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
2. This office action is in response to the original filing of 04/20/2023. Claim 1-15 are pending and have been considered below.
Claim Rejections - 35 USC § 112
3. The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 6 is rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 6 recites the limitation "processor subsystem" in line 5. There is insufficient antecedent basis for this limitation in the claim.
Claim Rejections - 35 USC § 101
4. 35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claim 15 is rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. In summary, claim 15 recites a “computer readable medium” comprising transitory or non-transitory data. Transitory signals are not patentable under 35 USC 101. The broadest, reasonable interpretation of “computer readable medium” encompasses non-statutory subject matter (transitory signals) that is unpatentable under 35 U.S.C. 101.
Appropriate correction is required.
Claim Rejections - 35 USC § 103
5. 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 1-15 are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al Convergence of Edge Computing and Deep Learning: A Comprehensive Survey (January 30, 2020) Hereinafter Wang in view of Rawat et al. (US 2021/0326757 A1).
Claim 1. Wang discloses an edge device for use in a federated learning system for training a federated model, the edge device being comprised in a set of multiple edge devices, wherein the federated model is trained on respective local training datasets of the respective multiple edge devices (the edge devices receive a global/federated model which is trained on local training data, fig. 18), the edge device comprising:
a storage interface for accessing the local training dataset of the edge device, the local training dataset comprising multiple training inputs and corresponding training outputs (fig. 18);
a processor system configured to iteratively train the federated model by, in an iteration: obtaining a current federated model, determining a model update for the current federated model based on the local training dataset, and sending the model update to one or more other devices of the federated learning system (§VII "deep learning training at edge", section B "Vanilla federated learning at edge" : "Without requiring uploading data for central cloud training, FL can allow edge devices to train their local DL models with their own collected data and upload only the updated model instead. ", i.e. the edge devices must necessarily have memory storage to store their local data and processing means to perform the training. The iterations are shown in fig.18, "one round", meaning "each round of the iteration"),
wherein determining the model update in the iteration comprises: applying the current federated model to a training input to obtain at least a model output for the training input (..FL iteratively solicits a random set of edge devices to 1) download the global DL model from an aggregation server (use “server” in following), 2) train their local models on the downloaded global model with their own data, and 3) upload only the updated model to the server for model averaging.) (p. 888 col. 1, lines 7-15) [the model is applied as part of the training to the training dataset];
Wang does not explicitly disclose if the model output does not match a training output corresponding to the training input, include the training input in a subset of filtered training inputs to be used in the iteration; and - determining the model update by training the current federated model on only the subset of filtered training inputs.
However, Rawat discloses if the model output does not match a training output corresponding to the training input (i.e. positive examples), include the training input in a subset of filtered training inputs to be used in the iteration; and - determining the model update by training the current federated model on only the subset of filtered training inputs ([0058], [0060]-[0062]). Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Wang with Rawat features. One would have been motivated to do so in order to enable federated learning with only positive labels.
Claim 2. Wang and Rawat disclose the edge device of claim 1, Wang further discloses wherein the processor system is configured to perform one or more iterations in which a model update is determined by training on the full local training dataset, followed by one or more iterations in which a model update is determined by training on only a subset of filtered training inputs (§VII "deep learning training at edge", section B "Vanilla federated learning at edge" : "Without requiring uploading data for central cloud training, FL can allow edge devices to train their local DL models with their own collected data and upload only the updated model instead. ", i.e. the edge devices must necessarily have memory storage to store their local data and processing means to perform the training. The iterations are shown in fig.18, "one round", meaning "each round of the iteration")…Wang further discloses an alternative way of reducing the training dataset (§IX "Lesson learned and open challenges", section C "Complete edge architecture for DL" : " For instance, in order to reduce the amount of training data and speeding up the training process, using unlabeled data to transfer knowledge between edge devices can be adopted").
Claim 3. Wang and Rawat disclose the edge device of claim 1, Wang further discloses wherein the edge device is an IoT device (.. as depicted in Fig. 1, we further divide common edge devices into end devices and edge nodes: the “end devices” (end level) is used to refer to mobile edge devices (including smartphones, smart vehicles, etc.) and various IoT devices, and the “edge nodes” (edge level) include Cloudlets, Road-Side Units (RSUs), Fog nodes, edge servers, MEC servers and so on, namely servers deployed at the edge of the network) (§II.A.4).
Claim 4. Wang and Rawat disclose the edge device of claim 1, Wang further discloses wherein the processor system is configured to train the current federated model by applying multiple training epochs to the subset of filtered training inputs (§VII "deep learning training at edge", section B "Vanilla federated learning at edge" : "Without requiring uploading data for central cloud training, FL can allow edge devices to train their local DL models with their own collected data and upload only the updated model instead. ", i.e. the edge devices must necessarily have memory storage to store their local data and processing means to perform the training. The iterations are shown in fig.18, "one round", meaning "each round of the iteration").. Wang further discloses an alternative way of reducing the training dataset (§IX "Lesson learned and open challenges", section C "Complete edge architecture for DL" : " For instance, in order to reduce the amount of training data and speeding up the training process, using unlabeled data to transfer knowledge between edge devices can be adopted")..
Claim 5. Wang and Rawat disclose the edge device of claim 1, Rawat further discloses wherein the processor system is further configured to determine a confidence score of the current federated model for the training input; and, if the confidence score does not exceed a threshold, include the training input in the subset of filtered training inputs ([0058], [0060]-[0062]) [such condition is similar to the one of claim 1]. One would have been motivated to do so in order to enable federated learning with only positive labels.
Claim 6. Wang and Rawat disclose the edge device of claim 1, Wang further discloses wherein the processor system is configured to train the current federated model in a forward pass in which the current federated model is applied to training inputs to determine model outputs, and wherein the processor subsystem is configured to determine the subset of filtered training inputs to be used in the iteration based on the model outputs determined in the forward pass ((§III "Fundamental of deep Learning", section A "Neural Networks in DL") : "..(§VII "deep learning training at edge", section B "Vanilla federated learning at edge" : "Without requiring uploading data for central cloud training, FL can allow edge devices to train their local DL models with their own collected data and upload only the updated model instead. ", i.e. the edge devices must necessarily have memory storage to store their local data and processing means to perform the training. The iterations are shown in fig.18, "one round", meaning "each round of the iteration")…Wang further discloses an alternative way of reducing the training dataset (§IX "Lesson learned and open challenges", section C "Complete edge architecture for DL" : " For instance, in order to reduce the amount of training data and speeding up the training process, using unlabeled data to transfer knowledge between edge devices can be adopted"). and Rawat further discloses a backward pass in which the model update is determined based on the determined model outputs ([0076]). One would have been motivated to do so in order to enable federated learning with only positive labels.
Claim 7. Wang and Rawat disclose the edge device of claim 1, Wang further discloses wherein the subset of filtered training inputs to be used in the iteration is smaller than a subset of filtered training inputs to be used in a previous iteration ((§IV "Fundamental of deep Learning", section A "Neural Networks in DL") : "..(§VII "Edge Computing for Deep Learning", section B "Communication and Computation Modes for Edge DL" 3) Vertical Collaboration: the intermediate output, e.g., high-resolution surveillance video streams, should be carefully designed much smaller than the raw input, therefore drastically reducing the network traffic required between the end and the edge (or the edge and the cloud)).
Claim 8. Wang and Rawat disclose the edge device of claim 1, Rawat further discloses wherein the processor system is configured to include a training input in the subset of filtered training inputs that was not included in a subset of filtered training inputs of a previous iteration ([0058], [0060]-[0062]) [such condition is similar to the one of claim 1]. One would have been motivated to do so in order to enable federated learning with only positive labels.
Claim 9. Wang and Rawat disclose the edge device of claim 1, Wang further discloses wherein the processor system is configured to receive the current federated model from an aggregation device of the federated learning system and to send the model update to the aggregation device (§VII "deep learning training at edge", section B "Vanilla federated learning at edge" : "Without requiring uploading data for central cloud training, FL can allow edge devices to train their local DL models with their own collected data and upload only the updated model instead. ", i.e. the edge devices must necessarily have memory storage to store their local data and processing means to perform the training. The iterations are shown in fig.18, "one round", meaning "each round of the iteration").. Wang further discloses an alternative way of reducing the training dataset (§IX "Lesson learned and open challenges", section C "Complete edge architecture for DL" : " For instance, in order to reduce the amount of training data and speeding up the training process, using unlabeled data to transfer knowledge between edge devices can be adopted")..
Claim 10. Wang and Rawat disclose the edge device of claim 1, Wang further discloses wherein the processor system is configured to determine an initial federated model by training on the local training dataset, and to send the initial federated model to one or more other devices of the federated learning system ((§VII "deep learning training at edge", section B "Vanilla federated learning at edge" FL [198], [199] can allow edge devices to train their local DL models with their own collected data and upload only the updated model instead. As depicted in Fig. 18, FL iteratively solicits a random set of edge devices to 1) download the global DL model from an aggregation server (use “server” in following), 2) train their local models on the downloaded global model with their own data, and 3) upload only the updated model to the server for model averaging).
Claim 11. Wang and Rawat disclose the edge device of claim 1, Wang further discloses wherein the model update comprises an updated set of parameters for the current federated model or a gradient for updating the current federated model (§VII "deep learning training at edge", section B "Vanilla federated learning at edge" : "Without requiring uploading data for central cloud training, FL can allow edge devices to train their local DL models with their own collected data and upload only the updated model instead. ", i.e. the edge devices must necessarily have memory storage to store their local data and processing means to perform the training. The iterations are shown in fig.18, "one round", meaning "each round of the iteration"…we can let FL clients communicate with the central server periodically (rather continually) to seek consensus on the shared DL model [202]. In addition, structured update, sketched update can help enhance the communication efficiency when clients uploading updates to the server as well. Structured update means restricting the model updates to have a pre-specified structure, specifically, 1) low-rank matrix; or 2) sparse matrix [202], [203]. On the other hand, for sketched update, full model updates are maintained.)
Claim 12. Supra claim 1 and Wang further discloses wherein the aggregation device is configured to iteratively train the federated model by, in an iteration: sending a current federated model to one or more edge devices; receiving model updates from the one or more edge devices, and updating the current federated model by aggregating the model updates ((§VII "deep learning training at edge", section B "Vanilla federated learning at edge": FL [198], [199] can allow edge devices to train their local DL models with their own collected data and upload
only the updated model instead. As depicted in Fig. 18, FL iteratively solicits a random set of edge devices to 1) download the global DL model from an aggregation server (use “server” in following), 2) train their local models on the downloaded global model with their own data, and 3) upload only the updated model to the server for model averaging. Privacy and security risks can be significantly reduced by restricting the training data to only the device side, and thus avoiding the privacy issues as in [195], incurred by uploading training data to the cloud.).
Claim 13 represents the computer-implemented edge method of claim 1 and is rejected under the same rationale.
Claim 14 represents the computer-implemented federated learning method of claim 12 and is rejected along the same rationale.
Claim 15. represents the computer-readable medium of claim 13 and is rejected along the same rationale.
Conclusion
6. The prior art made of record and not relied upon is considered pertinent to applicant's disclosure (See PTO-892).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Phenuel S. Salomon whose telephone number is (571) 270-1699. The examiner can normally be reached on Mon-Fri 7:00 A.M. to 4:00 P.M. (Alternate Friday Off) EST.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Usmaan Saeed can be reached on (571) 272-4046. The fax phone number for the organization where this application or proceeding is assigned is 571-273-3800.
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://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/PHENUEL S SALOMON/Primary Examiner, Art Unit 2146