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 .
DETAILED ACTION
This action is responsive to the following communication: Amendment filed Aug. 20, 2026. This Action is made Final.
Claims 1-18 and 21-22 are pending in the case. Claims 1, 9 and 17 are independent claims.
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 1-7, 9-15, 17-18 are rejected under 35 U.S.C. 103 as being unpatentable over Harshit et al. (hereinafter Harshit) “Canoe: A System for Collaborative Learning for Neural Nets“ 2021 in view of U.S. Patent Publication No. 2008/0295012 in view of Fernando et al. (hereinafter Fernando) “Collaborative and continual learning for classification tasks in a society of device” 2020 and further in view of Nakayama et al. (hereinafter Nakayama) U.S. Patent Pub. 2021/0406782.
With respect to independent claim 1, Harshit teaches a method for continual neural network training in an edge computing environment (see e.g., Fig. 1 Abstract – “. Canoe provides new system support for dynamically extracting significant parameters from a helper node’s neural network, and uses this with a multi-model boosting-based approach to improve the predictive performance of the target node.”), comprising:
deploying a plurality of copies of a centralized neural network respectively to a corresponding plurality of edge servers (see e.g., Fig. 1 - Step 1 The initial model is deployed on the edge node),
wherein each of the copies of the centralized neural network is independently operated and trained at a respective one of the edge servers based on inputs received at that edge server to create independently trained neural networks (see e.g., Page 3 col. 2 – “each edge node creates a customized model and requests for help when there is drift in the model performance … helper nodes can be identified from the model and node metadata, using techniques such as [14, 43, 53, 71], as done in [23]. The specific contributions made by Canoe are new techniques which enable knowledge transfer to be performed across neural network models among distributed nodes in a manner that boosts the over-all accuracy at the target node while maintaining low data transfer costs”);
sending, at periodic intervals, a corresponding fitness measure the independently trained neural networks from the plurality of edge servers to a cloud-based data center (see e.g., Page 2, column 2 lines 18-23); and
performing neural network breeding based on the copies of the independently trained neural networks sent from the plurality of edge servers (see e.g., page 7 lines 27-29).
Harshit does not expressly show copies of the independently trained neural networks are sent to a cloud-based data center and the centralized neural network at the cloud based data center is updated based on the copies of the independently trained neural network sent from the edge severs and performing neural network breeding is based on the copies of the independently trained neural networks from the edge servers. However, Harshit teaches performing centralized neural network breeding in a cloud-based data center. Harshit teaches the use of a federated learning to obtain centralized neural network (see Page 3 column 1 lines 3-7). Furthermore, Fernando teaches similar feature (page 7, lines 25-27 – “each of the devices creates and refines its own local model of the learning problem that is intended to be solved. For that, devices are continuously acquiring and storing new information through their sensors. This information is raw data, which must be locally preprocessed before being able to use it in a learning stage: noise detection, data transformation, feature extraction, data normalization, instance selection, etc. When local models are obtained, they are sent to the cloud where a new learning stage is performed to join the local knowledge, thus obtaining a global model.” edge sever sends copies of the independent trained neural networks and update the centralized neural network at the data center). Both Harshit and Fernando are directed to federated learning. Accordingly, it would have been obvious to the skilled artisan before the effective filing date of the claimed invention having Harshit and Fernando in front of them to modify the system of Harshit to include the above feature. The motivation to combine Harshit and Fernando comes from Fernando. Fernando discloses the motivation for “Glocal learning” which is incremental and adaptive learning over time (see e.g. page 7). This motivation for combination also applies to the remaining claims which depend on this combination.
Harshit-Fernando does not expressly show the amended features discussed below. However, Nakayama teaches a corresponding fitness measure, determined at the respective one of the edge servers (see e.g., Para [110] – “Performance of the local ML models, cluster models, and global models is uploaded”) and each of the copies of the independently trained neural networks being sent together with the corresponding fitness measure for that copy (see e.g., Para [110] – “Performance of the local ML models, cluster models, and global models is uploaded to the database system together with the models themselves.”). Both Harshit and Nakayama are directed to federated learning. Accordingly, it would have been obvious to the skilled artisan before the effective filing date of the claimed invention having Harshit and Nakayama in front of them to further modify the modified system of Harshit to include the above feature. The motivation to combine Harshit and Nakayama comes from Nakayama. Nakayama discloses the motivation to upload all models with corresponding performance data so that user can choose (see e.g. Para [110]). This motivation for combination also applies to the remaining claims which depend on this combination.
With respect to dependent 2, the modified Harshit teaches deploying a plurality of copies of the updated centralized neural network respectively to the plurality of edge servers (see e.g. Fernando page 7).
With respect to dependent 3, the modified Harshit teaches updating the centralized neural network is performed without sending any of the inputs received at the edge servers to the cloud-based data center (see e.g. Fernando page 7 line 25-26).
With respect to dependent 4, the modified Harshit teaches the periodic interval is defined by a preset period of time (The examiner notes that this is well-known in the art).
With respect to dependent 5, the modified Harshit teaches the periodic interval is defined by a preset amount of drift occurring in one or more of the copies of the independently trained neural networks (see e.g. page 7 col. 2 line 13-16).
With respect to dependent 6, the modified Harshit teaches the neural network breeding includes discarding any of the copies of the independently trained neural networks sent from the plurality of edge servers that have a corresponding fitness measure below a threshold (see e.g. Fernando page 14 line 35-41).
With respect to dependent 7, the modified Harshit teaches the neural network breeding includes selecting top performing ones of the copies of the independently trained neural networks sent from the plurality of edge servers based on the fitness measures (see e.g. Fernando page 14 line 35-41).
Claim 9 is rejected for the similar reasons discussed above with respect to claim 1.
Claim 10 is rejected for the similar reasons discussed above with respect to claim 2.
Claim 11 is rejected for the similar reasons discussed above with respect to claim 3.
Claim 12 is rejected for the similar reasons discussed above with respect to claim 4.
Claim 13 is rejected for the similar reasons discussed above with respect to claim 5.
Claim 14 is rejected for the similar reasons discussed above with respect to claim 6.
Claim 15 is rejected for the similar reasons discussed above with respect to claim 7.
Claim 17 is rejected for the similar reasons discussed above with respect to claim 1.
Claim 18 is rejected for the similar reasons discussed above with respect to claim 2.
Claims 8, 16 are rejected under 35 U.S.C. 103 as being unpatentable over Harshit in view of Fernando and further in view of Luders et al. “Continual and One-Shot Learning Through Neural Networks with Dynamic External Memory” (hereinafter Luders) 2017.
With respect to dependent 8, the Harshit does not expressly show the neural network breeding includes performing a hyperNEAT calculation that includes determining a weighted average, based on the fitness measures, of parameters extracted from neurons in one or more of the copies of the independently trained neural networks sent from the plurality of edge servers. However, Luders teaches similar feature (page 888, lines 15- page 891, line 25). Both Harshit and Luders are directed to continual learning. Accordingly, it would have been obvious to the skilled artisan before the effective filing date of the claimed invention having Harshit and Luders in front of them to further modify the modified system of Harshit to include the above feature. The motivation to combine Harshit and Luders comes from Luders. Luders discloses the motivation for performing a hyperNEAT calculation so that learning performance can be improved (see e.g. page 888-891).
Claim 21 is rejected under 35 U.S.C. 103 as being unpatentable over Harshit in view of Fernando, Nakayama and further in view of Song et al. (hereinafter Song) U.S. Patent Pub. 2022/0044117.
With respect to claim 21, Harshit does not expressly show the features discussed below. However, Song teaches deploying a plurality of copies of the updated centralized neural network respectively to the plurality of edge servers (see e.g. Para [25][27] - “The server 106 sends the global model to different edge devices 104 to update their local models.”); and repeating the sending and the updating at a next one of the periodic intervals using the copies of the updated centralized neural network deployed to the plurality of edge servers (see e.g. Claim 8 Para [25][27]-“ This process may repeat indefinitely, as new information is collected by the edge devices 104, or may be repeated until model convergence is reached. New models may be distributed by the server 106 periodically, or after a sufficient amount of change from a previously distributed model.”).
Both Harshit and Song are directed to federated learning. Accordingly, it would have been obvious to the skilled artisan before the effective filing date of the claimed invention having Harshit and Song in front of them to further modify the modified system of Harshit to include the above feature. The motivation to combine Harshit and Song comes from Song. Song discloses the motivation to redeploying updated global model to edge devices so that the edge devices can continue to improve learning (see e.g. Para [25]-[27]).
Claim 22 is rejected under 35 U.S.C. 103 as being unpatentable over Harshit in view of Fernando, Nakayama and further in view of Wang et al. (hereinafter Wang) “Federated Learning With Matched Averaging” Feb. 2020.
With respect to claim 22, Harshit does not expressly show the features discussed below. However, Wang teaches the neural network breeding includes: deconstructing the copies of the independently trained neural networks by layer (see e.g. Abstract and Introduction – “the Federated matched averaging (FedMA) algorithm designed for federated learning of mod ern neural network architectures e.g. convolutional neural networks (CNNs) and LSTMs. FedMA constructs the shared global model in a layer-wise manner by matching and averaging hidden elements”); selecting a layer of the copies of the independently trained neural networks (see e.g. Sect. 2.3 – “data center gathers only the weights of the first layers from the clients and performs one-layer matching described previously to obtain the first layer weights of the federated model.” ); extracting characteristics of a neuron in the selected layer from each of the copies of the independently trained neural networks (see e.g. Sect. 2.1 2.3 – “Let wjl belth neuron learned on dataset j (i.e. lth column of W(1)Πj in the previous example)” The neuron’s weight corresponds to the extracted characteristics); and determining, for the neuron in the updated centralized neural network, a weighted average of the extracted characteristics based on the fitness measures (see e.g. Sect. 2.3).
Both Harshit and Wang are directed to federated learning. Accordingly, it would have been obvious to the skilled artisan before the effective filing date of the claimed invention having Harshit and Wang in front of them to further modify the modified system of Harshit to include the above feature. The motivation to combine Harshit and Wang comes from Wang. Wang discloses the motivation to match neurons before averaging them so that the neural network performance can be improved (see e.g. Abstract and Introduction).
It is noted that any citation to specific pages, columns, lines, or figures in the prior art references and any interpretation of the references should not be considered to be limiting in any way. “The use of patents as references is not limited to what the patentees describe as their own inventions or to the problems with which they are concerned. They are part of the literature of the art, relevant for all they contain.” In re Heck, 699 F.2d 1331, 1332-33, 216 USPQ 1038, 1039 (Fed. Cir. 1983) (quoting In re Lemelson, 397 F.2d 1006, 1009, 158 USPQ 275, 277 (CCPA 1968)). Further, a reference may be relied upon for all that it would have reasonably suggested to one having ordinary skill the art, including nonpreferred embodiments. Merck & Co. v. Biocraft Laboratories, 874 F.2d 804, 10 USPQ2d 1843 (Fed. Cir.), cert. denied, 493 U.S. 975 (1989). See also Upsher-Smith Labs. v. Pamlab, LLC, 412 F.3d 1319, 1323, 75 USPQ2d 1213, 1215 (Fed. Cir. 2005); Celeritas Technologies Ltd. v. Rockwell International Corp., 150 F.3d 1354, 1361, 47 USPQ2d 1516, 1522-23 (Fed. Cir. 1998).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PEIYONG WENG whose telephone number is (571)270-1660. The examiner can normally be reached on Mon.-Fri. 8 am to 5 pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Matthew Ell, can be reached on (571) 270-3264. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://portal.uspto.gov/external/portal. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
/PEI YONG WENG/Primary Examiner, Art Unit 2141
Prosecution Insights