DETAILED ACTION
This action is in response to the filing on 03/13/2023. Claims 1-20, are pending and have been considered below.
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) filed 09/11/2023 has been fully considered by the examiner.
The Information Disclosure Statements (IDS) filed 03/13/203 and 11/04/2024 have been partially considered, the follow documents have been strikethrough and have not been considered by the examiner:
NPL cite no. 3, on IDS filed 03/13/2023, is incorrectly dated as 2018, but the document shows the date as 2019
NPL cite no. 11, on IDS filed 03/13/2023, is incorrectly dated as 2020, but the document shows the date as 2021
NPL cite no. 14, on IDS filed 03/13/2023, is incorrectly dated as 2020, but the document shows the date as 2021
NPL cite no. 16, on IDS filed 03/13/2023, is incorrectly dated as 2016, but the document shows the date as 2015
NPL cite no. 16, on IDS filed 11/04/2024, is incorrectly titled as “FedBE: Making Bayesian model ensemble applicable to federated learning”, however the document title shown is “FedDistill: Making Bayesian Model Ensemble Applicable to Federated Learning”.
Specification
The disclosure is objected to because of the following informalities:
Pg. 9, lines 8-9, recites "(erg eNodeB or gNodeB)", should recite -- (e.g., eNodeB or gNodeB) --.
Para. 53, lines 2, recites "the global model 11", should recite -- the global model 116 --.
Appropriate correction is required.
Claim Objections
Claims 8 and 16 objected to because of the following informalities:
Claim 8, line 3, recites "prior to initialization the local model", should recite -- prior to initialization of the local model --.
Claim 16, line 2, recites "prior to initialization the local model", should recite -- prior to initialization of the local model --.
Appropriate correction is required.
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.
Claims 1-4, 6, 8-12, 14, and 16-18 are rejected under 35 U.S.C. 103 as being unpatentable over NITTA et al. (US 2024/0005172 A1), hereinafter Nitta, in view of Wang et al. (US 2020/0125925 A1), hereinafter Wang.
Regarding claim 1, Nitta teaches a computing system comprising: (According to one embodiment, a learning system includes a plurality of local devices and a server. [see Nitta, Abstract]) a processing unit configured to execute instructions to cause the computing system to (Nitta discloses that the learning system device comprises a CPU to execute the instructions [see para. 72-73]):
receive, from a central server, a set of global parameters (Nitta discloses receiving a global model from the server, including a set of parameters for a neural network [see Nitta, para. 29]);
initialize a local model using the set of global parameters (Nitta discloses initializing the global model with random values and transmitting the initialized global model to the local devices [see Nitta, para. 29]), the local model including at least: a feature extraction subnetwork to extract input data, a normalization layer, and a final layer to generate a prediction output (Nitta discloses the local and global model may be neural networks used in machine learning [see Nitta, para. 27], may include a normalization layer [see Nitta, para. 37], and may be used for prediction for any data [see Nitta, para. 16]);
update the local model using data sampled from a local dataset (Nitta discloses uses random sampling on local data to train the local model [see Nitta, para. 17-18]);
transmit information about a state of the updated local model to the central server (Nitta discloses the local device transmitting to the server local parameter and local data information for which training is completed [see Nitta, para. 41]).
However, Nitta fails to teach a feature extraction subnetwork to extract a feature vector from input data, a normalization layer to normalize the feature vector, and a final layer to generate a prediction output from the normalized feature vector.
In the same field of endeavor, Wang teaches:
a feature extraction subnetwork to extract a feature vector from input data, a normalization layer to normalize the feature vector, and a final layer to generate a prediction output from the normalized feature vector (Wang discloses using a feature extraction subnetwork comprising the foreground attentive subnetwork, body part subnetwork, and feature fusion subnetwork to extract a feature vector, which is then normalized by the L2 normalization layer, and using the L2 normalized feature vector to learn the symmetric triplet loss for person re-identification and multi-target tracking [see Wang, Abstract, para. 26, and FIG. 1-2]).
It would have been obvious to one of ordinary skill, in the art at the time before the effective filing date of the invention to incorporate a feature extraction subnetwork to extract a feature vector from input data, a normalization layer to normalize the feature vector, and a final layer to generate a prediction output from the normalized feature vector as suggested in Wang into Nitta because both methods perform machine learning for detection (see Nitta, para. 16; see Wang, Abstract). Incorporating the teaching of Wang into Nitta would perform effective person re-identification (see Wang, para. 9).
Regarding claim 2, the combination of Nitta and Wang as applied in claim 1 above teaches all the limitations of claim 1 and further teaches wherein the processing unit is further configured to execute instructions to cause the computing system to (Nitta discloses that the learning system device comprises a CPU to execute the instructions [see para. 72-73]):
after transmitting the information about the state of the updated local model to the central server, receive, from the central server, a set of trained global parameters (Nitta discloses after the local device transmits the data of the updated local model to the server [see Nitta, para. 41 and FIG. 2], the global model sends a set of trained model parameters for the trained global model and repeats steps SA2-SA12 until the model has completed training [see Nitta, para. 48-49 and FIG. 2]);
apply the set of trained global parameters to the local model (Nitta discloses after the local device transmits the data of the updated local model to the server [see Nitta, para. 41 and FIG. 2], the global model sends a set of trained model parameters for the trained global model and repeats steps SA2-SA12 until the model has completed training [see Nitta, para. 48-49 and FIG. 2]. Thus, the local model applies the set of trained global model parameters to conduct each training cycle, and would finish by applying the completed training global model parameters);
deploy the local model after the applying (It would have been obvious to one of ordinary skill in the art, that after receiving the trained global model, that the local device can now deploy the trained model).
Regarding claim 3, the combination of Nitta and Wang as applied in claim 1 above teaches all the limitations of claim 1 and further teaches wherein the normalization layer is configured to:
receive the feature vector from the feature extraction subnetwork (Wang discloses using a feature extraction subnetwork comprising the foreground attentive subnetwork, body part subnetwork, and feature fusion subnetwork to extract a feature vector, which is then normalized by the L2 normalization layer [see Wang, Abstract, para. 26, and FIG. 1-2]);
normalize the feature vector based on a magnitude of the feature vector (Method 200 includes normalizing feature vectors on a unit sphere space (208). For example, L2 normalization layer 109 is used to regularize the magnitude of each feature vector to be unit. [see Wang, para. 26]);
output the normalized feature vector to the final layer (Wang discloses using a feature extraction subnetwork comprising the foreground attentive subnetwork, body part subnetwork, and feature fusion subnetwork to extract a feature vector, which is then normalized by the L2 normalization layer, and using the L2 normalized feature vector to learn the symmetric triplet loss for person re-identification and multi-target tracking [see Wang, Abstract, para. 26, and FIG. 1-2]);
Regarding claim 4, the combination of Nitta and Wang as applied in claim 3 above teaches all the limitations of claim 3 and further teaches:
wherein the normalization layer is configured to normalize the feature vector by dividing the feature vector by the magnitude of the feature vector (Wang discloses normalizing the feature vector by regularizing the magnitude to be the unit vector [see Wang, para. 26]. Thus, each feature vector is divided by it's magnitude to result in the unit vector with a magnitude of 1).
Regarding claim 6, the combination of Nitta and Wang as applied in claim 1 above teaches all the limitations of claim 1 and further teaches:
wherein the feature extraction subnetwork includes one or more convolutional layers (Nitta discloses that the model may be a convolutional neural network [see Nitta, para. 27], thus it would have at least one convolutional layer).
Regarding claim 8, the combination of Nitta and Wang as applied in claim 1 above teaches all the limitations of claim 1 and further teaches wherein the processing unit is further configured to execute instructions to cause the computing system to (Nitta discloses that the learning system device comprises a CPU to execute the instructions [see para. 72-73]):
prior to initialization the local model, receive, from the central server, a local model definition defining the local model to include at least the normalization layer (Nitta discloses the local device receiving the global model including the initialized parameters from the server [see Nitta, para. 29]. Thus, it would have been obvious to one of ordinary skill, that the server must also send the model structure including that the model includes a normalization layer, otherwise the model would not have a normalization layer for the model to apply parameters to).
Regarding claim 9, claim 9 contains substantially similar limitations to those found in claim 1. Therefore it is rejected for the same reason as claim 1 above. Additionally, the combination of Nitta and Wang further teaches:
a method at a computing system, the method comprising (According to one embodiment, a learning system includes a plurality of local devices and a server. [see Nitta, Abstract]).
Regarding claim 10, claim 10 contains substantially similar limitations to those found in claim 2 above. Consequently, claim 10 is rejected for the same reasons.
Regarding claim 11, claim 11 contains substantially similar limitations to those found in claim 3 above. Consequently, claim 11 is rejected for the same reasons.
Regarding claim 12, claim 12 contains substantially similar limitations to those found in claim 4 above. Consequently, claim 12 is rejected for the same reasons.
Regarding claim 14, claim 14 contains substantially similar limitations to those found in claim 6 above. Consequently, claim 14 is rejected for the same reasons.
Regarding claim 16, claim 16 contains substantially similar limitations to those found in claim 8 above. Consequently, claim 16 is rejected for the same reasons.
Regarding claim 17, Nitta teaches a computing system comprising (According to one embodiment, a learning system includes a plurality of local devices and a server. [see Nitta, Abstract]) a processing unit configured to execute instructions to cause the computing system to (Nitta discloses that the learning system device comprises a CPU to execute the instructions [see para. 72-73]):
transmit, to one or more clients, a model definition for a local model to be implemented at each of the one or more clients (Nitta discloses the local device receiving the global model including the initialized parameters from the server [see Nitta, para. 29]. Thus, it would have been obvious to one of ordinary skill, that the server must also send the model structure, otherwise how would the local device know how to apply the parameters without knowing the model structure), wherein the local model is defined to include at least: a feature extraction subnetwork to extract input data, a normalization layer, and a final layer to generate a prediction output (Nitta discloses the local device receiving the global model including the initialized parameters from the server [see Nitta, para. 29], further discloses that the local and global model may be neural networks used in machine learning [see Nitta, para. 27], may include a normalization layer [see Nitta, para. 37], and may be used for prediction for any data [see Nitta, para. 16]. Thus, it would have been obvious to one of ordinary skill, that the server must also send the model structure, otherwise how would the local device know how to apply the parameters without knowing the model structure);
implement a global model based on the model definition, the global model having a set of global parameters (Nitta discloses the local device receiving the global model including the initialized parameters from the server [see Nitta, para. 29]. Thus, it would have been obvious to one of ordinary skill, that the server must also send the model structure including that the model includes a normalization layer, otherwise the model would not have a normalization layer for the model to apply parameters to);
perform one or more rounds of training by, for each round of training (Nitta discloses repeating steps SA2-SA12 until the model has completed training [see Nitta, para. 49 and FIG. 2]);
transmitting a set of global parameters to one or more selected clients of the one or more clients (Nitta discloses the local device receiving the global model including the initialized parameters from the server [see Nitta, para. 29]);
receiving information about a state of a respective local model from each respective one or more selected clients (Nitta discloses the local device transmitting to the server local parameter and local data information for which training is completed [see Nitta, para. 41] and the server receiving the local parameter and local data information from the local devices [see Nitta, para. 42]);
aggregating the received information into an aggregated update (Nitta discloses receiving the local model parameters and local data information from the local devices [see Nitta, para. 42], generating an integrated parameter based on the received data information, for example weighted-averaging may be calculated [see Nitta, para. 43] and updating the global model based on the integrated parameter, for example replacing a parameter with a value obtained by weighted-averaging using the integrated parameter [see Nitta, para. 46]);
updating the set of global parameters using the aggregated update (Nitta discloses receiving the local model parameters and local data information from the local devices [see Nitta, para. 42], generating an integrated parameter based on the received data information, for example weighted-averaging may be calculated [see Nitta, para. 43] and updating the global model based on the integrated parameter, for example replacing a parameter with a value obtained by weighted-averaging using the integrated parameter [see Nitta, para. 46]);
after training is terminated, transmit the updated set of global parameters from a last round of training as a set of trained global parameters to all of the one or more clients (Nitta discloses after the local device transmits the data of the updated local model to the server [see Nitta, para. 41 and FIG. 2], the global model sends a set of trained model parameters for the trained global model and repeats steps SA2-SA12 until the model has completed training [see Nitta, para. 48-49 and FIG. 2]).
However, Nitta fails to teach wherein the local model includes at least: a feature extraction subnetwork to extract a feature vector from input data, a normalization layer to normalize the feature vector, and a final layer to generate a prediction output from the normalized feature vector.
In the same field of endeavor, Wang teaches:
wherein the local model includes at least: a feature extraction subnetwork to extract a feature vector from input data, a normalization layer to normalize the feature vector, and a final layer to generate a prediction output from the normalized feature vector (Wang discloses using a feature extraction subnetwork comprising the foreground attentive subnetwork, body part subnetwork, and feature fusion subnetwork to extract a feature vector, which is then normalized by the L2 normalization layer, and using the L2 normalized feature vector to learn the symmetric triplet loss for person re-identification and multi-target tracking [see Wang, Abstract, para. 26, and FIG. 1-2]).
It would have been obvious to one of ordinary skill, in the art at the time before the effective filing date of the invention to incorporate wherein the local model includes at least: a feature extraction subnetwork to extract a feature vector from input data, a normalization layer to normalize the feature vector, and a final layer to generate a prediction output from the normalized feature vector as suggested in Wang into Nitta because both methods perform machine learning for detection (see Nitta, para. 16; see Wang, Abstract). Incorporating the teaching of Wang into Nitta would perform effective person re-identification (see Wang, para. 9).
Regarding claim 18, the combination of Nitta and Wang as applied in claim 17 above teaches all the limitations of claim 17 and further teaches:
wherein the normalization layer is defined to normalize the feature vector by dividing the feature vector by a magnitude of the feature vector (Wang discloses normalizing the feature vector by regularizing the magnitude to be the unit vector [see Wang, para. 26]. Thus, each feature vector is divided by it's magnitude to result in the unit vector with a magnitude of 1).
Claims 5, 13, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over NITTA et al. (US 2024/0005172 A1), hereinafter Nitta, in view of Wang et al. (US 2020/0125925 A1), hereinafter Wang, as applied in claim 1 above, and further in view of TensorFlow API Docs ("tf.math.l2_normalize", published 01/30/2019. Retrieved from The Wayback Machine on Dec 15, 2025), hereinafter TensorFlow.
Regarding claim 5, the combination of Nitta and Wang as applied in claim 3 above teaches all the limitations of claim 3 and further teaches:
wherein the normalization layer is configured to normalize the feature vector by dividing the feature vector by the magnitude of the feature vector (Wang discloses normalizing the feature vector by regularizing the magnitude to be the unit vector [see Wang, para. 26]. Thus, each feature vector is divided by it's magnitude to result in the unit vector with a magnitude of 1).
However, the combination of Nitta and Wang fails to teach normalize the feature vector by dividing the feature vector by a larger of: the magnitude of the feature vector; or a selected threshold value.
In the same field of endeavor, TensorFlow teaches:
normalize the feature vector by dividing the feature vector by a larger of: the magnitude of the feature vector; or a selected threshold value (TensorFlow discloses normalizing a 1D feature tensor (i.e., a feature vector) by dividing the tensor by the larger of it's magnitude or epsilon [see TensorFlow, pg. 1: "output = x / sqrt(max(sum(x**2), epsilon) )"], where epsilon is the lower bound value [see TensorFlow, pg. 2, 2nd bullet point]).
It would have been obvious to one of ordinary skill, in the art at the time before the effective filing date of the invention to incorporate normalize the feature vector by dividing the feature vector by a larger of: the magnitude of the feature vector; or a selected threshold value as suggested in TensorFlow into the combination of Nitta and Wang because both methods normalize feature vectors with L2 normalization (see Wang, para. 26; see TensorFlow, pg. 1). Incorporating the L2 normalization equation of TensorFlow into the combination of Nitta and Wang would aid in extending the utility of the L2 normalization calculation to regularize the magnitude of each feature vector to be unit [see Wang, para. 26].
Regarding claims 13 and 19, claims 13 and 19 contains substantially similar limitations to those found in claim 5 above. Consequently, claims 13 and 19 are rejected for the same reasons.
Claims 7, 15, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over NITTA et al. (US 2024/0005172 A1), hereinafter Nitta, in view of Wang et al. (US 2020/0125925 A1), hereinafter Wang, and further in view of KIM et al. (US 2019/0205626 A1), hereinafter Kim.
Regarding claim 7, the combination of Nitta and Wang as applied in claim 1 above teaches all the limitations of claim 1 and further teaches:
wherein the feature extraction subnetwork includes one or more recurrent layers (Nitta discloses that the model may be a recurrent neural network [see Nitta, para. 27], thus it would have at least one recurrent layer).
However, the combination of Nitta and Wang fails to teach wherein the feature extraction subnetwork includes one or more long short-term memory (LSTM) layers.
In the same field of endeavor, Kim teaches:
wherein the feature extraction subnetwork includes one or more long short-term memory (LSTM) layers (Kim discloses a feature extraction subnetwork which incorporates LSTM layers (see Kim, para. 122-123 and FIG. 10)).
It would have been obvious to one of ordinary skill, in the art at the time before the effective filing date of the invention to incorporate wherein the feature extraction subnetwork includes one or more long short-term memory (LSTM) layers as suggested in Kim into the combination of Nitta and Wang because both methods perform machine learning with recurrent neural network layers (see Nitta, para. 27; see Kim, para. 123). Incorporating the teaching of Kim into the combination of Nitta and Wang would be a simple substitution of any recurrent layer as suggested by Nitta (see Nitta, para. 27) for choosing an LSTM recurrent layer to achieve the predictable result of feature extraction for machine learning as taught by Kim (see Kim, para. 122-123, and FIG. 10).
Regarding claims 15 and 20, claims 15 and 20 contains substantially similar limitations to those found in claim 7 above. Consequently, claims 15 and 20 are rejected for the same reasons.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
WU et al. (US 20230281462 A1) teaches performing federated learning with a model using a convolutional layer to extract the features, then using a normalization layer to normalize the features, and inputting the normalized features into a pooling layer to obtain the output as object space feature information.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JAKE BREEN whose telephone number is (571)272-0456. The examiner can normally be reached Monday - Friday, 7:00 AM - 3:00 PM 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, Jennifer Welch can be reached at (571) 272-7212. 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.
/J.T.B./Examiner, Art Unit 2143
/JENNIFER N WELCH/Supervisory Patent Examiner, Art Unit 2143