correct label) using the upload logic 540 and server interface 542 to the server computer 110”, wherein the BRI for a performance log encompasses if the quality analyzer located at the each of client device determines a quality result of the machine learning model parameters); and each performance log of the one or more performance logs being respectively indicative of how the corresponding machine learned model performs on the respective local data of the respective client unit (i.e. para. [0080], “The quality analyzer 535 can use various techniques to determine the quality of the results from the machine learning tool 510 for a given set of input data”, wherein the examiner notes in para. [0022] that input data may be collected respectively by and at each client device); and wherein the method further includes: evaluating, at the central server unit, performance of one or more machine learned models based on the corresponding one or more received performance logs (i.e. para. [0080], the quality analyzer 535 can mark the input data as data that was misclassified and the misclassified input data can be uploaded (with or without a correct label) using the upload logic 540 and server interface 542 to the server computer 110… the server computer can perform the incremental training using the uploaded input data to generate updated model parameters that can be redistributed).
Claim 9:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 7.
Burger teaches further comprising at least one of: determining, based upon the evaluating, whether or not to deploy a machine learned model locally at the respective client unit (i.e. para. [0108], “block 850, the updated operational parameters can be transmitted to the edge device. The updated operational parameters can be used by the edge device to configure the machine learning tool”, wherein the BRI for determining, based up on evaluating encompasses how the system has evaluated that the previous model’s output was below a threshold, and subsequently initiates incremental training to update the operational parameters of the neural network model. The examiner also notes that para. [0022] cites that client devices may also be referred to as edge devices. Wherein the BRI to deploy a machine learned model encompasses sending the updated operational parameters to a client device for use in their respective neural network model); determining, based upon the evaluating, whether or not to update one or more of the machine learned models locally at the respective client units to generate corresponding up-dated machine learned models, and initiating the corresponding update (i.e. para. [0099], block 730, “in response to determining a measure of prediction quality of the output of the machine learning tool is below a threshold, incremental training of the operational parameters can be initiated using the input data as training data for the machine learning tool”, wherein it is determined that updated operation parameters will be used by the client after the predicted accuracy is found to be below a threshold); determining, based upon the evaluating, whether or not to download one or more cross-site machine learned models from the central server unit to one or more of the local sites, the one or more cross-site machine learned models being trained at client units different from the one or more of the plurality of client units downloaded to (i.e. para. [0101], “application executing on the client device can update operational parameters for the server computer and all client devices communicating with the server”, wherein it may be determined updated that operation parameters should be calculated and subsequently sent to all client devices after the predicted accuracy is found to be below a threshold); comparing, based upon the evaluating, a plurality of machine learned models available at a client unit (i.e. para. [0079], “edge devices can initially be assigned low levels of trust. If a given edge device provides training data that is determined to be useful in increasing an accuracy of the model, then the trust-level can be increased for the given edge device”, wherein based on the accuracy evaluation of the models obtained from each respective client device, the client models are compared to each other in order to determine that edge devices that have higher models accuracy will have a higher trust level); and determining, based upon the evaluating, a value of the respective local data of one or more of the plurality of client units (i.e. para. [0073], the training data from the individual edge devices can be weighted based on the trust-level of the respective edge device).
Claim 11:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 1.
Burger teaches further comprising: receiving, from one or more of the plurality of client units, one or more configuration files locally generated at the one or more of the plurality of client units, each configuration file being respectively indicative of the local configuration of the respective machine learned model at the respective client unit (i.e. para. [0082], Fig. 5, “The quality analyzer 535 can also determine a quality of the results from the machine learning tool 510… . The quality analyzer 535 can mark the input data as data that was misclassified and the misclassified input data can be uploaded (with or without a correct label) using the upload logic 540 and server interface 542 to the server computer 110”, wherein the BRI for a performance log encompasses if the quality analyzer located at the client device determines a quality result of the machine learning model parameters); and storing the one or more configuration files at the central server unit (i.e. para. [0085], The gradient values can be stored in the collected training data set 550 and/or uploaded to the server computer using the upload logic 540).
Claim 12:
Claim 12 is the device claim reciting similar limitations to claim 1 and is rejected for similar reasons.
Claim 13:
Burger teaches the central server unit of claim 12
Burger teaches. wherein the memory unit includes a plurality of different types of machine learned models from different client units of the plurality of client units (i.e. para. [0094], Fig. 4, “the application executing on the client device can update operational parameters for the server computer and all client devices communicating with the server, such as by performing the incremental training of the neural network model and distributing the updated operational parameters to local server computer memory”, wherein the plurality of client devices may upload their incremental trained models to the server computer).
Claim 14:
Burger, Bjork, Grzybowski, and Rainero teach a non-transitory computer program product comprising program elements to induce a computing unit of a system for client-specific federated learning to perform the method of claim 1.
Burger further teaches when the program elements are loaded into a memory of the computing unit (i.e. para. [0095], Fig. 5, The updated operational parameters can be stored on a computer-readable medium such as memory or a storage device of the client device).
Claim 15:
Burger, Bjork, Grzybowski, and Rainero teach a non-transitory computer-readable medium storing program elements, readable and executable by a computing unit of a system for client-specific federated learning, to perform the method of claim 1.
Burger further teaches when the program elements are executed by the computing unit (i.e. para. [0017], Any of the disclosed methods can be implemented as computer-executable instructions stored on one or more computer-readable media).
Claim 16:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 2.
Burger further teaches wherein the toolset includes a plurality of different untrained and trainable machine learning algorithms, trainable with the respective local data using methods of machine learning (i.e. para. [0065], “The pre-defined APIs can include… and higher-level APIs (e.g., layers, convolutional neural networks, recurrent neural networks, linear classifiers, and so forth)”, wherein it is noted that higher level APIs may include convolutional neural networks, which are a series of algorithm layers that may be trained using the data provided by the client machine).
Claim 17:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 2.
Burger teaches further comprising: downloading, to one or more of the plurality of client units, one or more cross-site machine learned models from the central server unit, the one or more cross-site machine learned models being created at client units different from the one or more of the plurality of client units that the one or more cross-site machine learned models are downloaded to (i.e. para. [0094], “the application executing on the client device can update operational parameters for the server computer and all client devices communicating with the server, such as by performing the incremental training of the neural network model and distributing the updated operational parameters to local server computer memory and/or storage and to the individual client devices”, wherein the BRI for cross-site models encompasses a first client device may update a machine learning model to server and subsequently the updated model may be distributed to all the remaining client devices part of the system).
Claim 18:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 2
Burger teaches further comprising: receiving, from one or more of the plurality of client units, one or more updated machine learned models, the one or more updated machine learned models being machine learned models updated locally by the one or more of the plurality of client units based on the respective local data of the one or more of the plurality of client units (i.e. para. [0094], “the incremental training can be partially performed at the client device”, wherein it is noted in para. [0076], that the incremental training may be respectively based on a client’s local data, such as the retraining logic 570 which generates the updated model parameters to be stored in the model parameters 534); and storing the one or more updated machine learned models in the central server unit (i.e. para. [0094], application executing on the client device can update operational parameters for the server computer … such as by performing the incremental training of the neural network model and distributing the updated operational parameters to local server computer memory and/or storage).
Claim 19:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 2.
Burger teaches further comprising: receiving, from one or more of the plurality of client units, one or more configuration files locally generated at the one or more of the plurality of client units, each configuration file being respectively indicative of the local configuration of the respective machine learned model at the respective client unit (i.e. para. [0088], “ FIG. 6 illustrates a method 600 of updating operational parameters of a neural network model using a client computing device… incremental training of the neural network model can be initiated using the input data as training data for the neural network model”, wherein the BRI for one or more configuration files encompasses the updated operation parameters generated by the client device as a result of the incremental training based on the client’s input data); and storing the one or more configuration files at the central server unit (i.e. para. [0095], application executing on the client device can update operational parameters for the server computer … such as by performing the incremental training of the neural network model and distributing the updated operational parameters to local server computer memory and/or storage).
Claim 20:
Burger, Bjork, Grzybowski, and Rainero teach the a non-transitory computer-readable medium storing program elements, readable and executable by a computing unit of a system for client-specific federated learning, to perform the method of claim 2.
Burger further teaches when the program elements are executed by the computing unit (i.e. para. [0017], Any of the disclosed methods can be implemented as computer-executable instructions stored on one or more computer-readable media).
Claim(s) 10 is/are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication NO. 20200265301 “Burger” in light of U.S. Patent Application Publication NO. 20130173509 “Bjork”, in light of U.S. Patent Application Publication NO. 20110145272 “Grzybowski”, and further in light of U.S. Patent Application Publication NO. 20060026511 “Rainero”, as applied to Claim 1, and further in light of U.S. Patent Application Publication NO. 20180060744 “Achin”.
Claim 10:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 1.
Burger teaches further comprising: providing, to at least one of the plurality of client units, an incremental machine learned model, the learning of the incremental machine learned model being resumed by incremental machine learning (i.e. para. [0107], “incremental training of the operational parameters can be performed using the additional training data received from the edge device to generate updated operational parameters… he updated operational parameters can be used by the edge device to configure the machine learning tool”, wherein the incrementally trained model is provided to the client devices by sending the client devices incrementally updated operational parameters that can be used by the client’s machine learning tool).
Burger, Bjork, Grzybowski, and Rainero may not explicitly teach partitioning the respective local data of the one or more of the plurality of client units into a plurality of folds; and performing one or more cross-validating operations on the incremental machine learned model across the plurality of folds to obtain an updated incremental machine learned model and an associated performance log, indicative of how the updated incremental machine learned model performs on the respective local data of the at least one client unit of the plurality of client units, wherein the one or more cross-validating operations involve the continuous further training of the incremental machine learned model by incremental machine learning.
However, Achin teaches
partitioning the respective local data of the one or more of the plurality of client units into a plurality of folds (i.e. para. [0193], To facilitate cross-validation, predictive modeling system 100 may partition the dataset (or suggest a partitioning of the dataset) into K “folds”); and performing one or more cross-validating operations on the incremental machine learned model across the plurality of folds to obtain an updated incremental machine learned model (i.e. para. [0193], Cross-validation comprises fitting a predictive model to the partitioned dataset K times, such that during each fitting, a different fold serves as the test set and the remaining folds serve as the training set”, wherein it is noted that the BRI for an updated incremental machine learned model encompasses how the K-fold cross validation changes with each iteration of cross validation, thus the machine learned model is becoming incrementally updated) and an associated performance log, indicative of how the updated incremental machine learned model performs on the local data of the at least one client unit of the plurality of client units (i.e. para. [0193], Cross-validation can generate useful information about how the accuracy of a predictive model varies with different training data), wherein the one or more cross-validating operations involve the continuous further training of the incremental machine learned model by incremental machine learning (i.e. para. [0198], “the application of a double loop of k-fold cross validation may allow predictive modeling system 100 to simultaneously achieve five important goals: (1) tuning complex models with many hyper-parameters, (2) developing informative derived features, (3) tuning a blend of two or more models”, wherein the BRI for incremental machine learning encompasses how the K-fold cross validation may incrementally change parameters by tuning model parameters).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to add partitioning the local data of the at least one client unit of the plurality of client units into a plurality of folds; and performing one or more cross-validating operations on the incremental machine learned model across the plurality of folds to obtain an updated incremental machine learned model and an associated performance log, indicative of how the updated incremental machine learned model performs on the local data of the at least one client unit of the plurality of client units, wherein the cross-validating operations involve the continuous further training of the incremental machine learned model by incremental machine learning, to Burger-Bjork-Grzybowski-Rainero’s distributed and incrementally trained machine learning models, with how a machine learning model may be cross validated to further train the model and wherein the cross-validation obtains a log of information that is indicative of how the accuracy of the model varies, as taught by Achin. One would have been motivated to combine Achin and Burger-Bjork-Grzybowski-Rainero in order to give a user more insight into a machine learning model as k-fold cross validation is used to flag problems like overfitting or selection bias and gives insights on how the model will generalize to an independent dataset.
Claim(s) 22 is/are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication NO. 20200265301 “Burger” in light of U.S. Patent Application Publication NO. 20130173509 “Bjork”, in light of U.S. Patent Application Publication NO. 20110145272 “Grzybowski”, and further in light of U.S. Patent Application Publication NO. 20060026511 “Rainero”, as applied to Claim 1 above, and further in light of U.S. Patent Application Publication NO. 20180176097 “Russell”.
Claim 22:
Burger, Bjork, Grzybowski, and Rainero teach the method of claim 1.
While Burger-Bjork-Grzybowski-Rainero teaches receiving configuration data at a central server unit and configuring a machine learning toolset based on the received configuration data, Burger may not explicitly teach further comprising:
storing the configuration data of the one or more of the plurality of client units at the central server unit, and configuring the toolset by obtaining machine learning tools corresponding to the configuration data stored for a respective client unit.
However, Russell teaches
storing the configuration data of the one or more of the plurality of client units at the central server unit (i.e. para. [0092], “The server 220 can inquire from the device 240 what type it is and choose a design implementation and layout suitable to the known limitations of the device 240”, wherein the BRI for storing configuration data encompasses the server stored and known limitations of the one or more client devices), and configuring the toolset by obtaining ( (i.e. para. [0119], “The cloud application 210 can interrogate the client application 230 to determine the device 240 capabilities or limitations and choose a layout that conforms to the devices 240 capabilities or limitations. In addition, customers can choose features that can be “dropped” on lower end devices”, wherein the sent over tools are configured based on the data about the limitations of the hardware of the client).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to storing the configuration data of the one or more of the plurality of client units at the central server, to Burger-Bjork-Grzybowski-Rainero’s distributed and incrementally trained machine learning models, with storing the configuration data of the one or more of the plurality of client units at the central server, and configuring the toolset by obtaining machine learning tools corresponding to the configuration data stored for a respective client unit, as taught by Russell. One would have been motivated to combine Russell and Burger-Bjork-Grzybowski-Rainero in order to send an appropriate performing software service for a user’s optimal machine purpose.
Claim(s) 23 is/are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication NO. 20200265301 “Burger” in light of U.S. Patent Application Publication NO. 20130173509 “Bjork”, in light of U.S. Patent Application Publication NO. 20110145272 “Grzybowski”, in light of U.S. Patent Application Publication NO. 20060026511 “Rainero”, and further in light of U.S. Patent Application Publication NO. 20180176097 “Russell”, as applied to Claim 22 above, and further in light of U.S. Patent Application Publication NO. 20170277853 “Carlson”.
Claim 23:
Burger, Bjork, Grzybowski, Rainero, and Russell teach the method of claim 22.
Bjork further teaches
Wherein the toolset is configured with one or more algorithms selected based on the configuration data, wherein the toolset includes one or more algorithms selected based on the configuration data (para. [0036], “Similar to any conventional machine learning operations, the model training process requires multiple iterations over the data records in the dataset, i.e. the raw data 400c, when training the data model using some appropriate computation algorithm that has been selected depending on the field of use”, wherein the raw data may indicate a field of use and thus an appropriate machine learning algorithm tool for training may be selected).
While Bjork teaches selecting a machine learning algorithm for a model based on configuration data related to a field of use, Burger, Bjork, Grzybowski, Rainero, and Russell may not explicitly teach
the one or more algorithms including untrained and trainable machine learned algorithms.
However, Carlson teaches
the one or more algorithms including untrained and trainable machine learned algorithms (i.e. para. [00026], “After all inputs are added to the system, the initial prediction algorithm 212 along with performance metrics and feature importance is created. This information is presented to the user to review and determine if the performance and number of features are acceptable 214. The user is given the option to eliminate features 216 (this allows the algorithm to be deployed when fewer features are measured). If features are selected for elimination, the prediction algorithm training 212 is repeated to create a new algorithm with remaining features. This process iterates until user is satisfied with results or other prediction accuracy metrics are attained”, wherein the BRI for an untrained algorithm encompasses an initially created algorithm which is trainable over a series of iterations by a user).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to add the one or more algorithms including untrained and trainable machine learned algorithms, to Burger-Bjork-Grzybowski-Rainero-Russel’s calling of appropriate algorithms depending on configuration data, how created algorithms may be untrained and trainable, as taught by Carlson. One would have been motivated to combine Carlson and Burger-Bjork-Grzybowski-Rainero-Russel in order to more easily create a new action plan and is transparent about the new system being deployed.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
U.S. Patent Application Publication NO. 20190110753 “Zhang”, teaches in para. [0005], The weights for the lower layer(s) are then frozen, while the weights of the upper layer(s) are retrained using images from the relevant domain to identify output according to the desired diagnosis (e.g. identification or prediction of specific ophthalmic diseases or conditions). This approach allows the classifier to recognize distinguishing features of specific categories of images (e.g. images of the eye) far more quickly using significantly fewer training images and while requiring substantially less computational power.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DAVID H TAN whose telephone number is (571)272-7433. The examiner can normally be reached M-F 7:30-4:30.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Cesar Paula can be reached at (571) 272-4128. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/D.T./Examiner, Art Unit 2145
/CESAR B PAULA/Supervisory Patent Examiner, Art Unit 2145
Prosecution Insights