DETAILED ACTION
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 .
Drawings
The drawings filed on 09/13/2023 are accepted.
Specification
The specification filed on 09/13/2023 is accepted.
Information Disclosure Statement
The examiner has considered the information disclosure statements (IDS) submitted on 09/13/2023.
Claim Rejections - 35 USC § 112
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.
Claims 1-2, 4 and 8 are 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 1 is rejected under 35 U.S.C. 112(b) for reciting the limitation (among others) “after collecting enough parameters of the global shared model”. The term "enough" in claim 1 is a relative term which renders the claim indefinite. The term "enough" is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention.
For examination purposes, the above limitation of claim 1 is interpreted as “after collecting parameters of the global shared model”.
Claim 1 is also rejected under 35 U.S.C. 112(b) for reciting the limitation (among others) “the participants first dividing a small portion (such as 5% of the training data) from the obtained private training data as a validation set”. The term "small" in claim 1 is a relative term which renders the claim indefinite. The term "small" is not positively defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. The phrase "such as …" in the parenthesis constitutes intended use making the functionality following not carry any patentable weight since it never actually has to take place. Claims should be amended to recite more direct and positive language.
For the purpose of examination, the cited “such as …” is not considered limiting. See MPEP § 2173.05(d), and the above limitation of claim 1 is interpreted as “the participants first dividing the obtained private training data into a plurality of portions, and using one of the portions as a validation set”.
Claim 8 is rejected under 35 U.S.C. 112(b) for reciting the limitation (among others) “after collecting sufficient global shared models from the clients”. The term "sufficient" in claim 8 is a relative term which renders the claim indefinite. The term "sufficient" is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention.
For examination purposes, the above limitation of claim 1 is interpreted as “after collecting global shared models from the clients”.
Claims 2, 4 and 8 are rejected for being dependent on a rejected base claim, namely claim 1.
Allowable Subject Matter
Claims 1-2, 4 and 8 would be allowable if the 112 (b) rejections above were overcome.
The following is a statement of reasons for the indication of allowable subject matter:
Claim 1 is allowable for disclosing
An adaptive personalized federated learning method supporting heterogeneous model, comprising the following steps:
(1) initializing parameters of a global shared model by a central server;
(2) the central server sending the parameters of the global shared model to each participant of the federated learning, after receiving the parameters of the global shared model, the participants updating their own global shared model with the parameters;
(3) the participants performing learning for adaptability to update the weights of private models;
(4) the participants using newly obtained private training data to train both the private models and the globally shared model by using a stochastic gradient descent algorithm;
(5) the participants uploading the parameters of the globally shared model to the central server after one round of iterative training;
(6) after collecting enough parameters of the global shared model, the central server aggregating these model parameters to obtain new parameters of the global shared model, and then returning to step (2) to distribute the new parameters of the global shared model to each participant, and then circulating until the loss function of all models converges or reaches the maximum number of iterations in federated learning;
wherein, the private models are the models held by each participant of federated learning, and the structure and parameters are not disclosed, the structure of the private models held by each participant are different;
wherein, the specific implementation of step (3) is as follows: the participants first dividing a small portion (such as 5% of the training data) from the obtained private training data as a validation set, and inferring the private models and the global shared model on the validation set, obtaining the prediction output result ppri of the private models and the prediction output result Psha of the global shared models, then the participants updating the weight of the private models through the stochastic gradient descent method, and the update expression is as follows:
λ’i = λi – η
∇
λiLCE (Paen, y)
wherein, λi is the weight of the private model before update, λ’i is the weight of the private model after update, η represents learning rate,
∇
λi represents LCE(Paen, y) to λi gradient, LCE(Paen, y) represents cross entropy of Paen and y, Paen represents the weighted average result of ppri and Psha, y is the ground-truth label;
wherein, the loss function expression used for private model training in step (4) is as follows:
Lpri = LCE(Ppri,y) + DKL(Ppri||Psha) + LCE(Paen,y)
wherein, Lpri is the loss function of the private model, LCE(Ppri,y) represents cross entropy of Ppri and y, LCE(Paen,y) represents cross entropy of Paen and y, DKL(Ppri||Psha) represents the KL divergence of Ppri relative to Psha, Paen represents the weighted average result of Ppri and Psha, y is the ground-truth label, Psha is the prediction output result of the global shared model;
wherein, the expression of the loss function used for the global shared model training in step (4) is as follows:
Lsha = LCE(Psha,y) + DKL(Psha||Ppri) + LCE(Paen,y)
wherein, Lsha is the loss function of the global shared model, LCE(Psha,y) represents cross entropy of Psha and y, LCE(Paen,y) represents cross entropy of Paen and y, DKL(Psha||Ppri) represents the KL divergence of Psha relative to Ppri, Paen represents the weighted average result of Ppri and Psha, y is the ground-truth label, Ppri is the prediction output result of the private model, Psha is the prediction output result of the global shared model.
The closest references found
Liang et al. (Think Locally, Act Globally: Federated Learning with Local and Global Representations) on pages 4-5, algorithm 1, sections 3.2-3.3, discloses a process of learning a global model across all devices by: at each iteration, a server sends a copy of a global model to each device, each device train their local model and the received global model to obtain prediction. The local model parameters and global model parameters are updated by gradient-based method. After the updates, each device sends the updated global parameters back to the server which aggregates these updates using a weighted average.
Malik et al. (US Pub. 2021/0117780) in Fig. 8, paragraph 0117, discloses a process of training a neural network model using personalized federated learning. At each round of training, a server selects a subset of client devices to train a global neural network model together with the local personalization models of each selected client device. Each client receives a current version of the global neural network model from the server, trains both the received global neural network model together with the local personalization model using local data to generate a plurality of updated federated model parameters and a plurality of updated local model parameters. Each client then sends the trained global neural network model including the updated federated model parameters to server, the server aggregates the updated federated model parameters sent from the client devices to generate an updated global neural network model. The server may then proceed with the next round of training by sending the updated global neural network model to a new subset of client devices.
Shan et al. (US Pub. 2023/0196092) describes a method for training a federated learning model. In paragraphs 0023-0024, Shan discloses the global federated learning model is updated using the equation:
PNG
media_image1.png
52
338
media_image1.png
Greyscale
“where Wt is updated weight of the global federated learning model after incorporating parameters from the received training event Ek, Wt_1 is weight of the global federated learning model before incorporating the parameters of the received training event Ek, n is summated training dataset size from all the K clients, αt is staleness discount rate of the t-th iteration of training of the k-th client federated learning model, and pt,k is accuracy of the t-th iteration of training of the k-th client federated learning model”
However, the prior art of record do not teach or suggest, individually or in combination the claim limitations as a whole especially limitations:
“wherein, the specific implementation of step (3) is as follows: the participants first dividing a small portion (such as 5% of the training data) from the obtained private training data as a validation set, and inferring the private models and the global shared model on the validation set, obtaining the prediction output result ppri of the private models and the prediction output result Psha of the global shared models, then the participants updating the weight of the private models through the stochastic gradient descent method, and the update expression is as follows:
λ’i = λi – η
∇
λiLCE (Paen, y)
wherein, λi is the weight of the private model before update, λ’i is the weight of the private model after update, η represents learning rate,
∇
λi represents LCE(Paen, y) to λi gradient, LCE(Paen, y) represents cross entropy of Paen and y, Paen represents the weighted average result of ppri and Psha, y is the ground-truth label;
wherein, the loss function expression used for private model training in step (4) is as follows:
Lpri = LCE(Ppri,y) + DKL(Ppri||Psha) + LCE(Paen,y)
wherein, Lpri is the loss function of the private model, LCE(Ppri,y) represents cross entropy of Ppri and y, LCE(Paen,y) represents cross entropy of Paen and y, DKL(Ppri||Psha) represents the KL divergence of Ppri relative to Psha, Paen represents the weighted average result of Ppri and Psha, y is the ground-truth label, Psha is the prediction output result of the global shared model;
wherein, the expression of the loss function used for the global shared model training in step (4) is as follows:
Lsha = LCE(Psha,y) + DKL(Psha||Ppri) + LCE(Paen,y)
wherein, Lsha is the loss function of the global shared model, LCE(Psha,y) represents cross entropy of Psha and y, LCE(Paen,y) represents cross entropy of Paen and y, DKL(Psha||Ppri) represents the KL divergence of Psha relative to Ppri, Paen represents the weighted average result of Ppri and Psha, y is the ground-truth label, Ppri is the prediction output result of the private model, Psha is the prediction output result of the global shared model.”
Therefore, the combination of features is considered to be allowable.
Claims 2, 4 and 8 are considered to be allowable because they are dependent on claim 1.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TRI T NGUYEN whose telephone number is 571-272-0103. The examiner can normally be reached M-F, 8 AM-5 PM, (CT).
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, OMAR FERNANDEZ can be reached at 571-272-2589. 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.
/TRI T NGUYEN/Examiner, Art Unit 2128
/OMAR F FERNANDEZ RIVAS/Supervisory Patent Examiner, Art Unit 2128
Prosecution Insights