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
Applicant’s argument filed 07/01/2026 have been fully considered but the arguments provided for 103 rejections are not persuasive. The amendments have overcome the 112(a) and 112(b) rejections and thus, the rejections have been withdrawn.
Applicant’s Argument: On page 10-11 of Applicant’s response to rejections under 35 U.S.C. 103, applicant states that the cited references do not teach the new claim limitations of “wherein CURL creates compact representations of the unlabeled data that reduces dimensionality of the unlabeled data” and “the aggregating comprises concatenating the vectors”.
Examiner’s Response: Applicant’s argument is not persuasive. Applicant’s arguments have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
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 1-3, 7-9, and 13-15, are rejected under 35 U.S.C. 103 as being unpatentable over Milletari (US11804050B1), in view of Suzuki (US20220405606A1) and Wu, “Decentralized Unsupervised Learning of Visual Representations”.
Regarding claim 1, Milletari teaches:
“A computer-implemented method (CIM) for use with a federated computer system that includes a server subsystem and a plurality of client subsystems, (abstract, Figure 5, Figure 5 shows a plurality of client devices as training nodes and the results are sent to the aggregation servers) with each client subsystem including stored unlabeled data, with the server subsystem being in communication with each client subsystem through a communication network, and with each client subsystem not being permitted access to stored unlabeled data of other client subsystems of the federated computer system, the CIM comprising” (col. 6, lines 22-32; col. 10, lines 45-65; col. 27, lines 30-44; Figure 5, The training of the models can be unsupervised and using unlabeled data. The client devices train the models locally and communicate the results to the aggregation servers over a network. The model trainer at each client device uses a different training dataset and the client devices do not share training data. In one embodiment, vertical federated learning can be implemented between the client devices to ensure data privacy.)
“receiving, by the server subsystem, an ” ([col. 3, lines 28-37; col. 4, lines 16-25; col. 19, lines 47-55, Figure 5], Machine learning model (MLM) 108 may be collaboratively trained by training nodes and MLM 106. Federated learning may be used to train a central model by a plurality of local models. In at least one embodiment, a client-server network environment may include client devices as edge device for training MLM 106 and aggregation servers that train MLM 108. MLM 108 is trained by a plurality of MLM 106 and it is implied that during the first round of training, the client and server side starts off with an untrained model.)
“receiving, from each given client subsystem, over the communication network and by the server subsystem, a respective collection of representations that have been obtained at the given client subsystem through models located at the given client subsystem and trained on the unlabeled data of the given client subsystem using Contrastive Unsupervised Representation Learning (CURL), ” ([col. 3, lines 28-37; col. 4, lines 26-48; col. 10, lines 45-65; col. 19, lines 47-55; Figure 1A and 5], The plurality of client devices uses unsupervised learning to train a ML model locally and provide the learned parameters of weights and biases to the training aggregator server over the network. Training aggregator receives information from training nodes and uses the information to train MLM 108. A client server network environment may be used for collaborative training. The client devices train MLM 106 and provide information to the aggregation servers to train MLM 108. In one embodiment, machine learning models are trained using unsupervised learning and unsupervised training can be used to generate a self-organizing map, which can perform operations useful in reducing dimensionality of the data. A self-organizing map is an unsupervised neural network that trains on unlabeled input data.)
“aggregating, by the server subsystem, the representations of the collections of representations respectively received from each client subsystem of the plurality of client subsystems to obtain an input to the ” ([col. 4, lines 49-57; col. 11, lines 5-27; col. 19, lines 47-55, Figure 5], The parameter determiner aggregates the parameters of local machine learning models to generate parameters for the central model. Machine learning model (MLM) 108 may be collaboratively trained by training nodes and MLM 106. Federated learning may be used to train a central model by a plurality of local models. In at least one embodiment, a client-server network environment may include client devices as edge device for training MLM 106 and aggregation servers that train MLM 108. MLM 108 is trained by a plurality of MLM 106 and it is implied that during the first round of training, the client and server side starts off with an untrained model. Consensus determiner analyzes the representation data from the plurality of training nodes and may determine a consensus value by combining values of parameters for each training nodes.)
“inputting the input to the ” ([col. 3, lines 55-65; col. 4, lines 16-25 & 65-67; col. 5, lines 1-14; col. 27, lines 29-37], The aggregated parameters are input into the central model for training over multiple iterations. A contribution determiner may determine the amount of influence (plurality of patterns) for each training node in training the central model based on weights. The claim does not define the scope of the limitation “Self-Supervised Vertical Federated Learning”. It is unclear what constitutes as Self-Supervised Vertical Federated Learning without further defining the functionality of the claim limitation. Under the broadest reasonable interpretation, Milletari discloses the claim limitation. Machine learning model may be auto-encoders, which is a type of self-supervised learning process. It is also disclosed that the training of the models can consist of vertical federated learning.)
Milletari does not explicitly disclose an implementation of “untrained server-side ML model” and “wherein CURL creates compact representations of the unlabeled data that reduces dimensionality of the unlabeled data”.
However, Suzuki discloses in the same field of endeavor:
“receiving, by the server subsystem, an untrained server-side machine learning (ML) model” ([0042-0043], The server receives a base prediction model that is an untrained neural network.)
“aggregating, by the server subsystem, the representations of the collections of representations respectively received from each client subsystem of the plurality of client subsystems to obtain an input to the untrained server-side machine learning (ML) model ...” ([0042-0043; 0048-0049], The clients send model parameters to the server and the server performs an integration process of the parameter to generate an updated model.)
“inputting the input to the untrained server-side ML model to train the untrained server-side ML model ...” ([0042-0043; 0048-0050], The model parameter from each training base is combined at the server to train the untrained server-side machine learning model.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “untrained server-side ML model” from Suzuki into the teaching of Milletari. Doing so can improve training efficiency in federated learning system by using a plurality of client training devices to update a global ML model (Suzuki, abstract).
Milletari in view of Suzuki does not explicitly disclose an implementation of “wherein CURL creates compact representations of the unlabeled data that reduces dimensionality of the unlabeled data”.
However, Wu discloses in the same field of endeavor:
“receiving, from each given client subsystem, over the communication network and by the server subsystem, a respective collection of representations that have been obtained at the given client subsystem through models located at the given client subsystem and trained on the unlabeled data of the given client subsystem using Contrastive Unsupervised Representation Learning (CURL), wherein CURL creates compact representations of the unlabeled data that reduces dimensionality of the unlabeled data” ([pg. 2-3, Section 3, par. 1-2; pg. 4, Section 4.1, par. 1; pg. 5, Section 6, par. 1], Each client conducts local contrastive learning for multiple epochs with local features. The framework consist of ResNet-18 as the base encoder to compress the input data into the feature space. During each learning round, each client uploads its latest model and latest features of encrypted images to the server.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “wherein CURL creates compact representations of the unlabeled data that reduces dimensionality of the unlabeled data” from Wu into the teaching of Milletari in view of Suzuki. Doing so can improve federated learning systems to learn visual representations from unlabeled image data by implementing a collaborative contrastive learning framework (Wu, abstract).
Regarding Claim 7:
The claim recites an article of manufacture (“A computer program product”) that performs the method as described in claim 1. Therefore claim 7 is rejected under the same reasons mentioned for claim 1. The additional elements of claim 7 are addressed below with the Milletari reference:
“one or more computer readable storage media” ([col. 113, lines 1-43], A computer-readable storage medium stores the instructions to perform the operations.)
“computer code stored collectively in the one or more computer readable storage media, with the computer code including data and instructions to cause a processor(s) set to perform at least the following operations” ([col. 113, lines 1-43], A computer-readable storage medium stores the instructions to perform the operations.)
Regarding Claim 13:
The claim recites a system (“A computer system”) that performs the method as described in claim 1. Therefore claim 13 is rejected under the same reasons mentioned for claim 1. The additional elements of claim 13 are addressed below with the Milletari reference:
“a processor(s) set” ([col. 54, lines 47-57], Figure 16A shows a computer system consisting of a processor.)
“a set of storage device(s)” ([col. 6, lines 52-65, Figure 2A], Figure 2A shows multiple data storage components of the system.)
“computer code stored collectively in the set of storage device(s), with the computer code including data and instructions to cause a processor(s) set to perform at least the following operations” ([col. 6, lines 52-65], The hardware shown in Figure 2A and 2B may store data and instructions to perform the steps of the federated learning system.)
Regarding claims 2, 8, and 14, Milletari teaches:
“receiving, by each client subsystem of the plurality of client subsystems, new data for analysis” (col. 5, lines 13-19; col. 6, lines 20-32, The training of the models is performed over multiple iterations. Training dataset may be data generated on-site at the local training node.)
“inputting the new data to a respective trained ML model of each client subsystem of the plurality of client subsystems, to obtain a collection of new representations” ([col. 5, lines 13-19], Local models at each training node process new training datasets and a new set of parameters are determined.)
“sending, by each client subsystem, through the communication network and to the server subsystem, the collection of new representations” ([col. 5, lines 28-33], The parameters of the local models are sent to the training aggregator through the network.)
“aggregating, by the server subsystem, the representations of the collections of new representations to obtain a new input to the trained server-side ML model” ([col. 5, lines 34-39], The parameter of the local models are combined by the training aggregator.)
“using the trained server-side ML model to make a first prediction based on the new input” ([col. 54, lines 14-31], The federated learning system may be implemented in cloud-based servers and autonomous vehicle. The servers receive image data from the vehicle to perform inferencing of an object classification task.)
Regarding claims 3, 9, and 15, Milletari in view of Suzuki and Wu teaches:
“the collections of representations obtained at the given client subsystem are in the form of vectors” ([Milletari, col. 11, lines 28-35; col. 13, lines 1-8], The set of parameters from the local training nodes are provided as a vector)
“the aggregating comprises concatenating the vectors” ([Milletari, col. 15, lines 39-64], The parameter determiner from the training aggregator may aggregate one or more values of one or more corresponding parameters from one or more of training nodes using one or more corresponding weights. A plurality of parameter values from one or more training nodes are aggregated prior to updating the server model. Wu (pg. 2-3, Section 3, par. 1-2) further discloses the server combines the features received from the plurality of clients.)
“the unlabeled data of the client subsystems cannot be derived from the vectors” ([Wu, pg. 2-3, Section 3, par. 1-2], During local contrastive learning, images are encrypted by InstaHide prior to generating the features to upload to the server. The input data cannot be reconstructed.)
Claims 20, 22, and 24 are rejected under 35 U.S.C. 103 as being unpatentable over Milletari (US11804050B1), in view of Suzuki (US20220405606A1), Wu, “Decentralized Unsupervised Learning of Visual Representations”, and Han, "Adaptive Gradient Sparsification for Efficient Federated Learning: An Online Learning Approach".
Regarding claims 20, 22, and 24, Milletari in view of Suzuki and Wu teaches:
“inputting labeled data to the models on each client subsystem” ([Milletari, col. 10, lines 63-67; col. 11, lines 1-4, Semi-supervised learning may be used and it consists of a training dataset with both labeled and unlabeled data. The training nodes can receive labeled data for semi-supervised learning.)
“aggregating outputs of the models at a server that is classified as a label holder” ([Milletari, col. 4, lines 49-57; col. 11, lines 5-27; col. 19, lines 47-55, Figure 5], The parameter determiner aggregates the parameters of local machine learning models to generate parameters for the central model. Machine learning model (MLM) 108 may be collaboratively trained by training nodes and MLM 106. Federated learning may be used to train a central model by a plurality of local models.)
Milletari in view of Suzuki and Wu does not explicitly disclose an implementation of “computing, by the server, a loss using the outputs of the models” and “sending, by the server and to each of the client subsystems, gradient information, wherein the models on each client subsystem are updated according to the gradient information”. However, Han discloses in the same field of endeavor:
“computing, by the server, a loss using the outputs of the models” ([pg. 3, section A. par. 1-2; pg. 7-8, section E, par. 1-2], Each client computes losses on a data sample and the losses are sent to the server. The server also computes averages of the losses.)
“sending, by the server and to each of the client subsystems, gradient information, wherein the models on each client subsystem are updated according to the gradient information” ([pg. 3, section A. par. 1-5; pg. 7-8, section E, par. 4; Figure 3], The framework consists of aggregating the sparsified gradients after every local update step and model weight vector is updated in every training round. Figure 3 shows the communication between the server and clients. The server computes and sends the gradient information to all client devices.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “computing, by the server, a loss using the outputs of the models” and “sending, by the server and to each of the client subsystems, gradient information, wherein the models on each client subsystem are updated according to the gradient information” from Han into the teaching of Milletari in view of Suzuki and Wu. Doing so can reduce the communication overhead and improve the overall efficiency of federated learning by implementing gradient sparsification (Han, abstract).
Conclusion
THIS ACTION IS MADE FINAL. 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 GARY MAC whose telephone number is (703)756-1517. The examiner can normally be reached Monday - Friday 8:00 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, Abdullah Kawsar can be reached on (571) 270-3169. 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.
/GARY MAC/Examiner, Art Unit 2127
/ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127
Prosecution Insights