DETAILED ACTION
This communication is in response to the application filed 6/6/24 in which claims 1-20 were presented for examination.
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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 8/5/26 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Allowable Subject Matter
Claims 6-12, 18, and 19 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or non-obviousness.
Claims 1, 3, 4, and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Chen, Cheng, et al. "Fedcluster: Boosting the convergence of federated learning via cluster-cycling." 2020 IEEE international conference on big data (Big Data). IEEE, 2020 (“Chen”) in view of Zhang, Ying, et al. "Deep mutual learning." 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, 2018 (“Zhang”).
Regarding claim 1, Chen discloses [a] system for facilitating a distributed learning and activation model for a machine learning program in an environment, the system comprising:
a plurality of computing nodes in networked communication, the plurality of computing nodes comprising a first set of computing nodes and a second set of computing nodes, (Chen Section II FedCluster: Federated Learning with Cluster-Cycling (“To improve the optimization efficiency and flexibility of federated learning, we propose the FedCluster framework as illustrated in Fig. 1 (Right). To elaborate, in the FedCluster system, devices are grouped into multiple clusters using a certain clustering approach (elaborated later).”
PNG
media_image1.png
466
546
media_image1.png
Greyscale
))
wherein the first set of computing nodes manages a first set of machine learning models and the second set of computing node manages a second set of machine learning models, and (see Chen Fig.1 (right) (each cluster includes at least two nodes)).
Chen does not expressly teach:
wherein a first computing node in the first set of computing nodes and a second computing node in the second set of computing nodes cause mutual learning between two or more of the plurality of nodes by exchanging information in near real time to update the first set of machine learning models and the second set of machine learning models (but see Zhang page 3 Section 2.2 Optimization (“The mutual learning strategy is performed in each mini-batch based model update step and throughout the whole training process. At each iteration, we compute the predictions of the two models and update both networks’ parameters according to the predictions of the other. The optimization of Θ1 and Θ2 is conducted iteratively until convergence.”)).
It would have been obvious to one of ordinary skill at the time of the claimed invention to have modified Chen to incorporate the teachings of Zhang to apply mutual learning between a model in the first cluster and the second cluster throughout the training process, at least because mutual learning provides a simple but effective way to improve the generalization ability of a network by training collaboratively with a cohort of other networks. See Zhang page 2.
Regarding claim 3, Chen, in view of Zhang, discloses the invention of claim 1 as discussed above. Chen further discloses wherein updating the first set of machine learning models and the second set of machine learning models occurs independent of delays and packet drops associated with a network used to exchange the information (Chen Section II 3) Availability-based clustering (“Availability-based clustering: A more general approach is to divide each learning round into multiple time slots. Each device determines its available time slot to perform federated learning. Then, the available devices within each time slot form a cluster.”)).
Regarding claim 4, Chen, in view of Zhang, discloses the invention of claim 1 as discussed above. Chen further discloses wherein the first computing node and the second computing node are configured to variably update the distributed learning and activation model (Chen Section II (“Specifically, as illustrated in Fig. 1 (Right), in the first cycle of a learning round, a subset of devices of Cluster 1 are activated to perform federated learning, i.e., they download a model from the cloud server, perform local trainings using a certain algorithm (e.g., SGD) and upload the trained local models to the cloud for model averaging. Then, in the next cycle, a subset of devices of Cluster 2 are activated to perform federated learning. Following this strategy, all the clusters take turns to perform federated learning in a cyclic order.”)).
Regarding claim 13, Chen, in view of Zhang, discloses the invention of claim 1 as discussed above. Chen further discloses wherein each of the plurality of computing nodes is configured to update weights of instances associated with the first set of machine learning models and instances associated with the second set of machine learning models (Chen Section II (“In each learning round, the system performs multiple cycles of federated learning through the clusters in a cyclical way. Specifically, as illustrated in Fig. 1 (Right), in the first cycle of a learning round, a subset of devices of Cluster 1 are activated to perform federated learning, i.e., they download a model from the cloud server, perform local trainings using a certain algorithm (e.g., SGD) and upload the trained local models to the cloud for model averaging. Then, in the next cycle, a subset of devices of Cluster 2 are activated to perform federated learning.”) (federated learning involves updating weights associated with the model clusters)).
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over Chen and Zhang as applied to claim 1 above, and further in view of Sattler, Felix, Klaus-Robert Müller, and Wojciech Samek. "Clustered federated learning: Model-agnostic distributed multitask optimization under privacy constraints." IEEE transactions on neural networks and learning systems 32.8 (2020): 3710-3722 (“Sattler”).
Regarding claim 2, Chen, in view of Zhang, discloses the invention of claim 1 as discussed above. Chen and Zhang do not expressly teach wherein the first set of machine learning models is a superset of the second set of machine learning models, and wherein the first set of machine learning models is configured to be executed by at least a portion of the second set of machine learning models (but see Sattler Section 1 (“FEDERATED learning (FL) [1]–[5] is a distributed training framework, which allows multiple clients (typically, mobile or the IoT devices) to jointly train a single deep learning model on their combined data in a communication-efficient way, without requiring any of the participants to reveal their private training data to a centralized entity or to each other. FL realizes this goal via an iterative three-step protocol where, in every communication round t, the clients first synchronize with the server by downloading the latest master model θt. Every client then proceeds to improve the down loaded model by performing multiple iterations of stochastic gradient descent with minibatches sampled from its local data Di, resulting in a weight-update vector
PNG
media_image2.png
62
662
media_image2.png
Greyscale
Finally, all clients upload their computed weight-updates to the server, where they are aggregated by weighted averaging according to
PNG
media_image3.png
96
632
media_image3.png
Greyscale
to create the next master model (see [1]). The procedure is summarized in Algorithm 2.”)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen and Zhang to incorporate the teachings of Sattler to update the global model with the client models obtained by mutual learning for execution on the server, at least because doing so would enable training a single deep learning model on the combined data of the models belonging to each cluster in a communication-efficient way.
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Chen and Zhang as applied to claim 1 above, and further in view of Boulanger (US 2021/0150267 A1; published May 20, 2021).
Regarding claim 5, Chen, in view of Zhang, discloses the invention of claim 1 as discussed above. Chen and Zhang do not expressly teach: one or more databases for storing information corresponding to the environment; and object data associated with a plurality of objects, wherein the environment is a simulation environment (but see Boulanger ¶ 22 (“In various embodiments, the environment simulation system 102 includes a scene generator 104 and an infrared sensor simulator 106. The scene generator 104 is operable to build a virtual 3D environment utilizing data from an object database 112, which stores 3D models and other object data allowing a modeled 3D object to be placed in the scene. The scene generator 104 may also apply environmental effects 114 (such as weather conditions, temperature, time of day, etc.). The environment simulation system 102 may optionally include an optical image capture component 108 for capturing optical images of a generated scene.”)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen and Zhang to incorporate the teachings of Boulanger to store environment information and object data related to an environment simulation system, at least because doing so would enable training a neural network to classify images by training on a synthetic training dataset.
Claim 14 is rejected under 35 U.S.C. 103 as being unpatentable over Chen, Zhang, and Boulanger as applied to claim 5 above, and further in view of Revanth, N. R., and P. J. Narayanan. "Distributed massive model rendering." Proceedings of the Eighth Indian Conference on Computer Vision, Graphics and Image Processing. 2012 (“Revanth”).
Regarding claim 14, Chen, in view of Zhang, discloses the invention of claim 5 as discussed above. Chen and Zhang do not expressly teach wherein the first computing node and the second computing node are configured to negotiate which of the plurality of objects to select for rendering at the environment (but see Revanth Section 3.3 Distributed Rendering Pipeline (“The regular graphics pipeline does the frustum culling followed by rendering. However, we split this pipeline into four parallel modules (Figure 1) where each module is assigned to a different GPU. 1. Load balanced frustum division: The server handles this module. Here the scene is frustum culled and the frustum is divided into sub-frustums based on the objects in the frustum using a load balanced frustum division algorithm. 2. Visibility determination: This module runs on the first GPU of each client. Hardware Occlusion queries are used to identify the visible objects in each client’s sub-frustum. 3. Sub-frame rendering: This module runs on the second GPU of each client. The visible objects identified by the previous module are rendered to a frame buffer with the corresponding sub-frustum parameters. 4. Assemble Frame: The sub-frames rendered in the previous module are transferred back to the server to be assembled into a complete frame buffer.”)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen and Zhang to incorporate the teachings of Revanth to render scene objects in distributed fashion based on load balancing, at least because doing so would enable handling massive models. See Revanth Section 3 Distributed Massive Model Rendering.
Claims 15 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Chen, in view of Zhang and Wu, Chuhan, et al. "Communication-efficient federated learning via knowledge distillation." Nature communications 13.1 (2022): 2032 (“Wu”).
Regarding claim 15, Chen discloses [a] method for facilitating a distributed learning and activation model for a machine learning program in an environment, the method comprising:
providing a plurality of computing nodes in networked communication, the plurality of computing nodes comprising a first set of computing nodes and a second set of computing nodes; (Chen Section II FedCluster: Federated Learning with Cluster-Cycling (“To improve the optimization efficiency and flexibility of federated learning, we propose the FedCluster framework as illustrated in Fig. 1 (Right). To elaborate, in the FedCluster system, devices are grouped into multiple clusters using a certain clustering approach (elaborated later).”
PNG
media_image1.png
466
546
media_image1.png
Greyscale
))
managing a first set of machine learning models by a first computing node in the first set of computing nodes and managing a second set of machine learning models by a second computing node in the second set of computing nodes; (see Chen Fig.1 (right) (each cluster includes at least two nodes)).
enabling a mutual learning process between the first computing node and the second computing node by: (but see Zhang Section 2.2 Optimization (“The mutual learning strategy is performed in each mini-batch based model update step and throughout the whole training process. At each iteration, we compute the predictions of the two models and update both networks’ parameters according to the predictions of the other. The optimization of Θ1 and Θ2 is conducted iteratively until convergence. The optimization details are summarized in Algorithm1.
PNG
media_image4.png
544
980
media_image4.png
Greyscale
”)).
It would have been obvious to one of ordinary skill at the time of the claimed invention to have modified Chen to incorporate the teachings of Zhang to apply mutual learning between a model in the first cluster and the second cluster throughout the training process, at least because mutual learning provides a simple but effective way to improve the generalization ability of a network by training collaboratively with a cohort of other networks. See Zhang page 2.
Chen further discloses:
providing, by the first computing node and in near real time, a first set of information to dynamically update the first set of machine learning models, and (Chen Section II (“In each learning round, the system performs multiple cycles of federated learning through the clusters in a cyclical way. Specifically, as illustrated in Fig. 1 (Right), in the first cycle of a learning round, a subset of devices of Cluster 1 are activated to perform federated learning, i.e., they download a model from the cloud server, perform local trainings using a certain algorithm (e.g., SGD) and upload the trained local models to the cloud for model averaging.”))
providing, by the second computing node and in near real time, a second set of information to dynamically update the second set of machine learning models; (Chen Section II (“Then, in the next cycle, a subset of devices of Cluster 2 are activated to perform federated learning. Following this strategy, all the clusters take turns to perform federated learning in a cyclic order.”)).
Chen does not expressly teach:
wherein the mutual learning process is performed using compression when providing the first set of information and providing the second set of information, wherein the compression is a reversible compression between the first computing node and the second computing node and configured to reduce computational effort on the first computing node while reducing bandwidth usage when communicating between the first computing node and the second computing node (but see Wu page 2 2nd column (“In this work, we present a communication-efficient federated learning method based on knowledge distillation, named FedKD. Our method is mainly focused on cross-silo federated learning where the clients have relatively richer computing resources and larger local data volume than personal devices. Instead of directly communicating the large models between the clients and server, in FedKD there are a small model (mentee) and a large model (mentor) that learn and distill knowledge from each other, where only the mentee model is shared by different clients and learned collaboratively, which can effectively reduce the communication cost. In this way, different clients have different local mentor models, which can better adapt to the characteristics of local datasets to achieve personalized model learning. Motivated by the Protégé Effect29 ("while we teach, we learn”, Seneca the Younger) in human learning, we propose an adaptive mutual distillation method to enable the local mentor and mentee models to reciprocally learn from the knowledge distilled from their predicted soft labels and intermediate results, where the distillation intensity is self adaptively controlled by the correctness of their predictions.”)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen and Zhang to incorporate the teachings of Wu to distill the model information when exchanging model weights between clients, at least because doing so would reduce the communication cost (i.e., bandwidth).
Regarding claim 17, Chen, in view of Zhang and Wu, discloses the invention of claim 15 as discussed above. Chen further discloses wherein the first computing node and the second computing node variably update the first set of machine learning models or the second set of machine learning models, and the updates to the first set of machine learning models and the second set of machine learning models occur independent of delays and packet drops associated with a network used to exchange the first set of information and the second set of information (Chen Section II 3) Availability-based clustering (“Availability-based clustering: A more general approach is to divide each learning round into multiple time slots. Each device determines its available time slot to perform federated learning. Then, the available devices within each time slot form a cluster.”)).
Claim 16 is rejected under 35 U.S.C. 103 as being unpatentable over Chen, Zhang, and Wu as applied to claim 15 above, and further in view of Sattler.
Regarding claim 16, Chen, in view of Zhang and Wu, discloses the invention of claim 15 as discussed above. Chen, Zhang, and Wu do not expressly teach wherein the first set of machine learning models is a superset of the second set of machine learning models, and wherein the first set of machine learning models is configured to be executed by at least a portion of the second set of machine learning models (but see Sattler Section 1 (“FEDERATED learning (FL) [1]–[5] is a distributed training framework, which allows multiple clients (typically, mobile or the IoT devices) to jointly train a single deep learning model on their combined data in a communication-efficient way, without requiring any of the participants to reveal their private training data to a centralized entity or to each other. FL realizes this goal via an iterative three-step protocol where, in every communication round t, the clients first synchronize with the server by downloading the latest master model θt. Every client then proceeds to improve the down loaded model by performing multiple iterations of stochastic gradient descent with minibatches sampled from its local data Di, resulting in a weight-update vector
PNG
media_image2.png
62
662
media_image2.png
Greyscale
Finally, all clients upload their computed weight-updates to the server, where they are aggregated by weighted averaging according to
PNG
media_image3.png
96
632
media_image3.png
Greyscale
to create the next master model (see [1]). The procedure is summarized in Algorithm 2.”)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Chen and Zhang to incorporate the teachings of Sattler to update the global model with the client models obtained by mutual learning for execution on the server, at least because doing so would enable training a single deep learning model on the combined data of the models belonging to each cluster in a communication-efficient way.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Bhuyan, Neelkamal, Sharayu Moharir, and Gauri Joshi. "Multi-model federated learning with provable guarantees." EAI International Conference on Performance Evaluation Methodologies and Tools. Cham: Springer Nature Switzerland, 2022.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SHAHID KHAN whose telephone number is (571)270-0419. The examiner can normally be reached M-F, 9-5 est.
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, Usmaan Saeed can be reached at (571)272-4046. 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.
/SHAHID K KHAN/ Primary Examiner, Art Unit 2146