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 Arguments
Regarding the objections to claims 1–66 in the previous office action, amendments to the claim have overcome the objection; however, the amendments have led to a new objection, which is given below.
Regarding the rejection of claims as judicial exceptions under 35 U.S.C. 101, Examiner finds that amendments to the claims have overcome the rejections, which are withdrawn. In particular, Examiner finds that the use of a federation-of-federations for obtaining the predictions for voting (recited in claim 1 as “obtaining a student prediction from each of a plurality of respective student machine-learning models of each of a plurality of nodes of the cluster, wherein each node of the plurality of nodes comprises the respective plurality of student machine-learning models, performing a voting, at the cluster hub, of the student predictions from the plurality of respective student machine-learning models”) recites specific steps described as an improvement in the specification in at least paragraph 1013 (“The techniques described herein can provide a number of benefits over conventional approaches. A benefit of a federation of federations is that models can glean learnings from multiple different networks without communicating any private data”). The use of the plurality of nodes each with a plurality of models therefore integrates the abstract idea (the mental process of tallying votes, as recited in “performing a voting, at the cluster hub, of the student predictions from the plurality of respective student machine-learning models”) into a practical application.
Regarding the rejection of claims under 35 U.S.C. 103, Applicant’s arguments are directed towards amended portions of the claims which have not been previously examined, and for which new grounds of rejection are given below.
Claim Objections
Claims 1-66 objected to because of the following informality. In claim 1 and claim 34, Examiner respectfully suggests the word “a” should be inserted as underlined in “classifying, at a cluster hub of a cluster, a cluster training dataset of the cluster by, for each datapoint of the cluster training dataset,” or the limitation should be otherwise amended for clarification and grammar. Claims 2-33 and 35-66 objected to by dependency on claim 1 or 34.
Appropriate correction is required.
Claim Rejections - 35 USC § 103
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, 21, 33–34, 54, and 66 rejected under 35 U.S.C. 103 over Kim et al., US Pre-Grant Publication No. 2023/0075369 (hereafter Kim) in view of Wang et al., US Pre-Grant Publication No. 2022/0121999 (hereafter Wang).
Regarding claim 1 and analogous claim 34:
Kim teaches:
“A method comprising”: Kim, paragraph 0004, “Features, no matter how plentiful, may be used as training data only if they are associated with corresponding labels. For example, images used to train a network to output a label must each be associated with a label. Such data labelling often consumes significant time and resources, resulting in undesirable trade-offs between the cost of model training and the usefulness of the trained model. Systems to facilitate the generation of training data labels are desired.”; Kim, paragraph 0015, “The following description is provided to enable any person in the art to make and use the described embodiments and sets forth the best mode contemplated for carrying out some embodiments [a method].”
(bold only) “classifying, at a cluster hub of cluster, a cluster training dataset of the cluster by, for each datapoint of the cluster training dataset: obtaining a student prediction from each of a plurality of respective student machine-learning models of each of a plurality of nodes of the cluster”: Kim, paragraph 0032, “A pseudo-label corresponding to each feature is determined [classifying a … training dataset by, for each datapoint of the … training dataset] at S230 based on the respective pseudo-labels output by each model based on the feature [obtaining a student prediction from each of a plurality of respective student machine-learning models].”
(bold only) “performing a voting, at the cluster hub, of the student predictions from the plurality of respective student machine-learning models of at least a portion of the plurality of nodes of the cluster to determine a respective classification for the datapoint”: Kim, paragraph 0032, “A pseudo-label corresponding to each feature is determined at S230 based on the respective pseudo-labels output by each model based on the feature. According to system 400, label determination component 420 receives all labels output by each of the plurality of trained models, and generates pseudo-labels 430 for each of feature ul-um based thereon. For z=i to m, label determination component 420 determines pseudo-label pz corresponding to feature uz based on each pseudo-label which was output by trained models 120-1, 120-2 and 120-e based on feature uz (i.e., pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e). Pseudo-label pz1 may be determined based on pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e in any suitable manner, including but not limited to majority voting (i.e., choosing the most-often occurring value of the pseudo-labels) [performing a voting of the student predictions from the plurality of respective student machine-learning models … to determine a respective classification for the datapoint] or averaging the softmax outputs of each trained model.”
(bold only) “and labeling the datapoint of the cluster training dataset with the respective classification”: Kim, paragraph 0032, “A pseudo-label corresponding to each feature is determined at S230 based on the respective pseudo-labels output by each model based on the feature. According to system 400, label determination component 420 receives all labels output by each of the plurality of trained models, and generates pseudo-labels 430 for each of feature ul-um based thereon. For z=i to m, label determination component 420 determines pseudo-label pz corresponding to feature uz based on each pseudo-label which was output by trained models 120-1, 120-2 and 120-e based on feature uz (i.e., pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e). Pseudo-label pz1 may be determined based on pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e in any suitable manner, including but not limited to majority voting (i.e., choosing the most-often occurring value of the pseudo-labels) or averaging the softmax outputs of each trained model [labeling the datapoint of the … training dataset with the respective classification].”
(bold only) “training, at the cluster hub, a cluster machine-learning model of the cluster using the cluster training dataset, as classified”: Kim, paragraph 0016, “Briefly, some embodiments operate to train multiple models based on a set of labeled training data, to generate pseudo-labels corresponding to each of a plurality of features using the trained models, and to train a model based on the labeled training data and on pseudo-labeled training data comprised of the plurality of features and corresponding pseudo labels [training a … machine-learning model … using the … training dataset, as classified]. The quality of the generated pseudo-labels may be improved over prior systems due to regularization provided by the multiple trained models, resulting in improved the accuracy of the final trained model. Further, using the final trained model for subsequent inferences requires less time and fewer resources than use of the multiple trained models.”
Kim does not explicitly teach:
(bold only) “classifying, at a cluster hub of cluster, a cluster training dataset of the cluster by, for each datapoint of the cluster training dataset: obtaining a student prediction from each of a plurality of respective student machine-learning models of each of a plurality of nodes of the cluster”
“wherein each node of the plurality of nodes comprises the respective plurality of student machine-learning models”
(bold only) “performing a voting, at the cluster hub, of the student predictions from the plurality of respective student machine-learning models of at least a portion of the plurality of nodes of the cluster to determine a respective classification for the datapoint”
(bold only) “labeling the datapoint of the cluster training dataset with the respective classification”
(bold only) “training, at the cluster hub, a cluster machine-learning model of the cluster using the cluster training dataset, as classified”
Wang teaches:
(bold only) “classifying, at a cluster hub of cluster, a cluster training dataset of the cluster by, for each datapoint of the cluster training dataset: obtaining a student prediction from each of a plurality of respective student machine-learning models of each of a plurality of nodes of the cluster,” (bold only) “performing a voting, at the cluster hub, of the student predictions from the plurality of respective student machine-learning models of at least a portion of the plurality of nodes of the cluster to determine a respective classification for the datapoint,” (bold only) “labeling the datapoint of the cluster training dataset with the respective classification,” and (bold only) “training, at the cluster hub, a cluster machine-learning model of the cluster using the cluster training dataset, as classified”: Wang, paragraph 0039, “Referring now to FIGS. 3 and 4, after the server 108 receives the evaluation results 112 for each of the clients 102 [of each of a plurality of nodes of the cluster][at least a portion of the plurality of nodes of the cluster], the server [at a cluster hub of cluster] can determine an optimal subset of models and assign weights for the combination of selected models. For example, as shown in FIG. 4 model 1 may be provided a 0.4 weight, model 3 may be provided a 0.6 weight, and model 2 may not be selected. Details of the determination of these weights are provided below.”
“wherein each node of the plurality of nodes comprises the respective plurality of student machine-learning models”: Wang, Fig. 2,
PNG
media_image1.png
431
926
media_image1.png
Greyscale
[showing clients 1–3, a plurality of nodes, each with models 1–3, hence, comprises the respective plurality of student machine-learning models]; Wang, paragraph 0006, “In one embodiment, a computer implemented method includes distributing a plurality of prediction models, where each of a plurality of clients initially includes at least one associated prediction model from the plurality of prediction models, among all of the plurality of clients to provide each of the plurality of clients with each of the plurality of prediction models. Each of the plurality of prediction models is evaluated on at least a portion of a dataset resident on each of the plurality of clients to output a quantification indicating how each of the prediction models fit at least the portion of the local dataset of each of the plurality of clients.”
Wang and Kim are analogous arts as they are both related to ensemble models. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the multi-model clients of Wang with the teachings of Kim to arrive at the present invention, in order to combine the benefits of federated learning and ensemble learning, as stated in Wang, paragraph 0019, “By virtue of the concepts discussed herein, a system and method are provided that improves upon the approaches currently used in ensemble learning. These concepts can assure scalability and efficiency of such ensemble learning systems while minimizing the amount of data
required to be shared between clients with their own decentralized datasets and/or between such clients and a server.”
Regarding claim 21 and analogous claim 54:
Kim as modified by Wang teaches “The method of claim 1.”
Kim further teaches “wherein each of the plurality of respective student machine-learning models comprises one of an XGBoost model, a random forest model, a neural network model, a GNN model, a GAN model, a TFT model, an autoencoder-type model, an RL model, or an HRL model”: Kim, paragraph 0019, “Models as described herein may comprise any one or more types of artificial neural network model that are or become known [each of the plurality of respective student machine-learning models comprises … a neural network model], including but not limited to convolutional neural network models, recurrent neural network models, long short-term memory network models, deep reservoir computing and deep echo state network models, deep belief network models, and deep stacking network models.”
Regarding claim 33 and analogous claim 66:
Kim as modified by Wang teaches “The method of claim 1.”
Kim further teaches “wherein the voting comprises a majority voting”: Kim, paragraph 0032, “A pseudo-label corresponding to each feature is determined at S230 based on the respective pseudo-labels output by each model based on the feature. According to system 400, label determination component 420 receives all labels output by each of the plurality of trained models, and generates pseudo-labels 430 for each of feature ul-um based thereon. For z=i to m, label determination component 420 determines pseudo-label pz corresponding to feature uz based on each pseudo-label which was output by trained models 120-1, 120-2 and 120-e based on feature uz (i.e., pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e). Pseudo-label pz1 may be determined based on pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e in any suitable manner, including but not limited to majority voting (i.e., choosing the most-often occurring value of the pseudo-labels) [wherein the voting comprises a majority voting] or averaging the softmax outputs of each trained model.”
Claims 2–3 and 35–36 rejected under 35 U.S.C. 103 over Kim as modified by Wang in view of Miralles et al., US Pre-Grant Publication No. 2024/0386283 (hereafter Miralles).
Regarding claim 2 and analogous claim 35:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “providing model weights of the cluster machine-learning model, as trained, to the plurality of nodes of the cluster.”
Miralles teaches “providing model weights of the cluster machine-learning model, as trained, to the plurality of nodes of the cluster”: Miralles, paragraph 0038, “In particular, the invention concerns a coordination entity able to configure weights of models of neural networks of the same structure, of nodes from a set of nodes of a communication network, by federated learning of said weights in which said nodes locally train their models of neural networks and share the weights of their model with other nodes of said network [providing model weights of the cluster machine-learning model, as trained, to the plurality of nodes of the cluster].”
Miralles and Kim are analogous arts as they are both related to machine learning ensembles. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the federated learning clusters of Miralles with the teachings of Kim to arrive at the present invention, in order to improve the efficiency of the classification system, as stated in Miralles, paragraph 0007, “In recent years, the federated learning approach has attracted a lot of interest in many fields, such as healthcare, banking, industry 4.0 or smart cities, because it can help build better global models, while preserving the confidentiality of the local files (medical, financial files, etc.). It can provide a natural solution to the growing needs for personal data protection, while addressing the current technological challenges: decrease in the energy consumption, minimization of the latency, two challenges with the deployment of 5G technology. As explained above, the federated learning is thus a form of distributed learning where several nodes collaboratively solve a machine learning task.”
Regarding claim 3 and analogous claim 36:
Kim as modified by Wang and Miralles teaches “The method of claim 2.”
Kim further teaches “wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model to make inferences on incoming network data received at the node”: Kim, paragraph 0016, “Briefly, some embodiments operate to train multiple models based on a set of labeled training data, to generate pseudo-labels corresponding to each of a plurality of features using the trained models, and to train a model based on the labeled training data and on pseudo-labeled training data comprised of the plurality of features and corresponding pseudo labels. The quality of the generated pseudo-labels may be improved over prior systems due to regularization provided by the multiple trained models, resulting in improved the accuracy of the final trained model. Further, using the final trained model for subsequent inferences [uses the model weights in a respective node machine-learning model to make inferences] requires less time and fewer resources than use of the multiple trained models”; Kim, paragraph 0043, “Inference agent 718 may receive features [on incoming network data received at the node] and input the features to the trained final model to generate associated labels.”
Claim 4 and analogous claim 37 rejected under 35 U.S.C. 103 over Kim as modified by Wang and Miralles in view of Park et al., US Patent No. 10778705 (hereafter Park).
Kim as modified by Wang and Miralles teaches “The method of claim 3.”
Kim as modified by Wang and Miralles does not explicitly teach “wherein the respective node machine-learning model is used to make the inferences on the incoming network data in an intrusion detection system or an intrusion prevention system at the node.”
Park teaches “wherein the respective node machine-learning model is used to make the inferences on the incoming network data in an intrusion detection system or an intrusion prevention system at the node”: Park, paragraph 0001, “The present invention relates to a deep-learning-based intrusion detection method [an intrusion detection system], system and computer program for web applications, and more particularly, to a method, a system and a computer program for detecting whether traffic is a hacker attack by inputting a network traffic flowing into a server farm to a deep neural network (DNN) model and outputting data from the model.”
Park and Kim are analogous arts as they are both related to machine learning model training and inference. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the intrusion detection of Park with the teachings of Kim to arrive at the present invention, in order to apply the methods to defend against security attacks, as stated in Park, paragraph 0005, “Accordingly there are increasing needs for studies on providing web services more securely, detecting attacks, which exploit the web services, and defending against the attacks. There are needs for a specialized intrusion detection system for the web services to respond quickly to various web attacks and reduce false alarms.”
Claim 5 and analogous claim 38 rejected under 35 U.S.C. 103 over Kim as modified by Wang and Miralles in view of Villasante Marcos et al., US Pre-Grant Publication No. 2022/0240157 (hereafter Marcos).
Kim as modified by Wang and Miralles teaches “The method of claim 3.”
Kim as modified by Wang and Miralles does not explicitly teach “wherein the respective node machine-learning model is used to make the inferences on the incoming network data in a traffic routing function at the node.”
Marcos teaches “wherein the respective node machine-learning model is used to make the inferences on the incoming network data in a traffic routing function at the node”: Marcos, paragraph 0014, “An aspect of the disclosure provides a data traffic routing method for controlling data traffic in a communication network, the method comprising: receiving, at a first agent from a User Plane Function, communication network status information; calculating, by the first agent, data traffic routing instructions using a current routing model [make the inferences on the incoming network data in a traffic routing function]; sending by the first agent: the data traffic routing instructions to the User Plane Function; and experience information to a second agent; storing, at the second agent, the experience information; determining, at the second agent, if the number of instances of stored experience information exceeds a predetermined threshold; and if it is determined that the number of instances of stored experience information exceeds a predetermined threshold: training a neural network using the instances of stored experience information; and updating the current routing model using results of the neural network training. In this way, accurate routing instructions for data traffic may be promptly provided.”
Marcos and Kim are analogous arts as they are both related to machine learning model training and inference. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the traffic routing of Marcos with the teachings of Kim to arrive at the present invention, in order to apply the methods to efficiently route traffic, as stated in Marcos, paragraph 0012, “It is an object of the present disclosure to facilitate data traffic control in a communication network, such that the data traffic may be routed more efficiently.”
Claims 6, 8, 39, and 41 rejected under 35 U.S.C. 103 over Kim as modified by Wang and Miralles in view of Dodwell et al., US Pre-Grant Publication No. 2021/0174223 (hereafter Dodwell).
Regarding claim 6 and analogous claim 39:
Kim as modified by Wang and Miralles teaches “The method of claim 2.”
Miralles further teaches (bold only) “wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model that identifies items of interest in a security imagery feed received at the node”: Miralles, paragraphs 0016-0019, “In at least one embodiment, the configuration method comprises, during said federated learning: sending, to the aggregation node of said at least one cluster, a request to learn the weights of the models of the nodes of said cluster with the weights of a global model to the set of nodes [wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model]; receiving, from the aggregation node of said at least one cluster, the weights of said aggregate model of said cluster resulting from said learning; and updating the weights of the global model by aggregation of the received weights of the aggregate model of said at least one cluster.”
Miralles and Kim are combinable for the rationale given under claim 2.
Kim as modified by Wang and Miralles does not explicitly teach (bold only) “wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model that identifies items of interest in a security imagery feed received at the node.”
Dodwell teaches (bold only) “wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model that identifies items of interest in a security imagery feed received at the node”: Dodwell, paragraph 0033, “The bias mitigation server 202 can implement a slow path model during the first time period that shadows the fast path model to determine whether to provide captured images by the security camera of the premises to the first user device 218 according to the input data (images of people from the household) and the additional data (images of people who frequently visit the premises) based on the slow path model. In the example above, the slow path model may determine that the captured images should not be provided to the first user device 218 because there is no perceived security threat according to the additional data [machine-learning model that identifies items of interest in a security imagery feed]. Such a prediction can be done using machine learning techniques by the slow path model. Further, the determination of whether people in the images captured by the security camera of the premises are images of people of the household or images of people who frequently visit the premises can be performed using image recognition techniques.”
Dodwell and Kim are analogous arts as they are both related to machine learning model training and inference. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the security image processing of Dodwell with the teachings of Kim to arrive at the present invention, in order to apply the methods to identify images suspicious enough to present to a user, as stated in Dodwell, paragraph 0033, “. In the example above, the slow path model may determine that the captured images should not be provided to the first user device 218 because there is no perceived security threat according to the additional data.”
Regarding claim 8 and analogous claim 41:
Kim as modified by Wang and Miralles teaches “The method of claim 2.”
Miralles further teaches (bold only) “wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model that detects anomalies from one or more baselines at the node”: Miralles, paragraphs 0016-0019, “In at least one embodiment, the configuration method comprises, during said federated learning: sending, to the aggregation node of said at least one cluster, a request to learn the weights of the models of the nodes of said cluster with the weights of a global model to the set of nodes [wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model]; receiving, from the aggregation node of said at least one cluster, the weights of said aggregate model of said cluster resulting from said learning; and updating the weights of the global model by aggregation of the received weights of the aggregate model of said at least one cluster.”
Miralles and Kim are combinable for the rationale given under claim 2.
Kim as modified by Wang and Miralles does not explicitly teach (bold only) “wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model that detects anomalies from one or more baselines at the node.”
Dodwell teaches (bold only) “wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model that detects anomalies from one or more baselines at the node”: Dodwell, paragraph 0033, “The bias mitigation server 202 can implement a slow path model during the first time period that shadows the fast path model to determine whether to provide captured images by the security camera of the premises to the first user device 218 according to the input data (images of people from the household) and the additional data (images of people who frequently visit the premises) based on the slow path model. In the example above, the slow path model may determine that the captured images should not be provided to the first user device 218 because there is no perceived security threat according to the additional data [that detects anomalies from one or more baselines at the node]. Such a prediction can be done using machine learning techniques by the slow path model. Further, the determination of whether people in the images captured by the security camera of the premises are images of people of the household or images of people who frequently visit the premises can be performed using image recognition techniques.”
Dodwell and Kim are analogous arts as they are both related to machine learning model training and inference. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the security image processing of Dodwell with the teachings of Kim to arrive at the present invention, in order to apply the methods to identify images suspicious enough to present to a user, as stated in Dodwell, paragraph 0033, “. In the example above, the slow path model may determine that the captured images should not be provided to the first user device 218 because there is no perceived security threat according to the additional data.”
Claim 7 and analogous claim 40 rejected under 35 U.S.C. 103 over Kim as modified by Wang and Miralles in view of Chen et al., US Pre-Grant Publication No. 2021/0392173 (hereafter Chen).
Kim as modified by Wang and Miralles teaches “The method of claim 2.”
Miralles further teaches (bold only) “wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model that performs natural language processing at the node for spam detection”: Miralles, paragraphs 0016-0019, “In at least one embodiment, the configuration method comprises, during said federated learning: sending, to the aggregation node of said at least one cluster, a request to learn the weights of the models of the nodes of said cluster with the weights of a global model to the set of nodes [wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model]; receiving, from the aggregation node of said at least one cluster, the weights of said aggregate model of said cluster resulting from said learning; and updating the weights of the global model by aggregation of the received weights of the aggregate model of said at least one cluster.”
Miralles and Kim are combinable for the rationale given under claim 2.
Kim as modified by Wang and Miralles does not explicitly teach (bold only) “wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model that performs natural language processing at the node for spam detection.”
Chen teaches (bold only) “wherein each node of the plurality of nodes uses the model weights in a respective node machine-learning model that performs natural language processing at the node for spam detection”: Chen, paragraph 0007, “FIG. 6 illustrates an example process for performing a spam detection analysis on a call request based on one or more machine learning models [spam detection], and the refinement of such models based on one or more actions taken after the spam detection analysis was performed”; Chen, paragraph 0013, “For example, the MLSDC may use Natural Language Processing (“NLP”) [natural language processing], pattern matching, and/or other suitable techniques to identify words and/or phrases included in the communications, use voice recognition techniques and/or other types of audio analysis to identify voice signatures or other audible features included in the communications, and/or other suitable types of analysis.”
Chen and Kim are analogous arts as they are both related to machine learning model training and inference. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the spam detection of Chen with the teachings of Kim to arrive at the present invention, in order to apply the methods to block unwanted communications, as stated in Chen, paragraph 0001, “Service providers, such as wireless networks, voice call providers, and/or other entities, may offer packet-based messaging or voice call services. Such services may be used in ways that may be undesirable to certain users, such as the use of such services to send unauthorized or undesired communications, such as voice calls. Such communications may sometimes be referred to as ‘spam.’”
Claim 9-11 and 42-44 rejected under 35 U.S.C. 103 over Kim as modified by Wang in view of Zhang, “Reliable face recognition by random subspace Support Vector Machine ensemble,” 2012, Proceedings of the 2012 International Conference on Machine Learning and Cybernetics (hereafter Zhang).
Regarding claim 9 and analogous claim 42:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein performing the voting further comprises using confidence filtering with a predetermined confidence value.”
Zhang teaches “wherein performing the voting further comprises using confidence filtering with a predetermined confidence value”: Zhang, section 3, paragraph 5, “Obviously, t is a tunable threshold that control the rejection rate, and we use t to relate the consensus degree from the voting [the voting] to confidence measure and abstain to classify ambiguous samples in classification”; Zhang, Section 1, paragraph 4, “With the proposed SVM ensemble operating on the four feature descriptions, a sample can be either classified or rejected. An instance will be accepted for classification when a confidence from ensemble decision is larger than a prescribed threshold and rejected for classification otherwise [comprises using confidence filtering with a predetermined confidence value]. The conformity from ensemble is thus linked to confidence of classification. The reject option can subsequently improve classification reliability and leave the control of classification accuracy to user. The efficiency of the proposed system was verified using a realistic face database the author created.”
Zhang and Kim are analogous arts as they are both related to ensemble machine learning. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the confidence measure of Zhang with the teachings of Kim to arrive at the present invention, in order to improve reliability, as stated in Zhang, Section 1, paragraph 4, “The conformity from ensemble is thus linked to confidence of classification. The reject option can subsequently improve classification reliability and leave the control of classification accuracy to user.”
Regarding claim 10 and analogous claim 43:
Kim as modified by Wang teaches “The method of claim 1.”
Kim further teaches (bold only) “before performing the voting, for each node of the plurality of nodes of the cluster: performing consistency voting on the student predictions from the plurality of respective student machine-learning models of the node to determine if the node will participate in the voting”: Kim, paragraph 0032, “A pseudo-label corresponding to each feature is determined at S230 based on the respective pseudo-labels output by each model based on the feature. According to system 400, label determination component 420 receives all labels output by each of the plurality of trained models, and generates pseudo-labels 430 for each of feature ul-um based thereon. For z=i to m, label determination component 420 determines pseudo-label pz corresponding to feature uz based on each pseudo-label which was output by trained models 120-1, 120-2 and 120-e based on feature uz (i.e., pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e). Pseudo-label pz1 may be determined based on pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e in any suitable manner, including but not limited to majority voting (i.e., choosing the most-often occurring value of the pseudo-labels) [performing the voting] or averaging the softmax outputs of each trained model.”
Wang further teaches (bold only) “before performing the voting, for each node of the plurality of nodes of the cluster: performing consistency voting on the student predictions from the plurality of respective student machine-learning models of the node to determine if the node will participate in the voting”: Wang, paragraph 0039, “Referring now to FIGS. 3 and 4, after the server 108 receives the evaluation results 112 for each of the clients 102 [for each node of the plurality of nodes of the cluster], the server can determine an optimal subset of models and assign weights for the combination of selected models. For example, as shown in FIG. 4 model 1 may be provided a 0.4 weight, model 3 may be provided a 0.6 weight, and model 2 may not be selected. Details of the determination of these weights are provided below.”
Wang and Kim are combinable for the rationale given under claim 1.
Kim as modified by Wang does not explicitly teach (bold only) “before performing the voting, for each node of the plurality of nodes of the cluster: performing consistency voting on the student predictions from the plurality of respective student machine-learning models of the node to determine if the node will participate in the voting.”
Zhang teaches (bold only) “before performing the voting, for each node of the plurality of nodes of the cluster: performing consistency voting on the student predictions from the plurality of respective student machine-learning models of the node to determine if the node will participate in the voting”: Zhang, section 3, paragraph 5, “To design SVM ensemble with rejection option, a major issue is to decide under what conditions an instance should be accepted for classification and under what conditions it should be rejected. With classifier ensemble, a simple intuition can be exploited to implement rejection option. An ensemble of SVM classifiers forms a ‘committee of experts’, which can decide an abstention from decision based on the consensus from the members. Specifically, with a test example x ∈ ℝn, each member of the committee makes a prediction of x's label. With an ensemble of size M1, labels l1, l2, …, lM are predicted from the committee. In forming the decision from the M1 classifiers (committee members), a sample x is assigned the class for which there is a predefined consensus degree, or when at least t of the members are agreed on the label [performing consistency voting on the student predictions from the plurality of respective student machine-learning models of the node], where
PNG
media_image2.png
97
501
media_image2.png
Greyscale
Otherwise, the example is rejected. Obviously, t is a tunable threshold that control the rejection rate, and we use t to relate the consensus degree from the voting to confidence measure and abstain to classify ambiguous samples in classification [to determine if the node will participate].”
Zhang and Kim are analogous arts as they are both related to ensemble machine learning. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the consistency measure of Zhang with the teachings of Kim to arrive at the present invention, in order to improve reliability, as stated in Zhang, Section 1, paragraph 4, “The conformity from ensemble is thus linked to confidence of classification. The reject option can subsequently improve classification reliability and leave the control of classification accuracy to user.”
Regarding claim 11 and analogous claim 44:
Kim as modified by Wang and Zhang teaches “The method of claim 10.”
Zhang further teaches “wherein the consistency voting determines if the student predictions from the plurality of respective student machine-learning models satisfy a predetermined agreement threshold”: Zhang, section 3, paragraph 5, “To design SVM ensemble with rejection option, a major issue is to decide under what conditions an instance should be accepted for classification and under what conditions it should be rejected. With classifier ensemble, a simple intuition can be exploited to implement rejection option. An ensemble of SVM classifiers forms a ‘committee of experts’, which can decide an abstention from decision based on the consensus from the members. Specifically, with a test example x ∈ ℝn, each member of the committee makes a prediction of x's label. With an ensemble of size M1, labels l1, l2, …, lM are predicted from the committee. In forming the decision from the M1 classifiers (committee members), a sample x is assigned the class for which there is a predefined consensus degree [wherein the consistency voting determines if the student predictions from the plurality of respective student machine-learning models satisfy a predetermined agreement threshold], or when at least t of the members are agreed on the label, where
PNG
media_image2.png
97
501
media_image2.png
Greyscale
Otherwise, the example is rejected. Obviously, t is a tunable threshold that control the rejection rate, and we use t to relate the consensus degree from the voting to confidence measure and abstain to classify ambiguous samples in classification.”
Zhang and Kim are combinable for the rationale given under claim 10.
Claims 12–15 and 45–48 rejected under 35 U.S.C. 103 over Kim as modified by Wang in view of Ghosh et al., “Self Training with Ensemble of Teacher Models,” 2021, arXiv:2107.08211v1 (hereafter Ghosh).
Regarding claim 12 and analogous claim 45:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein each of the plurality of respective student machine-learning models is trained based on teacher predictions provided by a plurality of respective teacher machine-learning models associated with each of the plurality of respective student machine-learning models.”
Ghosh teaches “wherein each of the plurality of respective student machine-learning models is trained based on teacher predictions provided by a plurality of respective teacher machine-learning models associated with each of the plurality of respective student machine-learning models”: Ghosh, section 2, “Self-Training using Student-Teacher Models”: “This class of methods [Xie et al., 2020] for SSL iteratively use a trained (teacher) model to pseudo-label a set of unlabeled examples, and then re-train the model (now student) on the labelled plus the pseudo-labelled examples [trained based on teacher predictions]. Usually the same model assumes the dual role of the the [sic] student (as the learner) and the teacher (it generates labels, which are then used by itself as a student for learning)”; Ghosh, section 3, “Model Training”: “We train k separate models on each of the k samples of the training data [provided by a plurality of respective teacher machine-learning models associated with each of the plurality of respective student machine-learning models]. The models are also chosen to be different architectures with their separate parameters Θ1, Θ2, …, Θk. The unlabeled examples XU are then fed to each of the k trained models to infer their corresponding probability vectors. We obtain (fΘ1(x), fΘ2(x); … ; fΘk(x)) for each x ∈ XU, and fΘi(x) ∈ ℝc∀I”; Ghosh, section 3, “Pseudo-labeling”: “For assigning pseudo-labels to the unlabeled examples, we take the ensemble of the predictions of the individual k models.
PNG
media_image3.png
82
423
media_image3.png
Greyscale
The unlabeled examples are then sorted in decreasing order of the entropy of the ensemble prediction vectors fensmbl(x), and first p examples with the lowest entropy are selected. We
assign pseudo-labels ^y to these p examples as follows:
PNG
media_image4.png
42
425
media_image4.png
Greyscale
”
Ghosh and Kim are analogous arts as they are both related to machine learning ensembles. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the student-teacher training of Ghosh with the teachings of Kim to arrive at the present invention, in order to improve model accuracy, as stated in Ghosh, Abstract, “Our proposed algorithm carefully avoids common pitfalls in utilizing unlabeled data and leads to a more accurate and calibrated supervised model compared to vanilla self-training based student-teacher algorithms.”
Regarding claim 13 and analogous claim 46:
Kim as modified by Wang and Ghosh teaches “The method of claim 12.”
Ghosh further teaches “wherein a voting of the teacher predictions is used to label datapoints in a student training dataset”: Ghosh, section 3, “Model Training”: “We train k separate models on each of the k samples of the training data. The models are also chosen to be different architectures with their separate parameters Θ1, Θ2, …, Θk. The unlabeled examples XU are then fed to each of the k trained models to infer their corresponding probability vectors. We obtain (fΘ1(x), fΘ2(x); … ; fΘk(x)) for each x ∈ XU, and fΘi(x) ∈ ℝc∀I”; Ghosh, section 3, “Pseudo-labeling”: “For assigning pseudo-labels to the unlabeled examples, we take the ensemble of the predictions of the individual k models.
PNG
media_image3.png
82
423
media_image3.png
Greyscale
The unlabeled examples are then sorted in decreasing order of the entropy of the ensemble prediction vectors fensmbl(x), and first p examples with the lowest entropy are selected. We
assign pseudo-labels ^y to these p examples as follows:
PNG
media_image4.png
42
425
media_image4.png
Greyscale
[wherein a voting of the teacher predictions is used to label datapoints in a student training dataset]”
Ghosh and Kim are combinable for the rationale given under claim 12.
Regarding claim 14 and analogous claim 47:
Kim as modified by Wang and Ghosh teaches “The method of claim 13.”
Ghosh further teaches “wherein each of the plurality of respective student machine- learning models is trained using the student training dataset, as labeled”: Ghosh, section 2, “Self-Training using Student-Teacher Models”: “This class of methods [Xie et al., 2020] for SSL iteratively use a trained (teacher) model to pseudo-label a set of unlabeled examples, and then re-train the model (now student) on the labelled plus the pseudo-labelled examples [trained using the student training dataset, as labeled]. Usually the same model assumes the dual role of the the [sic] student (as the learner) and the teacher (it generates labels, which are then used by itself as a student for learning)”;
Ghosh and Kim are combinable for the rationale given under claim 12.
Regarding claim 15 and analogous claim 48:
Kim as modified by Wang and Ghosh teaches “The method of claim 12.”
Ghosh further teaches “wherein each of the plurality of respective teacher machine- learning models is trained using a respective subset of a labeled training dataset before the plurality of respective teacher machine-learning models provide the teacher predictions”: Ghosh, section 3, “Model Training”: “We train k separate models on each of the k samples of the training data [each of the plurality of respective teacher machine- learning models is trained using a respective subset of a labeled training dataset]. The models are also chosen to be different architectures with their separate parameters Θ1, Θ2, …, Θk. The unlabeled examples XU are then fed to each of the k trained models to infer their corresponding probability vectors [before the plurality of respective teacher machine-learning models provide the teacher predictions]. We obtain (fΘ1(x), fΘ2(x); … ; fΘk(x)) for each x ∈ XU, and fΘi(x) ∈ ℝc∀I”;
Ghosh and Kim are combinable for the rationale given under claim 12.
Claims 16–18 and 49–51 rejected under 35 U.S.C. 103 over Kim as modified by Wang in view of Ghosh and Miralles.
Regarding claim 16 and analogous claim 49:
Kim as modified by Wang teaches “The method of claim 1.”
Wang further teaches (bold only) “wherein the cluster machine-learning model of the cluster and other cluster machine-learning models of other clusters provide cluster predictions to label a global training dataset”: Wang, paragraph 0039, “Referring now to FIGS. 3 and 4, after the server 108 receives the evaluation results 112 for each of the clients 102 [the cluster machine-learning model of the cluster], the server can determine an optimal subset of models and assign weights for the combination of selected models. For example, as shown in FIG. 4 model 1 may be provided a 0.4 weight, model 3 may be provided a 0.6 weight, and model 2 may not be selected. Details of the determination of these weights are provided below.”
Wang and Kim are combinable for the rationale given under claim 1.
Kim as modified by Wang does not explicitly teach (bold only) “wherein the cluster machine-learning model of the cluster and other cluster machine-learning models of other clusters provide cluster predictions to label a global training dataset.”
Miralles teaches (bold only) “wherein the cluster machine-learning model of the cluster and other cluster machine-learning models of other clusters provide cluster predictions to label a global training dataset”: Miralles, paragraphs 0016-0019, “In at least one embodiment, the configuration method comprises, during said federated learning: sending, to the aggregation node of said at least one cluster, a request to learn the weights of the models of the nodes of said cluster with the weights of a global model to the set of nodes [the cluster machine-learning model of the cluster of the cluster and other cluster machine-learning models of other clusters]; receiving, from the aggregation node of said at least one cluster, the weights of said aggregate model of said cluster resulting from said learning; and updating the weights of the global model by aggregation of the received weights of the aggregate model of said at least one cluster.”
Miralles and Kim are analogous arts as they are both related to machine learning ensembles. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the federated learning clusters of Miralles with the teachings of Kim to arrive at the present invention, in order to improve the efficiency of the classification system, as stated in Miralles, paragraph 0007, “In recent years, the federated learning approach has attracted a lot of interest in many fields, such as healthcare, banking, industry 4.0 or smart cities, because it can help build better global models, while preserving the confidentiality of the local files (medical, financial files, etc.). It can provide a natural solution to the growing needs for personal data protection, while addressing the current technological challenges: decrease in the energy consumption, minimization of the latency, two challenges with the deployment of 5G technology. As explained above, the federated learning is thus a form of distributed learning where several nodes collaboratively solve a machine learning task.”
Ghosh teaches (bold only) “wherein the cluster machine-learning model of the cluster and other cluster machine-learning models of other clusters provide cluster predictions to label a global training dataset”: Ghosh, section 3, “Model Training”: “We train k separate models on each of the k samples of the training data. The models are also chosen to be different architectures with their separate parameters Θ1, Θ2, …, Θk. The unlabeled examples XU are then fed to each of the k trained models to infer their corresponding probability vectors. We obtain (fΘ1(x), fΘ2(x); … ; fΘk(x)) for each x ∈ XU, and fΘi(x) ∈ ℝc∀I [machine-learning models … provide … predictions to label a global training dataset].”
Ghosh and Kim are analogous arts as they are both related to machine learning ensembles. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the student-teacher training of Ghosh with the teachings of Kim to arrive at the present invention, in order to improve model accuracy, as stated in Ghosh, Abstract, “Our proposed algorithm carefully avoids common pitfalls in utilizing unlabeled data and leads to a more accurate and calibrated supervised model compared to vanilla self-training based student-teacher algorithms.”
Regarding claim 17 and analogous claim 50:
Kim as modified by Wang, Ghosh, and Miralles teaches “The method of claim 16.”
Ghosh further teaches “the global training dataset is used to train a global machine-learning model”: Ghosh, Fig. 1,
PNG
media_image5.png
361
600
media_image5.png
Greyscale
[showing that the pseudo-labelled data produced by the ensemble is ultimately used to retrain all (i.e., globally) the models of the ensemble, hence, the global training dataset is used to train a global machine-learning model].
Ghosh and Kim are combinable for the rationale given under claim 16.
Miralles further teaches “and model weights of the global machine-learning model, as trained, are provided to the plurality of nodes of the cluster and other nodes of the other clusters”: Miralles, paragraphs 0137-0138, “In one particular embodiment, the aggregate model of each cluster is sent to the cluster of lower level, for example conditionally, such as after a constant number of iterations. The aggregate models can thus go up to the coordination entity which can aggregate these models into an updated version of the overall model. This global model can then go down to all the nodes for a new implementation of the method either directly or via the aggregation nodes [and model weights of the global machine-learning model, as trained, are provided to the plurality of nodes of the cluster and other nodes of the other clusters].”
Miralles and Kim are combinable for the rationale given under claim 16.
Regarding claim 18 and analogous claim 51:
Kim as modified by Wang, Ghosh, and Miralles teaches “The method of claim 16.”
Kim further teaches (bold only) “wherein a voting of the cluster predictions is used to label datapoints in the global training dataset”: Kim, paragraph 0032, “A pseudo-label corresponding to each feature is determined at S230 based on the respective pseudo-labels output by each model based on the feature. According to system 400, label determination component 420 receives all labels output by each of the plurality of trained models, and generates pseudo-labels 430 for each of feature ul-um based thereon. For z=i to m, label determination component 420 determines pseudo-label pz corresponding to feature uz based on each pseudo-label which was output by trained models 120-1, 120-2 and 120-e based on feature uz (i.e., pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e). Pseudo-label pz1 may be determined based on pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e in any suitable manner, including but not limited to majority voting (i.e., choosing the most-often occurring value of the pseudo-labels) [a voting of the … predictions is used to label datapoints in the global training dataset] or averaging the softmax outputs of each trained model.”
Wang further teaches (bold only) “wherein a voting of the cluster predictions is used to label datapoints in the global training dataset”: Wang, paragraph 0039, “Referring now to FIGS. 3 and 4, after the server 108 receives the evaluation results 112 for each of the clients 102 [cluster predictions], the server can determine an optimal subset of models and assign weights for the combination of selected models. For example, as shown in FIG. 4 model 1 may be provided a 0.4 weight, model 3 may be provided a 0.6 weight, and model 2 may not be selected. Details of the determination of these weights are provided below.”
Wang and Kim are combinable for the rationale given under claim 1.
Claims 19–20 and 52–53 rejected under 35 U.S.C. 103 over Kim as modified by Wang in view of Miralles and Cook, US Pre-Grant Publication No. 2022/0343221 (hereafter Cook).
Regarding claim 19 and analogous claim 52:
Kim as modified by Wang teaches “The method of claim 1.”
Kim further teaches (bold only) “the cluster machine-learning model of the cluster and other cluster machine-learning models of other clusters provide cluster predictions to label an industry-specific training dataset for an industry based on a weighted voting of the cluster predictions”: Kim, paragraph 0032, “A pseudo-label corresponding to each feature is determined at S230 based on the respective pseudo-labels output by each model based on the feature. According to system 400, label determination component 420 receives all labels output by each of the plurality of trained models, and generates pseudo-labels 430 for each of feature ul-um based thereon. For z=i to m, label determination component 420 determines pseudo-label pz corresponding to feature uz based on each pseudo-label which was output by trained models 120-1, 120-2 and 120-e based on feature uz (i.e., pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e). Pseudo-label pz1 may be determined based on pseudo-label pz-1, pseudo-label pz-2, and pseudo-label pz-e in any suitable manner, including but not limited to majority voting (i.e., choosing the most-often occurring value of the pseudo-labels) [provide … predictions to label an … training dataset … based on a weighted voting of the … predictions] or averaging the softmax outputs of each trained model.”
Wang further teaches (bold only) “the cluster machine-learning model of the cluster and other cluster machine-learning models of other clusters provide cluster predictions to label an industry-specific training dataset for an industry based on a weighted voting of the cluster predictions”: Wang, paragraph 0039, “Referring now to FIGS. 3 and 4, after the server 108 receives the evaluation results 112 for each of the clients 102 [the cluster machine-learning model of the cluster], the server can determine an optimal subset of models and assign weights for the combination of selected models. For example, as shown in FIG. 4 model 1 may be provided a 0.4 weight, model 3 may be provided a 0.6 weight, and model 2 may not be selected. Details of the determination of these weights are provided below.”
Wang and Kim are combinable for the rationale given under claim 1.
Kim as modified by Wang does not explicitly teach:
(bold only) “the cluster machine-learning model of the cluster and other cluster machine-learning models of other clusters provide cluster predictions to label an industry-specific training dataset for an industry based on a weighted voting of the cluster predictions.”
“and a subset of the cluster predictions are weighted heavier in the weighted voting when the subset of the cluster predictions are provided by clusters associated with the industry”
Miralles teaches (bold only) “the cluster machine-learning model of the cluster and other cluster machine-learning models of other clusters provide cluster predictions to label an industry-specific training dataset for an industry based on a weighted voting of the cluster predictions”: Miralles, paragraphs 0016-0019, “In at least one embodiment, the configuration method comprises, during said federated learning: sending, to the aggregation node of said at least one cluster, a request to learn the weights of the models of the nodes of said cluster with the weights of a global model to the set of nodes [the cluster machine-learning model of the cluster of the cluster and other cluster machine-learning models of other clusters]; receiving, from the aggregation node of said at least one cluster, the weights of said aggregate model of said cluster resulting from said learning; and updating the weights of the global model by aggregation of the received weights of the aggregate model of said at least one cluster.”
Miralles and Kim are analogous arts as they are both related to machine learning ensembles. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the federated learning clusters of Miralles with the teachings of Kim to arrive at the present invention, in order to improve the efficiency of the classification system, as stated in Miralles, paragraph 0007, “In recent years, the federated learning approach has attracted a lot of interest in many fields, such as healthcare, banking, industry 4.0 or smart cities, because it can help build better global models, while preserving the confidentiality of the local files (medical, financial files, etc.). It can provide a natural solution to the growing needs for personal data protection, while addressing the current technological challenges: decrease in the energy consumption, minimization of the latency, two challenges with the deployment of 5G technology. As explained above, the federated learning is thus a form of distributed learning where several nodes collaboratively solve a machine learning task.”
Cook teaches (bold only) “the cluster machine-learning model of the cluster and other cluster machine-learning models of other clusters provide cluster predictions to label an industry-specific training dataset for an industry based on a weighted voting of the cluster predictions” and “and a subset of the cluster predictions are weighted heavier in the weighted voting when the subset of the cluster predictions are provided by clusters associated with the industry”: Cook, paragraph 0129, “In another example, multiple model types can be created in accordance with the embodiments disclosed herein. In one example, a first model may include a first set of predictands and/or parameters, whereas a second model may include a second set of predictands and/or parameters that are different from the first model. The first model may be executed against a first domain. The second model may be executed against a second domain. As noted above, these domains may be geographically independent, or may overlap geographically. A resulting forecast or disaster model/prediction may include an ensemble of these two model outputs. In some embodiments, the ensemble can include an average or other statistical calculation or operation performed on the model outputs. In one example, the systems and methods may determine a weighted average, where one or more of the model outputs for one or more of the domains are weighted. In one embodiment, the systems and methods could apply weighting to models based on proximity to a specific area of interest. For example, a domain that is closest to a desired area of prediction may have its model output weighted higher than areas that are farther away or geographical areas having general climatological difference that are distinct from the domain of interest [a subset of the cluster predictions are weighted heavier in the weighted voting when the subset of the cluster predictions are provided by clusters associated with the industry]. To be sure, any number of different models can be executed against any number of overlapping or non-overlapping domains.”
Cook and Kim are analogous arts as they are both related to weighted ensembles. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the domain-based weighting of Cook with the teachings of Kim to arrive at the present invention, in order to construct a model specific to a particular domain, as stated in Cook, paragraph 0129, “In one example, the systems and methods may determine a weighted average, where one or more of the model outputs for one or more of the domains are weighted. In one embodiment, the systems and methods could apply weighting to models based on proximity to a specific area of interest.”
Regarding claim 20 and analogous claim 53:
Kim as modified by Wang, Miralles, and Cook teaches “The method of claim 19.”
Cook further teaches (bold only) “the industry-specific training dataset is used to train an industry-specific machine learning model for the industry; and model weights of the industry-specific machine-learning model, as trained, are provided to nodes of the clusters associated with the industry”: Cook, paragraph 0129, “In another example, multiple model types can be created in accordance with the embodiments disclosed herein. In one example, a first model may include a first set of predictands and/or parameters, whereas a second model may include a second set of predictands and/or parameters that are different from the first model. The first model may be executed against a first domain. The second model may be executed against a second domain. As noted above, these domains may be geographically independent, or may overlap geographically. A resulting forecast or disaster model/prediction may include an ensemble of these two model outputs. In some embodiments, the ensemble can include an average or other statistical calculation or operation performed on the model outputs. In one example, the systems and methods may determine a weighted average, where one or more of the model outputs for one or more of the domains are weighted. In one embodiment, the systems and methods could apply weighting to models based on proximity to a specific area of interest. For example, a domain that is closest to a desired area of prediction may have its model output weighted higher than areas that are farther away or geographical areas having general climatological difference that are distinct from the domain of interest. To be sure, any number of different models can be executed against any number of overlapping or non-overlapping domains [train an industry-specific machine learning model for the industry].”
Cook and Kim are combinable for the rationale given under claim 19.
Miralles further teaches (bold only) “the industry-specific training dataset is used to train an industry-specific machine learning model for the industry; and model weights of the industry-specific machine-learning model, as trained, are provided to nodes of the clusters associated with the industry”: Miralles, paragraphs 0137-0138, “In one particular embodiment, the aggregate model of each cluster is sent to the cluster of lower level, for example conditionally, such as after a constant number of iterations. The aggregate models can thus go up to the coordination entity which can aggregate these models into an updated version of the overall model. This global model can then go down to all the nodes for a new implementation of the method either directly or via the aggregation nodes [model weights …, as trained, are provided to nodes of the clusters].”
Miralles and Kim are combinable for the rationale given under claim 19.
Claims 22, 30, 55, and 63 rejected under 35 U.S.C. 103 over Kim as modified by Wang in view of Sivaramakrishnan et al., US Patent No. 11,182,221 (hereafter Sivaramakrishnan).
Regarding claim 22 and analogous claim 55:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein at least one of the plurality of nodes comprises a SmartNIC.”
Sivaramakrishnan teaches “wherein at least one of the plurality of nodes comprises a SmartNIC“: Sivaramakrishnan, col. 28, lines 60-63, “The respective processing nodes are also operatively coupled to respective pluralities of Network Interface Controllers (NICs) or Smart Network Interface Controllers (SmartNICs).”
Sivaramakrishnan and Kim as modified by Miralles are analogous arts as they are both related to network nodes. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the identification of a SmartNIC as a network node of Sivaramakrishnan with the teachings of Kim to arrive at the present invention, in order to apply the node-based methods to a SmartNIC in particular, as stated in Sivaramakrishnan, col. 28, lines 60-63, “The respective processing nodes are also operatively coupled to respective pluralities of Network Interface Controllers (NICs) or Smart Network Interface Controllers (SmartNICs).”
Regarding claim 30 and analogous claim 63:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein at least one of the plurality of nodes comprises a virtual device context.”
Sivaramakrishnan teaches “wherein at least one of the plurality of nodes comprises a virtual device context”: Sivaramakrishnan, col 6., lines 55-60, “A processing node ( or node) is an addressable application running on a hardware device or virtual device that attaches to a network, and is capable of sending, receiving, or forwarding information over a communications channel to 60 or from other processing nodes.”
Sivaramakrishnan and Kim as modified by Miralles are analogous arts as they are both related to network nodes. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the identification of a virtual device as a network node of Sivaramakrishnan with the teachings of Kim to arrive at the present invention, in order to apply the node-based methods to a virtual device in particular, as stated in Sivaramakrishnan, col 6., lines 55-60, “A processing node ( or node) is an addressable application running on a hardware device or virtual device that attaches to a network, and is capable of sending, receiving, or forwarding information over a communications channel to 60 or from other processing nodes.”
Claims 23-26, 28-29, 31-32, 56-59, 61-62, and 64-65 rejected under 35 U.S.C. 103 over Kim as modified by Wang in view of Roberts et al., US Pre-Grant Publication No. 2020/0344146 (hereafter Roberts).
Regarding claim 23 and analogous claim 56:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein at least one of the plurality of nodes comprises a router.”
Roberts teaches “wherein at least one of the plurality of nodes comprises a router”: Roberts, paragraph 0007, “In another example aspect, a system includes a first control node assigned a first zone identifier; a second control node assigned a second zone identifier; a first workload configured to establish a first routing session with the first control node based, at least in part, on the first zone identifier, wherein the first workload is configured as a primary provider of a service; and a second workload configured to establish a second routing session with the second control node based, at least in part, on the second zone identifier, wherein the second workload is configured as a secondary provider of the service; wherein the first workload is further configured to receive one or more first service requests via a virtual router managed by the first control node.”
Roberts and Kim as modified by Wang are analogous arts as they are both related to network nodes. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the identification of a router as a network node of Roberts with the teachings of Kim to arrive at the present invention, in order to apply the node-based methods to a router in particular, as stated in Roberts, paragraph 0007, “In another example aspect, a system includes a first control node assigned a first zone identifier; a second control node assigned a second zone identifier; a first workload configured to establish a first routing session with the first control node based, at least in part, on the first zone identifier, wherein the first workload is configured as a primary provider of a service; and a second workload configured to establish a second routing session with the second control node based, at least in part, on the second zone identifier, wherein the second workload is configured as a secondary provider of the service; wherein the first workload is further configured to receive one or more first service requests via a virtual router managed by the first control node.”
Regarding claim 24 and analogous claim 57:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein at least one of the plurality of nodes comprises a switch.”
Roberts teaches “wherein at least one of the plurality of nodes comprises a switch”: Roberts, paragraph 0031, “In some examples, chassis switches 18 may operate as spine nodes and TOR switches 16 may operate as leaf nodes in data center 10A.”
Roberts and Kim as modified by Wang are analogous arts as they are both related to network nodes. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the identification of a switch as a network node of Roberts with the teachings of Kim to arrive at the present invention, in order to apply the node-based methods to a switch in particular, as stated in Roberts, paragraph 0031, “In some examples, chassis switches 18 may operate as spine nodes and TOR switches 16 may operate as leaf nodes in data center 10A.”
Regarding claim 25 and analogous claim 58:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein at least one of the plurality of nodes comprises a firewall.”
Roberts teaches “wherein at least one of the plurality of nodes comprises a firewall”: Roberts, paragraph 0004, “For example, a first compute node may be configured as a primary provider of a firewall service while a second compute node may be configured as a secondary or backup firewall service to the primary firewall service.”
Roberts and Kim as modified by Wang are analogous arts as they are both related to network nodes. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the identification of a firewall as a network node of Roberts with the teachings of Kim to arrive at the present invention, in order to apply the node-based methods to a firewall in particular, as stated in Roberts, paragraph 0004, “For example, a first compute node may be configured as a primary provider of a firewall service while a second compute node may be configured as a secondary or backup firewall service to the primary firewall service.”
Regarding claim 26 and analogous claim 59:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein at least one of the plurality of nodes comprises a load balancer.”
Roberts teaches “wherein at least one of the plurality of nodes comprises a load balancer”:
Roberts, paragraph 0036, “For example, SDN controller 132 implements high-level requests from orchestration engine 130 by configuring physical switches, e.g. TOR switches 16, chassis switches 18, and switch fabric 20; physical routers; physical service nodes such as firewalls and load balancers; and virtual services such as virtual firewalls in a virtualized environment.”
Roberts and Kim as modified by Wang are analogous arts as they are both related to network nodes. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the identification of a load balancer as a network node of Roberts with the teachings of Kim to arrive at the present invention, in order to apply the node-based methods to a load balancer in particular, as stated in Roberts, paragraph 0036, “For example, SDN controller 132 implements high-level requests from orchestration engine 130 by configuring physical switches, e.g. TOR switches 16, chassis switches 18, and switch fabric 20; physical routers; physical service nodes such as firewalls and load balancers; and virtual services such as virtual firewalls in a virtualized environment.”
Regarding claim 28 and analogous claim 61:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein at least one of the plurality of nodes comprises a processing container running on a network device.”
Roberts teaches “wherein at least one of the plurality of nodes comprises a processing container running on a network device”: Roberts, paragraph 0045, “The control nodes having one zone identifier can be on separate processes and/or physical hardware from control nodes having a different zone identifier to reduce the likelihood of a single point of failure. Workloads, such as virtual machines or containers, can establish routing sessions such as Border Gateway Protocol as a Service (BGPaaS) routing sessions using different zone identifiers to ensure that separate control nodes provide routing management services for the primary and secondary compute nodes or primary and secondary workloads associated with a high availability service.”
Roberts and Kim as modified by Wang are analogous arts as they are both related to network nodes. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the identification of a processing container as a network node of Roberts with the teachings of Kim to arrive at the present invention, in order to apply the node-based methods to a processing container in particular, as stated in Roberts, paragraph 0045, “The control nodes having one zone identifier can be on separate processes and/or physical hardware from control nodes having a different zone identifier to reduce the likelihood of a single point of failure. Workloads, such as virtual machines or containers, can establish routing sessions such as Border Gateway Protocol as a Service (BGPaaS) routing sessions using different zone identifiers to ensure that separate control nodes provide routing management services for the primary and secondary compute nodes or primary and secondary workloads associated with a high availability service.”
Regarding claim 29 and analogous claim 62:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein at least one of the plurality of nodes comprises a virtual machine.”
Roberts teaches “wherein at least one of the plurality of nodes comprises a virtual machine”: Roberts, paragraph 0045, “The control nodes having one zone identifier can be on separate processes and/or physical hardware from control nodes having a different zone identifier to reduce the likelihood of a single point of failure. Workloads, such as virtual machines or containers, can establish routing sessions such as Border Gateway Protocol as a Service (BGPaaS) routing sessions using different zone identifiers to ensure that separate control nodes provide routing management services for the primary and secondary compute nodes or primary and secondary workloads associated with a high availability service.”
Roberts and Kim as modified by Wang are analogous arts as they are both related to network nodes. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the identification of a virtual machine as a network node of Roberts with the teachings of Kim to arrive at the present invention, in order to apply the node-based methods to a virtual machine in particular, as stated in Roberts, paragraph 0045, “The control nodes having one zone identifier can be on separate processes and/or physical hardware from control nodes having a different zone identifier to reduce the likelihood of a single point of failure. Workloads, such as virtual machines or containers, can establish routing sessions such as Border Gateway Protocol as a Service (BGPaaS) routing sessions using different zone identifiers to ensure that separate control nodes provide routing management services for the primary and secondary compute nodes or primary and secondary workloads associated with a high availability service.”
Regarding claim 31 and analogous claim 64:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein at least one of the plurality of nodes comprises an operating system.”
Roberts teaches “wherein at least one of the plurality of nodes comprises an operating system”: Roberts, paragraph 0068, “The failure may be a total failure, such as a loss of power to a server hosting the control node or an operating system crash on a physical or virtual machine hosting the control node.”
Roberts and Kim as modified by Wang are analogous arts as they are both related to network nodes. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the identification of an operating system as a network node of Roberts with the teachings of Kim to arrive at the present invention, in order to apply the node-based methods to an operating system in particular, as stated in Roberts, paragraph 0068, “The failure may be a total failure, such as a loss of power to a server hosting the control node or an operating system crash on a physical or virtual machine hosting the control node.”
Regarding claim 32 and analogous claim 65:
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein at least one of the plurality of nodes comprises an application.”
Roberts teaches “wherein at least one of the plurality of nodes comprises an application”: Roberts, paragraph 0002, “In a typical cloud data center environment, a large collection of interconnected servers provide computing (e.g., compute nodes) and/or storage capacity to run various applications.”
Roberts and Kim as modified by Wang are analogous arts as they are both related to network nodes. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the identification of an application as a network node of Roberts with the teachings of Kim to arrive at the present invention, in order to apply the node-based methods to an application in particular, as stated in Roberts, paragraph 0002, “In a typical cloud data center environment, a large collection of interconnected servers provide computing (e.g., compute nodes) and/or storage capacity to run various applications.”
Claims 27 and analogous claim 60 rejected under 35 U.S.C. 103 over Kim as modified by Wang in view of Ma, US Pre-Grant Publication No. 2019/0149530 (hereafter Ma).
Kim as modified by Wang teaches “The method of claim 1.”
Kim as modified by Wang does not explicitly teach “wherein at least one of the plurality of nodes comprises an IoT device.”
Ma teaches “wherein at least one of the plurality of nodes comprises an IoT device”: Ma, paragraph 0007, “In one aspect of the present disclosure, a method is provided for deploying an IoT device node in an IoT network, and the IoT device node has a primary communication link capable of communicating with a management server and a secondary communication link capable of communicating with a deployment device.”
Ma and Kim as modified by Wang are analogous arts as they are both related to network nodes. It would have been obvious to a person having ordinary skill in the art prior to the effective filing date of the claimed invention to have combined the identification of an IoT device as a network node of Roberts with the teachings of Ma to arrive at the present invention, in order to apply the node-based methods to an IoT device in particular, as stated in Ma, paragraph 0007, “In one aspect of the present disclosure, a method is provided for deploying an IoT device node in an IoT network, and the IoT device node has a primary communication link capable of communicating with a management server and a secondary communication link capable of communicating with a deployment device.”
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Lin et al., “Ensemble Distillation for Robust Model Fusion in Federated Learning,” 2021, arXiv:2006.07242v3, discloses a method of federated learning in which trains the central model using heterogenous client models that results in fewer training rounds.
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 VINCENT SPRAUL whose telephone number is (703)756-1511. The examiner can normally be reached M-F 9:00 am - 5:00 pm.
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, MICHAEL HUNTLEY can be reached at (303) 297-4307. 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.
/VAS/ Examiner, Art Unit 2129
/MICHAEL J HUNTLEY/Supervisory Patent Examiner, Art Unit 2129
Prosecution Insights