DETAILED ACTION
This Office Action is in response to communications filed on February, 2nd, 2026 for Application No. 18/164,518, in which claims 1-20 are presented for examination. The amendments filed on February, 2nd, 2026 have been entered, where claims 1, 7-9, 15-17, and 19-20 are amended.
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 .
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-2, 9-10, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Zhou et al. (hereinafter Zhou) (“Two-Layer Federated Learning With Heterogeneous Model Aggregation for 6G Supported Internet of Vehicles”) in view of Han et al. (hereinafter Han) (“FedMes: Speeding Up Federated Learning With Multiple Edge Servers”) and Feuz et al. (hereinafter Feuz) (“Ranking and automatic selection of machine learning models Abstract”).
Regarding Claim 1, Zhou teaches a method for vehicular assisted hierarchical federated learning, the method comprising (Pg. 5309, Col. 1, Para. 2, “In this study, we propose a two-layer federated learning model based on convolutional neural network (TFL-CNN), which makes use of the local and global contexts of individual vehicles and RSUs to perform hierarchical and heterogeneous model selection and aggregation at the edge and cloud level”; Pg. 5313, Col. 1, Para. 4, “Experiments are conducted and discussed to demonstrate the usefulness and effectiveness of the proposed method comparing with several baseline methods”):
responsive to joining a hierarchical federated learning network, obtaining vehicular system conditions of a vehicle (Pg. 5311, Col. 1, Para. 1, “RSUs are the middle brokers to collect and aggregate not only learning parameters but also contextual information, such as vehicle locations and navigation direction, from all the connected vehicles within the coverage to facilitate parameter aggregation”; Pg. 5311, Col. 2, Para. 7, “the vehicular contextual information, such as location and navigation information, can be directly acquired by RSUs via 6G technologies [37] to conduct the proposed weighted aggregation in RSU”, where “vehicular contextual information, such as location and navigation information” is within the broadest reasonable interpretation of vehicular system conditions, which is “acquired by the RSUs” responsive to a vehicle joining the hierarchical federated learning network and being “connected” within the “RSUs” “coverage”);
exchanging data between the vehicle and a plurality of edge servers of the hierarchical federated learning network (Pg. 5310, Col. 2, Fig. 1, “Two-Layer Federated Learning Framework in 6G Supported Vehicular Networks”, where the “Two-Layer Federated Learning Framework” is a hierarchical federated learning network, where data is exchanged, as indicated by the lower-level green and blue arrows, between the vehicles, “Vehicles”, and a plurality of edge servers, the plurality of “RSU[s]”, see Pg. 5310-5311, Col. 2-1, Para. 4-1, “In the middle layer, each RSU has limited caching and computing capabilities, and is responsible for supervising all the interconnected vehicles in its coverage . . . RSUs are the middle brokers to collect and aggregate . . . from all the connected vehicles within the coverage . . . RSUs then communicate with the central cloud server”, where “RSUs” are servers because the provide services to network clients, the “connected vehicles”; and where a given “vehicle” exchanges data directly between the “connected” “RSU” and indirectly between the other “RSUs” through the “central cloud server”, see also Pg. 5310, Col. 2, Fig. 1)
according to a vehicle-to-edge server association protocol that is based on the vehicular system conditions (Pg. 5310, Col. 2, Fig. 1 and Pg. 5313, Col. 1, Fig. 3, “4: for each data owner vi ∈ V do 5: if vi supervised by rj do . . . 8: Submit the local training parameter wi (t) to rj”, where the exchange of data is based on the vehicle-to edge server association protocols of clustered “data owner[s] vi” “supervised by rj” “RSU[s]”, where vehicular system conditions, “contextual information, such as vehicle locations”, determines which “RSU” association protocol the “vehicles” will be selected for, to be “connected” to as a supervisor see Pg. 5311, Col. 1, Para. 1, “RSUs are the middle brokers to collect and aggregate not only learning parameters but also contextual information, such as vehicle locations and navigation direction, from all the connected vehicles within the coverage to facilitate parameter aggregation”);
identifying a machine learning model for the vehicle . . . [and providing the identified model parameters to] the vehicle, wherein identifying the machine learning model further comprises . . . [performing operations] at the vehicle . . . using the data acquired by the vehicle . . . [to select] the machine learning model for the vehicle . . . (Pg. 5313, Col. 1, Fig. 3, “13: Calculate the global parameters w(t+1) for model M . . . 15: Broadcast w(t+1) to the network” and Pg. 5311, Col. 1, Para. 1, “the updated model parameters will be dispatched from the central cloud server to RSUs and then to individual vehicles”, where the “updated” “global . . . model M” is identified at the “central cloud server” and provided to the “individual vehicles”, which comprises performing operations at the vehicle using the data acquired by the vehicle, “computed via local training by vi”, to select the machine learning model for the vehicle, generate “parameters” for “secure aggregation” to select “Global Updates”, see Pg. 5312, Fig. 2, “Architecture of TFL-CNN” and Pg. 5311, Col. 2, Para. 1, “we assume data owners are secure against a curious RSUs, which means an RSU cannot compromise the privacy of data owners [38]. To achieve this, parameters computed via local training by vi need to be encrypted and then sent to the corresponding RSU which conducts secure aggregation without exposing privacy information about any data owners”); and
at least one of: training the identified machine learning model to perform a task using the data acquired by the vehicle to produce a locally trained machine learning model; and applying the data acquired by the vehicle to the identified machine learning model to perform a task (Pg. 5311, Col. 2, Para. 3, “xi is the input samples of a data owner vi, the CNN model, which is introduced to perform an object detection task (e.g., traffic sign recognition, pedestrian detection, or object avoidance), can be represented as the hypothesis h(xi, ω), and trained locally by the data owner vi”; see also Pg. 5311, Col. 1, Para. 4, “any individual data owner vi in this framework can keep its own data di and train the object detection model locally” and Pg. 5311, Col. 1, Para. 2, “each vehicle generates raw data including both the captured photos/videos by built-in camera and the contextual information (e.g., GPS locality data, driving information, etc.). The computation capability of individual vehicle is able to support a relatively light computing task (e.g., training a learning model for object detection or road sign recognition)”).
Zhou does not explicitly disclose . . . from a plurality of machine learning models hosted on the plurality of edge servers using data acquired by . . . for each of the plurality of machine learning models, determining a model-specific accuracy measure . . . , and selecting, as . . . , one of the plurality of machine learning models hosted by an edge server of the plurality of edge servers that yields a highest accuracy measure . . . .
However, Han teaches . . . [identifying a machine learning model for the client] from a plurality of machine learning models hosted on the plurality of edge servers using data acquired by [the client]. . . and selecting . . . one of the plurality of machine learning models hosted by an edge server of the plurality of edge servers (Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where “clients” receive “multiple models” from the plurality of models hosted by the plurality of “different Ess” and the multiple models are identified, which requires that a machine learning model be identified and selected, based on whether the “clients [are] in the overlapping areas”; Pg. 3872, Col. 2, Para. 3, “We call this region in which the client can reliably communicate with multiple ESs overlapping cell area”, where a “client” is determined to be in an “overlapping cell area” “communication” data acquired by the “client” and from “multiple Ess”).
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the identifying a machine learning model for a vehicle, wherein identifying the machine learning model further comprises performing operations at the vehicle using the data acquired by the vehicle to select the machine learning model for the vehicle, and providing the identified model parameters to the vehicle of Zhou with identifying a machine learning model for the client from a plurality of machine learning models hosted on the plurality of edge servers using data acquired by the client and selecting one of the plurality of machine learning models hosted by an edge server of the plurality of edge servers of Han in order to significantly reduce training time while maintaining model synchronization across edge server clusters by sending and receiving models between multiple vehicles and edge servers (Han, Pg. 3870, Col. 1, Abstract, “the proposed scheme does not require costly communications with the central cloud server (located at the higher tier of edge servers) for model synchronization, significantly reducing the overall training time compared to the conventional cloud-based FL systems. Extensive experimental results show remarkable performance gains of our scheme compared to existing methods”).
Additionally, Feuz teaches . . . for each of the plurality of machine learning models, determining a model-specific accuracy measure [using application-specific data] . . . (Pg. 2, Para. 1, “the present disclosure is directed to an API for ranking . . . competing machine learning models that can perform a particular task . . . from a registry of machine learning models . . . such that the application specified . . . performance requirements are met”, where, for each of the plurality of machine learning models, “a registry of machine learning models”, a model-specific performance is determined, “ranking . . . competing machine learning models that can perform a particular task”; see also Pg. 4, Para. 1, “This disclosure describes an application programming interface (API) that enables software applications to automatically select a machine learning model that optimally suits the application-specific data and problem statement out of several available models from service providers”, where “application-specific data” is used; see also Pg. 6-7, Para. -1, “an API function such as lastRecognitionWasCorrect(boolean correctness) can be used as a feedback signal in order to guide selection of models to process future queries from the same app” and Pg. 8, Para. 2, “A ranking mechanism can be implemented in several ways, e.g., A neural network can be used for predicting quality of a certain model on given samples learned from the user feedback, e.g., the neural network is provided a (model, sample) pair as input and is trained to produce an estimated quality score as output”, where performance as determined by “correctness” “on given samples” “to produce an estimated quality score” is within the broadest reasonable interpretation of an accuracy measure)
[and selecting] as [the machine learning model for the local device, one of the plurality of machine learning models] . . . that yields a highest accuracy measure . . . (Pg. 2, Para. 1, “the present disclosure is directed to an API for ranking and automatic selection from competing machine learning models that can perform a particular task . . . The one or more models are selected from a registry of machine learning models, their task specialties, cost, and performance, such that the application specified cost and performance requirements are met” and Pg. 6, Para. 2, “An app developer can use the API to select one of the pre-trained ML models from a service provider to perform the tasks. The selected models can run . . . on the same device as the app, e.g., a consumer device”, where a machine learning model is selected from a plurality of models, “The one or more models are selected from a registry of machine learning models”, for a local device, “The selected models can run . . . on the same device as the app, e.g., a consumer device”, where “the application specified cost and performance requirements are met”, such that, instances where “performance requirements” of “the application” are dispositive with regard to the “specified cost”, the model with the highest “performance”, which as discussed above, is an accuracy measure, will be selected).
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the identifying a machine learning model for the vehicle from a plurality of machine learning models hosted on the plurality of edge servers using data acquired by the vehicle and selecting the machine learning model for the vehicle from a plurality of machine learning models hosted by an edge server of the plurality of edge servers of Zhou in view of Han with the determining, for each of the plurality of machine learning models, a model-specific accuracy measure using application-specific data and selecting as the machine learning model for the local device, one of the plurality of machine learning models that yields a highest accuracy measure of Feuz in order to automatically select and use the model that is most suited for the problem specific to their software applications (Feuz, Pg. 3, Para. 2, “In situations where competing machine learning models are available, e.g., in a marketplace of models, developers that wish to employ machine learning in their software applications do not currently have a mechanism to automatically select and use the model that is most suited for the problem specific to their software applications”) without requiring significant expenditures of resources or specific technical expertise (Feuz, Pg. 6, Para. 2, “As explained previously, the API enables software applications from application developers to automatically select one or more models from the registered ML models that is suited to the specific problem, e.g., based on quality and cost. For example, many app developers may not have the resources, data or expertise to train a machine learning model for tasks of interest, e.g., tasks that need to be performed to provide the app functionality. An app developer can use the API to select one of the pre-trained ML models from a service provider to perform the tasks. The selected models can run in a remote computer, e.g., a cloud-based server, or on the same device as the app, e.g., a consumer device”).
Regarding Claim 2, Zhou in view of Han and Feuz teach the method of claim 1, further comprising:
selecting the vehicle-to-edge server association protocol from a plurality of vehicle-to-edge server association protocols using the vehicular system conditions (Zhou, Pg. 5310, Col. 2, Fig. 1 and Zhou, Pg. 5313, Col. 1, Fig. 3, “4: for each data owner vi ∈ V do 5: if vi supervised by rj do . . . 8: Submit the local training parameter wi (t) to rj”, where the exchange of data is based on the vehicle-to edge server association protocols of clustered “data owner[s] vi” “supervised by rj” “RSU[s]”, where vehicular system conditions, “contextual information, such as vehicle locations”, determines which “RSU” association protocol the “vehicles” will be selected for, to be “connected” to as a supervisor see Zhou, Pg. 5311, Col. 1, Para. 1, “RSUs are the middle brokers to collect and aggregate not only learning parameters but also contextual information, such as vehicle locations and navigation direction, from all the connected vehicles within the coverage to facilitate parameter aggregation”).
Regarding Claim 9, Zhou teaches a vehicle, comprising: a communication circuit configured to exchange communications with edge servers of a hierarchical federated learning network; a memory storing instructions; and one or more processors communicably coupled to the memory and configured to execute the instructions to (Zhou, Pg. 5310, Col. 2, Fig. 1; Zhou, Pg. 5311, Col. 1, Para. 2, “each vehicle generates raw data including both the captured photos/videos by built-in camera and the contextual information (e.g., GPS locality data, driving information, etc.). The computation capability of individual vehicle is able to support a relatively light computing task (e.g., training a learning model for object detection or road sign recognition)”; Zhou, Pg. 5311, Col. 2, Para. 1, “parameters computed via local training by vi need to be encrypted and then sent to the corresponding RSU which conducts secure aggregation without exposing privacy information about any data owners”, where the “raw data” “generat[ion]” and “training a learning model” require a memory to store program instructions executed by processors; and where a communications hardware, that is within the broadest reasonable interpretation of a communication circuit, is required for the “sen[ding]” of “parameters” to “corresponding RSUs”) . . . .
The remaining limitations are substantially the same as limitations of Claim 1, therefore it is rejected under the same rationale.
Regarding Claim 10, the additional elements of the dependent claim are substantially the same as limitations of Claim 2, therefore it is rejected under the same rationale.
Regarding Claim 17, Zhou in view of Han and Feuz teach a server of a hierarchical federated learning network, the server comprising (Zhou, Pg. 5309, Col. 1, Para. 2, “In this study, we propose a two-layer federated learning model based on convolutional neural network (TFL-CNN), which makes use of the local and global contexts of individual vehicles and RSUs to perform hierarchical and heterogeneous model selection and aggregation at the edge and cloud level”; Zhou, Pg. 5310, Col. 2, Fig. 1, “Two-Layer Federated Learning Framework in 6G Supported Vehicular Networks”, where the “Two-Layer Federated Learning Framework” is a hierarchical federated learning network, and where the plurality of “RSU[s]” are servers, see Zhou, Pg. 5310-5311, Col. 2-1, Para. 4-1, “In the middle layer, each RSU has limited caching and computing capabilities, and is responsible for supervising all the interconnected vehicles in its coverage . . . RSUs are the middle brokers to collect and aggregate . . . from all the connected vehicles within the coverage . . . RSUs then communicate with the central cloud server”, where “RSUs” are servers because the provide services to network clients, the “connected vehicles”):
a communication circuit configured to exchange communications with at least one vehicle of a hierarchical federated learning network (Zhou, Pg. 5311, Col. 1, Para. 1, “the updated model parameters will be dispatched from the central cloud server to RSUs and then to individual vehicles”, where hardware, that is within the broadest reasonable interpretations of a communications server is required for “dispatch[ing]” of “updated model parameters” from the “RSUs . . . to individual vehicles”; Zhou, Pg. 5310, Col. 2, Fig. 1, “Two-Layer Federated Learning Framework in 6G Supported Vehicular Networks”, where the “Vehicles” are part of a hierarchical federated learning network);
a memory storing instructions and a plurality of machine learning models; and one or more processors communicably coupled to the memory and configured to execute the instructions to (Zhou, Pg. 5310-5311, Col. 2-1, Para. 4-1, “In the middle layer, each RSU has limited caching and computing capabilities, and is responsible for supervising all the interconnected vehicles in its coverage . . . RSUs are the middle brokers to collect and aggregate . . . from all the connected vehicles within the coverage”, where the “computing”, “supervising”, and “caching” require a memory storing instructions executed by a processor; Zhou, Pg. 5311, Col. 1, Para. 1, “the updated model parameters will be dispatched from the central cloud server to RSUs and then to individual vehicles”, where the transmission from the “RSUs” “to individual vehicles” requires the memory to store “model parameters”, where, in view of Han, are a plurality of machine learning “models”, see Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different Ess”; see also Feuz, Pg. 6, Para. 2, “An app developer can use the API to select one of the pre-trained ML models from a service provider to perform the tasks. The selected models can run . . . on the same device as the app, e.g., a consumer device”):
exchange data with the at least one vehicle of the hierarchical federated learning network (Zhou, Pg. 5310, Col. 2, Fig. 1, “Two-Layer Federated Learning Framework in 6G Supported Vehicular Networks”, where the “Two-Layer Federated Learning Framework” is a hierarchical federated learning network, where data is exchanged, as indicated by the lower-level green and blue arrows, between the vehicles, “Vehicles”, and a plurality of edge servers, the plurality of “RSU[s]”, see Zhou, Pg. 5310-5311, Col. 2-1, Para. 4-1, “In the middle layer, each RSU has limited caching and computing capabilities, and is responsible for supervising all the interconnected vehicles in its coverage . . . RSUs are the middle brokers to collect and aggregate . . . from all the connected vehicles within the coverage . . . RSUs then communicate with the central cloud server”)
according to a vehicle-to-edge server association protocol selected based on vehicular system conditions of the at least one vehicle (Zhou, Pg. 5310, Col. 2, Fig. 1 and Zhou, Pg. 5313, Col. 1, Fig. 3, “4: for each data owner vi ∈ V do 5: if vi supervised by rj do . . . 8: Submit the local training parameter wi (t) to rj”, where the exchange of data is based on the vehicle-to edge server association protocols of clustered “data owner[s] vi” “supervised by rj” “RSU[s]”, where vehicular system conditions, “contextual information, such as vehicle locations”, determines which “RSU” association protocol the “vehicles” will be selected for, to be “connected” to as a supervisor see Zhou, Pg. 5311, Col. 1, Para. 1, “RSUs are the middle brokers to collect and aggregate not only learning parameters but also contextual information, such as vehicle locations and navigation direction, from all the connected vehicles within the coverage to facilitate parameter aggregation”);
transmit, to the at least one vehicle, an identification selecting one of the plurality of machine learning models as a machine learning model for the vehicle (Zhou, Pg. 5313, Col. 1, Fig. 3, “13: Calculate the global parameters w(t+1) for model M . . . 15: Broadcast w(t+1) to the network” and Zhou, Pg. 5311, Col. 1, Para. 1, “the updated model parameters will be dispatched from the central cloud server to RSUs and then to individual vehicles”, where the “updated” “global . . . model M” is identified at the “central cloud server” and transmitted as an identification to the “individual vehicles”; see also Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where “clients” receive “multiple models” from the plurality of models hosted by the plurality of “different Ess” and the multiple models are identified, which requires that a machine learning model be identified and selected, based on whether the “clients [are] in the overlapping areas”; Han, Pg. 3872, Col. 2, Para. 3, “We call this region in which the client can reliably communicate with multiple ESs overlapping cell area”, where a “client” is determined to be in an “overlapping cell area” “communication” data acquired by the “client” and from “multiple Ess”),
wherein the selected machine learning model corresponds to a highest model-specific accuracy measure computed at the vehicle using the data acquired by the vehicle (Zhou, Pg. 5312, Fig. 2, “Architecture of TFL-CNN” and Zhou, Pg. 5311, Col. 2, Para. 1, “we assume data owners are secure against a curious RSUs, which means an RSU cannot compromise the privacy of data owners [38]. To achieve this, parameters computed via local training by vi need to be encrypted and then sent to the corresponding RSU which conducts secure aggregation without exposing privacy information about any data owners”, which comprises performing operations at the vehicle using the data acquired by the vehicle, “computed via local training by vi”, to select the machine learning model for the vehicle, generate “parameters” for “secure aggregation” to select “Global Updates; see also Feuz, Pg. 2, Para. 1, “the present disclosure is directed to an API for ranking . . . competing machine learning models that can perform a particular task . . . from a registry of machine learning models . . . such that the application specified . . . performance requirements are met”, where, for each of the plurality of machine learning models, “a registry of machine learning models”, a model-specific performance is determined, “ranking . . . competing machine learning models that can perform a particular task”, where, in view of Feuz, the computing is for the selection of the machine learning model corresponding to a highest model-specific accuracy measure Feuz, Pg. 4, Para. 1, “This disclosure describes an application programming interface (API) that enables software applications to automatically select a machine learning model that optimally suits the application-specific data and problem statement out of several available models from service providers”, where “application-specific data” is used; see also Feuz, Pg. 6-7, Para. -1, “an API function such as lastRecognitionWasCorrect (boolean correctness) can be used as a feedback signal in order to guide selection of models to process future queries from the same app” and Feuz, Pg. 8, Para. 2, “A ranking mechanism can be implemented in several ways, e.g., A neural network can be used for predicting quality of a certain model on given samples learned from the user feedback, e.g., the neural network is provided a (model, sample) pair as input and is trained to produce an estimated quality score as output”, where performance as determined by “correctness” “on given samples” “to produce an estimated quality score” is within the broadest reasonable interpretation of an accuracy measure);
based on the vehicle-to-edge server association, receive a model trained locally by the at least one vehicle using data acquired by the vehicle (Zhou, Pg. 5310, Col. 2, Fig. 1 and Zhou, Pg. 5313, Col. 1, Fig. 3, “4: for each data owner vi ∈ V do 5: if vi supervised by rj do . . . 8: Submit the local training parameter wi (t) to rj”, where the exchange of data is based on the vehicle-to edge server association protocols of clustered “data owner[s] vi” “supervised by rj” “RSU[s]”, where vehicular system conditions, “contextual information, such as vehicle locations”, determines which “RSU” a “vehicles” will be “connected” to as a supervisor, see Zhou, Pg. 5311, Col. 1, Para. 1, “RSUs are the middle brokers to collect and aggregate not only learning parameters but also contextual information, such as vehicle locations and navigation direction, from all the connected vehicles within the coverage to facilitate parameter aggregation”; Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”; Han, Pg. 3872, Col. 2, Para. 1, “Now each client sends the updated model to the PS, and the PS aggregates the model”);
and aggregate the selected machine learning model and the locally trained model to generate an aggregate machine learning model (Zhou, Pg. 5311, Col. 1, Para. 1, “In the middle layer, each RSU has limited caching and computing capabilities, and is responsible for supervising all the interconnected vehicles in its coverage . . . RSUs are the middle brokers to collect and aggregate . . . from all the connected vehicles within the coverage . . . RSUs then communicate with the central cloud server”; see also Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different Ess”; Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”; and Feuz, Pg. 6, Para. 2, “An app developer can use the API to select one of the pre-trained ML models from a service provider to perform the tasks. The selected models can run . . . on the same device as the app, e.g., a consumer device”).
The reasons of obviousness have been discussed in regard to the rejection of claim 1 above and remain applicable here.
Claims 3, 5, 11, and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Zhou in view of Han, Feuz, and Dai et al. (hereinafter Dai) (“Joint Offloading and Resource Allocation in Vehicular Edge Computing and Networks”).
Regarding Claim 3, Zhou in view of Han and Feuz teach the method of claim 1, wherein vehicular system conditions comprises . . . [location information] of the vehicle . . . and . . . [navigation information] of the vehicle (Zhou, Pg. 5311, Col. 1, Para. 1, “RSUs are the middle brokers to collect and aggregate not only learning parameters but also contextual information, such as vehicle locations and navigation direction, from all the connected vehicles within the coverage to facilitate parameter aggregation”).
Zhou in view of Han do not explicitly disclose . . . at least one of: computational resources . . . a privacy requirement settings . . . .
However, Dai teaches . . . [a vehicular edge computing method, wherein vehicular system conditions comprises] . . . at least one of: computational resources of the vehicle and a privacy requirement settings of the vehicle (Pg. 2, Col. 2, Para. 7, “Let fi denote the computational resource of vehicle i, which varies for different users and can be obtained through offline measurement [13]”; see generally Pg. 1, Col. 1, Abstract, “Vehicular Edge Computing (VEC) is a new computing paradigm with a high potential to improve vehicular services by offloading computation-intensive tasks to the VEC servers”).
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the method for hierarchical federated learning with vehicles and edge servers, wherein vehicular system conditions of a vehicle are used as a basis for association protocols of Zhou in view of Han and Feuz with the vehicular edge computing method, wherein vehicular system conditions comprise computational resources of the vehicle of Dai in order to utilize the computational resources of VEC servers, while not overloading the VEC servers with tasks that the vehicles have sufficient computational resources to complete (Dai, Pg. 1, Col. 1, Abstract, “Vehicular Edge Computing (VEC) is a new computing paradigm with a high potential to improve vehicular services by offloading computation-intensive tasks to the VEC servers. Nevertheless, as the computation resource of each VEC server is limited, offloading may not be efficient if all vehicles select the same VEC server to offload their tasks. To address this problem, in this paper, we propose offloading with resource allocation. We incorporate the communication and computation to derive the task processing delay. We formulate the problem as a system utility maximization problem, and then develop a low-complexity algorithm to jointly optimize offloading decision and resource allocation. Numerical results demonstrate the superior performance of our Joint Optimization of Selection and Computation (JOSC) algorithm compared to state of the art solutions”), which will allow local training decisions to be based on the specific computational power of the vehicle in the federated learning network (Zhou, Pg. 5309, Col. 1, Para. “With small quantity of
data (or low data quality depending on the camera resolution) and varying computational power of individual vehicles, local training is usually limited by its accuracy”).
Regarding Claim 5, Zhou in view of Han, Feuz, and Dai teach the method of claim 1, further comprising: determining that the vehicle comprises insufficient computational resources (Dai, Pg. 2, Col. 1, Para. 5-6, “Each task can either be offloaded to a selected VEC server to process, or be executed locally at the vehicle . . . If vehicle i chooses to offload task Di to a selected VEC server to process”, where the decision to “offload task Di” is determined based on whether the vehicle has sufficient computational resources to avoid “bottlenecks” that compromise “Quality of Service”, see Dai Pg. 1, Col. 1, Para. 2, “resource-constrained vehicles can be strained by computation intensive applications, resulting in bottlenecks and making it challenging for the vehicles to ensure the required level of Quality of Service”, with reference to “maximum allowed latency”, see Dai, Pg. 3, Col. 1, Para. 5, “we formulate the joint offloading and resource allocation scheme as an optimization problem . . . The first constraint (8b) guarantees that the task processing time cannot exceed the maximum allowed latency”)
for storing and running each machine learning model of the plurality of machine learning models (Zhou, Pg. 5311, Col. 1, Para. 4, “any individual data owner vi in this framework can keep its own data di and train the object detection model locally”, which, in view of Han, are they plurality of models, see Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas
receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where the “client” stores and runs the plurality of models, collectively, as the “averaged model”; see also Zhou, Pg. 5309, Col. 1, Para. “With small quantity of data (or low data quality depending on the camera resolution) and varying computational power of individual vehicles, local training is usually limited by its accuracy”),
wherein selecting the vehicle-to-edge server association protocol from the plurality of vehicle-to-edge server association protocols is based on the determination (Zhou, Pg. 5310, Col. 2, Fig. 1 and Zhou, Pg. 5313, Col. 1, Fig. 3, “4: for each data owner vi ∈ V do 5: if vi supervised by rj do . . . 8: Submit the local training parameter wi (t) to rj”, where the exchange of data is based on the vehicle-to edge server association protocol of clustered “data owner[s] vi” “supervised by rj” “RSU[s]”, where vehicular system conditions, “contextual information, such as vehicle locations”, determines which “RSU” a “vehicles” will be “connected” to as a supervisor see Zhou, Pg. 5311, Col. 1, Para. 1, “RSUs are the middle brokers to collect and aggregate not only learning parameters but also contextual information, such as vehicle locations and navigation direction, from all the connected vehicles within the coverage to facilitate parameter aggregation”, where in view of Dai, the contextual information includes the offloading decision of the “vehicle”, see Dai, Pg. 2, Col. 2, Para. 7, “Let fi denote the computational resource of vehicle i, which varies for different users and can be obtained through offline measurement [13]”).
The reasons of obviousness have been discussed in the rejection of claim 1, in regard to the combination of Zhou with Han, and the rejection of claim 3, in regard to the combination of Zhou and Han with Dai, and remain applicable here.
Regarding Claim 11, the additional elements of the dependent claim are substantially the same as limitations of Claim 3, therefore it is rejected under the same rationale.
Regarding Claim 13, the additional elements of the dependent claim are substantially the same as limitations of Claim 5, therefore it is rejected under the same rationale.
Claims 4, 6, 12, 14, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Zhou in view of Han, Feuz, Dai, and Zhao et al. (hereinafter Zhao) (“Privacy-Aware Federated Learning for Page Recommendation”).
Regarding Claim 4, Zhou in view of Han, Feuz, and Dai teach the method of claim 3, wherein vehicular system conditions comprises . . . [location information, navigation information,] and computational resources of the vehicle (Zhou, Pg. 5311, Col. 1, Para. 1, “RSUs are the middle brokers to collect and aggregate not only learning parameters but also contextual information, such as vehicle locations and navigation direction, from all the connected vehicles within the coverage to facilitate parameter aggregation”, which, in view of Dai, includes computational resources of the vehicle, see Dai, Pg. 2, Col. 2, Para. 7, “Let fi denote the computational resource of vehicle i, which varies for different users and can be obtained through offline measurement [13]”).
The reasons of obviousness have been discussed in regard to the rejection of claim 3 above and remain applicable here.
Zhou in view of Han, Feuz, and Dai do not explicitly disclose . . . privacy requirements . . . .
However, Zhao teaches [a federated learning method, where entity conditions comprise] . . . privacy requirements . . . (Pg. 1071, Col. 1, Abstract, “We propose Fed4Rec, a privacy-preserving framework for page recommendation based on federated learning (FL) and model-agnostic meta-learning (MAML), which allows machine learning models to train on data collected from both public users, who share data with the server, and private users, who do not share data with the server”).
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the method for hierarchical federated learning with vehicles and edge servers, wherein vehicular system conditions of a vehicle, which include computational resources, are used as a basis for association protocols of Zhou in view of Han, Feuz, and Dai with the federated learning method, where entity conditions comprise privacy requirements of Zhao or order to train models on all user data by adjusting data sharing protocols depending on user requirements (Zhao, Pg. 1071, Col. 1, Abstract, “Fed4Rec . . . allows machine learning models to train on data collected from both public users, who share data with the server, and private users, who do not share data with the server. Fed4Rec enables recommendations for both public users, computed at the server, and private users, computed at their local devices”), which improves model accuracy (Zhao, Pg. 1071, Col. 1, Abstract, “The results show that Fed4Rec outperforms the baselines in terms of recommendation accuracy”).
Regarding Claim 6, Zhou in view of Han, Feuz, Dai, and Zhao teach the method of claim 5, further comprising: responsive to the determination that the vehicle comprises insufficient computational resources (Dai, Pg. 2, Col. 1, Para. 5-6, “Each task can either be offloaded to a selected VEC server to process, or be executed locally at the vehicle . . . If vehicle i chooses to offload task Di to a selected VEC server to process”, where the decision to “offload task Di” is determined based on whether the vehicle has sufficient computational resources to avoid “bottlenecks” that compromise “Quality of Service”, see Dai Pg. 1, Col. 1, Para. 2, “resource-constrained vehicles can be strained by computation intensive applications, resulting in bottlenecks and making it challenging for the vehicles to ensure the required level of Quality of Service”, with reference to “maximum allowed latency”, see Dai, Pg. 3, Col. 1, Para. 5, “we formulate the joint offloading and resource allocation scheme as an optimization problem . . . The first constraint (8b) guarantees that the task processing time cannot exceed the maximum allowed latency”)
for storing and running each machine learning model of the plurality of machine learning models (Zhou, Pg. 5311, Col. 1, Para. 4, “any individual data owner vi in this framework can keep its own data di and train the object detection model locally”, which, in view of Han, are they plurality of models, see Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where the “client” stores and runs the plurality of models, collectively, as the “averaged model”; see also Zhou, Pg. 5309, Col. 1, Para. “With small quantity of data (or low data quality depending on the camera resolution) and varying computational power of individual vehicles, local training is usually limited by its accuracy”),
checking privacy requirement settings of the vehicle, wherein selecting the vehicle-to-edge server association protocol from the plurality of vehicle-to-edge server association protocols is based on the privacy requirement settings (Zhou, Pg. 5310, Col. 2, Fig. 1 and Zhou, Pg. 5313, Col. 1, Fig. 3, “4: for each data owner vi ∈ V do 5: if vi supervised by rj do . . . 8: Submit the local training parameter wi (t) to rj”, where the exchange of data is based on the vehicle-to edge server association protocol of clustered “data owner[s] vi” “supervised by rj” “RSU[s]”, where vehicular system conditions, “contextual information, such as vehicle locations”, determines which “RSU” a “vehicles” will be “connected” to as a supervisor see Zhou, Pg. 5311, Col. 1, Para. 1, “RSUs are the middle brokers to collect and aggregate not only learning parameters but also contextual information, such as vehicle locations and navigation direction, from all the connected vehicles within the coverage to facilitate parameter aggregation”, where in view of Zhao, includes a determination of “public” or “private” privacy requirement settings, which determine which of the plurality of association protocols, “share data with server” or “do not share data with the server”, are selected, see Zhao, Pg. 1071, Col. 1, Abstract, “We propose Fed4Rec, a privacy-preserving framework for page recommendation based on federated learning (FL) and model-agnostic meta-learning (MAML), which allows machine learning models to train on data collected from both public users, who share data with the server, and private users, who do not share data with the server”).
The reasons of obviousness have been discussed in the rejection of claim 1, in regard to the combination of Zhou with Han, the rejection of claim 3, in regard to the combination of Zhou and Han with Dai, and the rejection of claim 4, in regard to the combination of Zhou, Han, and Dai, with Zhao, and remain applicable here.
Regarding Claim 12, the additional elements of the dependent claim are substantially the same as limitations of Claim 4, therefore it is rejected under the same rationale.
Regarding Claim 14, the additional elements of the dependent claim are substantially the same as limitations of Claim 6, therefore it is rejected under the same rationale.
Regarding Claim 18, the additional elements of the dependent claim are substantially the same as limitations of Claim 4, therefore it is rejected under the same rationale.
Claims 7-8, 15-16, and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Zhou in view of Han, Feuz, Dai, Zhao, and Karb et al. (hereinafter Karb) (“A Network-Based Transfer Learning Approach to Improve Sales Forecasting of New Products”).
Regarding Claim 7, Zhou in view of Han, Feuz, Dai, and Zhao teach the method of claim 6, further comprising: responsive to the privacy requirement settings being set to first privacy requirements, requesting data . . . . (Zhao, Pg. 1074, Col. 1, Para. 1, “This global model can then be downloaded by private users to perform page recommendations on their devices or can be used at the server for page recommendations for public users”, where “private users” have first privacy requirements)
from each edge server of the plurality of edge servers . . . of each machine learning model . . . [based on data of] the vehicle derived from the data acquired by the vehicle (Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where “clients” receive “multiple models” from the plurality of models hosted by the plurality of “different Ess” and the multiple models are identified, which requires that a machine learning model be identified, based on whether the “clients [are] in the overlapping areas”; Han, Pg. 3872, Col. 2, Para. 3, “We call this region in which the client can reliably communicate with multiple ESs overlapping cell area”, where a “client” is determined to be in an “overlapping cell area” “communication” data acquired by the “client” and from “multiple Ess”),
wherein identifying the machine learning model for the vehicle from the plurality of machine learning models hosted on the plurality of edge servers comprises . . . the vehicle (Zhou, Pg. 5313, Col. 1, Fig. 3, “13: Calculate the global parameters w(t+1) for model M . . . 15: Broadcast w(t+1) to the network” and Zhou, Pg. 5311, Col. 1, Para. 1, “the updated model parameters will be dispatched from the central cloud server to RSUs and then to individual vehicles”, where the “updated” “global . . . model M” is identified at the “central cloud server” and provided to the “individual vehicles”, where, in view of Han, if from a plurality of machine learning models hosted on the plurality of edge servers, see Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas
receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where “clients” receive “multiple models” from the plurality of models hosted by the plurality of “different Ess” and the multiple models are identified, which requires that a machine learning model be identified, based on whether the “clients [are] in the overlapping areas”; Han, Pg. 3872, Col. 2, Para. 3, “We call this region in which the client can reliably communicate with multiple ESs overlapping cell area”, where a “client” is determined to be in an “overlapping cell area” “communication” data acquired by the “client” and from “multiple Ess”).
The reasons of obviousness have been discussed in the rejection of claim 1, in regard to the combination of Zhou with Han, and the rejection of claim 4, in regard to the combination of Zhou, Han, and Dai, with Zhao, and remain applicable here.
Zhou in view of Han, Feuz, Dai, and Zhao do not explicitly disclose . . . feature information . . . data feature information representative . . . performing feature matching between data feature information of each machine learning and data feature information of . . . identifying the machine learning model corresponding to data feature information that is closest to the data feature information of . . . .
However, Karb teaches [a model identification method, wherein] . . . [data] feature information [of a plurality of machine learning models, wherein the] . . . data feature information [is] representative [of each of the models, is used] . . . (Pg. 8, Para. 2, “For each of the 14 source products one model is trained independently and the network architectures, including all parameters, were saved after training”; Pg. 7, Para. 2, “As described in Section 3, the success of the transfer is highly depended on the similarity between source and target. We analyze three different dimensions to compare product similarities, based on the information available at the time of the first forecast in week three . . . The values for the source products in this table are based on two years of available data. Within the three dimensions, the source and target products are assigned to clusters based on the respective values. This clustering approach is used to systematically search for the most suitable source models”, where the method uses data feature information, “the three dimensions”, as representative of the “suitability” of the “source models”)
performing feature matching between data feature information of each machine learning and data feature information of [a target model application] (Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values”, where the clustering uses the data feature information of both “source” and “target”, see Pg. 7, Para. 2, “For the target product, additionally to the sales price, the mean and standard deviation of the hourly sales as well as the share of promotion . . . The values for the source products in this table are based on two years of available data”) . . .
identifying the machine learning model corresponding to data feature information that is closest to the data feature information of [the target model application] . . . (Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values. This clustering approach is used to systematically search for the most suitable source models”, where, as discussed above, “clustering” is based on the data feature information of the source models and target model application, and where “clustering” is used to identify the “model” with the most “similarity” to the target, for example, see Pg. 11, Para. 1, “it can be concluded that the same price cluster is a good indicator for the similarity of the domains”).
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the requesting data, based on a privacy setting, from each edge server storing a machine learning, wherein the transmitting of data is based on the data derived from data acquired by the vehicle of Zhou in view of Han, Feuz, Dai, and Zhao with the model identification method, wherein data feature information of a plurality of machine learning models and data feature information of a target application model are compared to identify the machine learning model with data feature information closest to the data feature information of the target model application of Karb in order to select source models with similar data to the target application (Karb, Pg. 3-4, Para. 4-1, “The TL concept is based on the assumption that there are similarities between Ds and Dt and that it can be useful to transfer knowledge between these domains”), which improves model accuracy for the target application (Karb, Pg. 1, Abstract, “The experimental results show, that the prediction accuracy of deep neural networks for food sales forecasting can be effectively increased using the proposed approach”).
Regarding Claim 8, Zhou in view of Han, Feuz, Dai, Zhao, and Karb teach the method of claim 6, further comprising: responsive to the privacy requirement settings being set to second privacy requirements, transmitting, to each edge server of the plurality of edge servers, data feature information of the vehicle derived from the data acquired by the vehicle (Zhao, Pg. 1074, Col. 1, Para. 1, “This global model can then be downloaded by private users to perform page recommendations on their devices or can be used at the server for page recommendations for public users”, where “public users” have second privacy requirements; Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where “clients” receive “multiple models” from the plurality of models hosted by the plurality of “different Ess” and the multiple models are identified, which requires that a machine learning model be identified, based on whether the “clients [are] in the overlapping areas”; Han, Pg. 3872, Col. 2, Para. 3, “We call this region in which the client can reliably communicate with multiple ESs overlapping cell area”, where a “client” is determined to be in an “overlapping cell area” “communication” data acquired by the “client” and from “multiple Ess”, which requires that the client data be transmitted, either directly or indirectly, to each edge server, where in view of Karb, the data is data feature information, see Karb, Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values”, where the clustering uses the data feature information of both “source” and “target”);
and receiving data matching results from each edge server of the plurality of edge servers (Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where “clients” receive “multiple models” from the plurality of models hosted by the plurality of “different Ess”, where, in view Zhao, the computations occur at the server, Zhao, Pg. 1074, Col. 1, Para. 1, “This global model can then be downloaded by private users to perform page recommendations on their devices or can be used at the server for page recommendations for public users”, and in view of Karb, the computations include the data matching, and therefore the receiving of data includes the data matching results, see Karb, Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values. This clustering approach is used to systematically search for the most suitable source models”, where, as discussed above, “clustering” is based on the data feature information of the source models and target model application, and where “clustering” is used to identify the “model” with the most “similarity” to the target, for example, see Karb, Pg. 11, Para. 1, “it can be concluded that the same price cluster is a good indicator for the similarity of the domains”),
wherein the data matching results from each edge sever comprises a measure of similarity between data feature information of the machine learning model hosted by a respective edge server and the data feature information of the vehicle, wherein identifying the machine learning model for the vehicle from the plurality of machine learning models hosted on the plurality of edge servers comprises identifying the machine learning model corresponding to a data matching result having highest measure of similarity (Karb, Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values”, where the clustering uses the data feature information of both “source” and “target”, see Karb, Pg. 7, Para. 2, “For the target product, additionally to the sales price, the mean and standard deviation of the hourly sales as well as the share of promotion . . . The values for the source products in this table are based on two years of available data” where, “clustering” is based on the data feature information of the source models and target model application, and where “clustering” is used to identify the “model” with the most “similarity” to the target, for example, see Karb, Pg. 11, Para. 1, “it can be concluded that the same price cluster is a good indicator for the similarity of the domains” , where the course models are the edge servers and the target is the vehicle, discussed above in regard to Zhou and Han, see generally Zhou, Pg. 5309, Col. 1, Para. 2, “In this study, we propose a two-layer federated learning model based on convolutional neural network (TFL-CNN), which makes use of the local and global contexts of individual vehicles and RSUs to perform hierarchical and heterogeneous model selection and aggregation at the edge and cloud level”; and Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”).
The reasons of obviousness have been discussed in the rejection of claim 1, in regard to the combination of Zhou with Han, the rejection of claim 4, in regard to the combination of Zhou, Han, and Dai, with Zhao, and the rejection of claim 7, in regard to the combination of Zhou, Han, Dai, and Zhao, with Karb, and remain applicable here.
Regarding Claim 15, the additional elements of the dependent claim are substantially the same as limitations of Claim 7, therefore it is rejected under the same rationale.
Regarding Claim 16, the additional elements of the dependent claim are substantially the same as limitations of Claim 8, therefore it is rejected under the same rationale.
Regarding Claim 19, Zhou in view of Han, Feuz, Dai, Zhao, and Karb teach the server of claim 18, wherein the one or more processors are further configured to execute the instructions to (Zhou, Pg. 5310-5311, Col. 2-1, Para. 4-1, “In the middle layer, each RSU has limited caching and computing capabilities, and is responsible for supervising all the interconnected vehicles in its coverage . . . RSUs are the middle brokers to collect and aggregate . . . from all the connected vehicles within the coverage”, where the “computing”, “supervising”, and “caching” require a memory storing instructions executed by a processor):
based on the privacy requirement settings of the at least one vehicle being set to first privacy requirements, transmit data feature information representative of the machine learning model (Zhao, Pg. 1074, Col. 1, Para. 1, “This global model can then be downloaded by private users to perform page recommendations on their devices or can be used at the server for page recommendations for public users”, where “private users” have first privacy requirements; Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where “clients” receive “multiple models” from the plurality of models hosted by the plurality of “different Ess” and the multiple models are identified, which requires that a machine learning model be identified, based on whether the “clients [are] in the overlapping areas”; Han, Pg. 3872, Col. 2, Para. 3, “We call this region in which the client can reliably communicate with multiple ESs overlapping cell area”, where a “client” is determined to be in an “overlapping cell area” “communication” data acquired by the “client” and from “multiple Ess”, where in view of Karb, the data is data feature information, see Karb, Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values”, where the clustering uses the data feature information of both “source” and “target”);
and based on a comparison of the data feature information of the machine learning model with data feature information of the at least one vehicle (Karb, Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values. This clustering approach is used to systematically search for the most suitable source models”, where, as discussed above, “clustering” is based on the data feature information of the source models and target model application, and where “clustering” is used to identify the “model” with the most “similarity” to the target, for example, see Karb, Pg. 11, Para. 1, “it can be concluded that the same price cluster is a good indicator for the similarity of the domains”, where the course models are the edge servers and the target is the vehicle, discussed above in regard to Zhou and Han, see generally Zhou, Pg. 5309, Col. 1, Para. 2, “In this study, we propose a two-layer federated learning model based on convolutional neural network (TFL-CNN), which makes use of the local and global contexts of individual vehicles and RSUs to perform hierarchical and heterogeneous model selection and aggregation at the edge and cloud level” and Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”);
transmitting model parameters of the machine learning model to the at least one vehicle (Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where “clients” receive “multiple models” from the plurality of models hosted by the plurality of “different Ess” and the multiple models are identified, which requires that a machine learning model be identified, based on whether the “clients [are] in the overlapping areas”; Han, Pg. 3872, Col. 2, Para. 3, “We call this region in which the client can reliably communicate with multiple ESs overlapping cell area”, where a “client” is determined to be in an “overlapping cell area” “communication” data acquired by the “client” and from “multiple Ess”, where in view of Karb, the data is data feature information, see Karb, Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values”, where the clustering uses the data feature information of both “source” and “target”; where the course models are the edge servers and the target is the vehicle, discussed above in regard to Zhou and Han, see generally Zhou, Pg. 5309, Col. 1, Para. 2, “In this study, we propose a two-layer federated learning model based on convolutional neural network (TFL-CNN), which makes use of the local and global contexts of individual vehicles and RSUs to perform hierarchical and heterogeneous model selection and aggregation at the edge and cloud level”; and Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”).
The reasons of obviousness have been discussed in the rejection of claim 1, in regard to the combination of Zhou with Han, the rejection of claim 4, in regard to the combination of Zhou, Han, and Dai, with Zhao, and the rejection of claim 7, in regard to the combination of Zhou, Han, Dai, and Zhao, with Karb, and remain applicable here.
Regarding Claim 20, Zhou in view of Han, Dai, Zhao, and Karb teach the server of claim 18, wherein the one or more processors are further configured to execute the instructions to (Zhou, Pg. 5310-5311, Col. 2-1, Para. 4-1, “In the middle layer, each RSU has limited caching and computing capabilities, and is responsible for supervising all the interconnected vehicles in its coverage . . . RSUs are the middle brokers to collect and aggregate . . . from all the connected vehicles within the coverage”, where the “computing”, “supervising”, and “caching” require a memory storing instructions executed by a processor):
based on the privacy requirement settings of the at least one vehicle being set to second privacy requirements, receiving data feature information of the at least one vehicle derived from the data acquired by the vehicle (Zhao, Pg. 1074, Col. 1, Para. 1, “This global model can then be downloaded by private users to perform page recommendations on their devices or can be used at the server for page recommendations for public users”, where “public users” have second privacy requirements, so they send data that is received “at the server”; Han, Pg. 3872, Col. 2, Para. 3, “We call this region in which the client can reliably communicate with multiple ESs overlapping cell area”, where a “client” is determined to be in an “overlapping cell area” “communication” data acquired by the “client” and from “multiple Ess”, which requires that the client data be transmitted, either directly or indirectly, to each edge server, where in view of Karb, the data is data feature information, see Karb, Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values”, where the clustering uses the data feature information of both “source” and “target”);
performing data matching data feature information of the machine learning model and the received data feature information of the at least one vehicle (Karb, Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values. This clustering approach is used to systematically search for the most suitable source models”, where, as discussed above, “clustering” is based on the data feature information of the source models and target model application, and where “clustering” is used to identify the “model” with the most “similarity” to the target, for example, see Karb, Pg. 11, Para. 1, “it can be concluded that the same price cluster is a good indicator for the similarity of the domains”, where the course models are the edge servers and the target is the vehicle, discussed above in regard to Zhou and Han, see generally Zhou, Pg. 5309, Col. 1, Para. 2, “In this study, we propose a two-layer federated learning model based on convolutional neural network (TFL-CNN), which makes use of the local and global contexts of individual vehicles and RSUs to perform hierarchical and heterogeneous model selection and aggregation at the edge and cloud level”; and Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”; and where, as discussed above, in view of Zhao the computations occur at the server, see Zhao, Pg. 1074, Col. 1, Para. 1, “This global model can then be downloaded by private users to perform page recommendations on their devices or can be used at the server for page recommendations for public users”);
transmitting data matching results to the at least one vehicle (Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where “clients” receive “multiple models” from the plurality of models hosted by the plurality of “different Ess”, where, in view Zhao, the computations occur at the server, and in view of Karb, the computations include the data matching, and therefore the receiving of data includes the data matching results, see Karb, Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values. This clustering approach is used to systematically search for the most suitable source models”);
and responsive to a request from the at least one vehicle, transmitting model parameters of the machine learning model to the at least one vehicle, wherein the request is based on the transmitted data matching results (Han, Pg. 3870, Col. 1, Abstract, “in the model-downloading stage, the clients in the overlapping areas receive multiple models from different ESs, take the average of the received models, and then update the averaged model with their local data”, where “clients” receive “multiple models” from the plurality of models hosted by the plurality of “different Ess” and the multiple models are identified, which requires that a machine learning model be identified, based on whether the “clients [are] in the overlapping areas”, where, in view of Karb, it is based on the transmitted data matching results, see Karb, Pg. 7, Para. 2, “Within the three dimensions, the source and target products are assigned to clusters based on the respective values. This clustering approach is used to systematically search for the most suitable source models”, where, as discussed above, “clustering” is based on the data feature information of the source models and target model application, and where “clustering” is used to identify the “model” with the most “similarity” to the target, for example, see Karb, Pg. 11, Para. 1, “it can be concluded that the same price cluster is a good indicator for the similarity of the domains”).
The reasons of obviousness have been discussed in the rejection of claim 1, in regard to the combination of Zhou with Han, the rejection of claim 4, in regard to the combination of Zhou, Han, and Dai, with Zhao, and the rejection of claim 7, in regard to the combination of Zhou, Han, Dai, and Zhao, with Karb, and remain applicable here.
Response to Arguments
Applicant's arguments filed on February 2nd, 2026 have been fully considered. Each argument is addressed in detail below.
I. Applicant argues the objections to the drawings should be withdrawn (Applicant’s Remarks,
02/02/2026, Pg. 12, Section “Objections to the Drawings”).
Applicant’s amendments to the drawings and the specification have overcome each and every objection to the drawings, as previously set forth in the February 2nd, 2026 Office Action. As a result, these objections have been withdrawn.
II. Applicant argues the rejections of the claims under 35 USC § 112 should be withdrawn (Applicant’s Remarks, 02/02/2026, Pg. 12-13, Section “Rejections under 35 U.S.C. § 112(b)”).
Applicant’s amendments to the claims have overcome each and every rejection of the claims under 35 USC § 112, as previously set forth in the February 2nd, 2026 Office Action. As a result, the rejections of the claims under 35 USC § 112 have been withdrawn.
III. Applicant argues the rejections of the claims under 35 USC § 103 should be withdrawn (Applicant’s Remarks, 02/02/2026, Pg. 13-17, Section “Rejections under 35 U.S.C. § 103”).
In response to Applicant’s amendments, the previously communicated rejections under 35
U.S.C. § 103, have been withdrawn. However, Applicants arguments are not persuasive in light of the
new grounds for rejection, under 35 U.S.C. § 103, discussed in detail above. The new grounds of
rejection rely on new prior art of record to teach the new combination of elements in the amended independent claims, which were not presented in this arrangement in any of the previously presented claims. As a result, Applicant’s arguments are rendered moot.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MATTHEW BRYCE GOLAN whose telephone number is (571)272-5159. The examiner can normally be reached Monday through Friday, 8:00 AM to 5:00 PM ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Alexey Shmatov can be reached at (571) 270-3428. 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.
/MATTHEW BRYCE GOLAN/Examiner, Art Unit 2123
/ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123
Prosecution Insights