Notice of Pre-AIA or AIA Status
1. 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 action is in response to the original filing on 02/26/2024. Claims 1-26 and 28-30 are pending and have been considered below.
Information Disclosure Statement
3. The information disclosure statement (IDS(s)) submitted on 02/26/2024, 05/02/2025 is/are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Rejections – 35 USC § 103
4. 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 of this title, 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.
5. Claims 1-6, 10, 15-20, 24, and 30 are rejected under 35 U.S.C. 103 as being unpatentable over McMahan et al. (U.S. Patent Application Pub. No. US 20190340534 A1) in view of Chen et al. (FEDBE: Making Bayesian Model Ensemble Applicable to Federated Learning, arXiv, published 30 Jan 2021, pages 1-19) and further in view of Gu et al. (Fast Federated Learning in the Presence of Arbitrary Device Unavailability, arXiv, published 8 Jun 2021, pages 1-44).
Claim 1: McMahan teaches a method for distributed machine learning (ML) at a central computing device (i.e. federated learning is a machine learning framework that enables training of a high-quality centralized model based on training data distributed over a large number of client computing devices; para. [0019]), the method comprising:
providing a global ML model to a plurality of local computing devices (i.e. Server 104 can be configured to access machine learning model 106, and to provide model 106 to a plurality of client devices 102; para. [0055]), wherein the global ML model includes a plurality of parameters (i.e. In an example federated learning framework, an objective is to learn a model with parameters embodied in a real matrix W; para. [0027]);
receiving, from each local computing device in a subset of the plurality of local computing devices, a local ML model updated based on the global ML model (i.e. a subset of clients are selected; each client in the subset updates the model based on their local data; the updated models or model updates are sent by each client to the server; para. [0022]), wherein the local ML model includes weights with values corresponding to one or more of the plurality of parameters (i.e. The updated local models are Wt 1, Wt 2, . . . , Wt n; para. [0028]);
updating the global ML model by averaging the representative values for each weight for each of the plurality of parameters (i.e. the server aggregates the updates (e.g., by averaging the updates) and improves the global mode; para. [0022, 0029, 0060]).
McMahan does not explicitly teach constructing, for each weight value in each of the received local ML models, a probability distribution for each of the plurality of parameters with corresponding received weight values; sampling, using the constructed probability distribution for each weight value for each of the plurality of parameters, for all of the plurality of local computing devices to generate representative values for each weight.
However, Chen teaches constructing, for each weight value in each of the received local ML models, a probability distribution for each of the plurality of parameters with corresponding received weight values (i.e. we propose to fit a diagonal Gaussian distribution N(µ,Σdiag) to the clients’ models {wi}; Section 3.3, pages 4-5); sampling, using the constructed probability distribution for each weight value for each of the plurality of parameters, of the plurality of local computing devices to generate representative values for each weight (i.e. Construct global model distribution p(w|D) from {wi;i ∈ S}, Sample M global models {w(m) ∼ p(w|D)}M m=1; Algorithm 1, page 5).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the invention of McMahan to include the feature of Chen. One would have been motivated to make this modification because FedAvg can degrade under non-IID client and that fitting a Gaussian or Dirichlet distribution to local models and sampling global models provides more robust aggregation.
However, Gu teaches for all of the plurality of local computing devices (i.e. our algorithm MIFA updates the model with the latest available accumulated gradients for all devices; Section 4, pages 4-5).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of McMahan and Chen to include the feature of Gu. One would have been motivated to make this modification because it reduces bias caused by partial participation and improves robustness of the global model.
Claim 2: McMahan, Chen, and Gu teach the method according to claim 1. McMahan further teaches wherein, after updating the global ML model, the steps of the method are repeated for a predetermined number of iterations i (i.e. systems implementing federated learning can perform the following actions in each of a plurality of rounds of model optimization: a subset of clients are selected; each client in the subset updates the model based on their local data; the updated models or model updates are sent by each client to the server; the server aggregates the updates (e.g., by averaging the updates) and improves the global model; and the server redistributes the global model to all the clients. Performing a plurality of rounds of the above actions iteratively improves the global model based on training data stored at the client devices; para. [0022, 0090]).
Claim 3: McMahan, Chen, and Gu teach the method according to claim 2. McMahan further teaches wherein updating of each of the received local ML models based on the global ML model is according to:
PNG
media_image1.png
426
636
media_image1.png
Greyscale
(i.e. In round t≥0, the server distributes the current model Wt to a subset St of nt clients (for example, to a selected subset of clients whose devices are plugged into power, have access to broadband, and are idle). Some or all of these clients independently update the model based on their local data. The updated local models are Wt 1, Wt 2, . . . , Wt n. Let the updates be: H t i =W t i −W t ,i∈S t; para. [0028]).
Claim 4: McMahan, Chen, and Gu teach the method according to claim 3. McMahan does not explicitly teach wherein the constructed probability distribution for each weight value for each of the plurality of parameters is according to:S N(f p), p=1, . . . ,W, where
W is the total number of weights in the global ML model;
SN is the function for sampling the N local computing devices; and
fp is the constructed probability distribution for each weight value for each of the plurality of parameters.
However, Chen further teaches wherein the constructed probability distribution for each weight value for each of the plurality of parameters is according to:S N(f p), p=1, . . . ,W, where
W is the total number of weights in the global ML model;
SN is the function for sampling the N local computing devices; and
fp is the constructed probability distribution for each weight value for each of the plurality of parameters (Section 3.3 and Algorithm 1).
PNG
media_image2.png
452
444
media_image2.png
Greyscale
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the invention of McMahan to include the feature of Chen. One would have been motivated to make this modification because FedAvg can degrade under non-IID client and that fitting a Gaussian or Dirichlet distribution to local models and sampling global models provides more robust aggregation.
Claim 5: McMahan, Chen, and Gu teach the method according to claim 4. McMahan further teaches wherein updating the global ML model by averaging the representative values for each weight for each of the plurality of parameters (i.e. the server aggregates the updates (e.g., by averaging the updates) and improves the global mode, Each client then sends the update back to the server, where the global update is computed by aggregating all the client-side updates. For example, one aggregation scheme can be as follows: Wt+1=Wt+ηtHt,Ht:=1nt∑i∈StHti; para. [0022, 0029, 0060]).
McMahan does not explicitly teach G1=1N∑K=1NSN(ft), t=1,…,W, where
G1 is the updated global ML model;
W is the total number of weights in the global ML model;
SN is the function for sampling the N local computing devices; and
ft is the constructed probability distribution for each weight value for each of the plurality of parameters.
However, Chen further teaches G1=1N∑K=1NSN(ft), t=1,…,W, where
G1 is the updated global ML model;
W is the total number of weights in the global ML model;
SN is the function for sampling the N local computing devices; and
ft is the constructed probability distribution for each weight value for each of the plurality of parameters (i.e. we propose to fit a diagonal Gaussian distribution N(µ,Σdiag) to the clients’ models {wi} following equation 5, Sample {w(m) ∼ p(w|D)}M m=1 and perform ensemble; Section 3.3).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the invention of McMahan to include the feature of Chen. One would have been motivated to make this modification because FedAvg can degrade under non-IID client and that fitting a Gaussian or Dirichlet distribution to local models and sampling global models provides more robust aggregation.
However, Gu further teaches SN is the function for sampling the N local computing devices (i.e. a Federated Learning setting: min w∈Rd f(w) := 1 N N i=1fi(w) := 1 N N i=1Eξi [fi(w,ξi)], (1) where w is the optimization variable, e.g., parameters of a machine learning model, N is the number of participating devices, fi is the local loss function on device i, and ξi describes the randomness in local data distribution. We say a device participates or is active at round t if it can complete the computation task and send back the update at the end of round t … our algorithm MIFA updates the model with the latest available accumulated gradients for all devices; Sections 3, 4, Algorithm 1, pages 3-5).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of McMahan and Chen to include the feature of Gu. One would have been motivated to make this modification because it reduces bias caused by partial participation and improves robustness of the global model.
Claim 6: McMahan, Chen, and Gu teach the method according to claim 1. McMahan further teaches wherein the global ML model is one or a combination of: a convolutional neural network (CNN), a artificial neural network (ANN), and a recurrent neural network (RNN) (i.e. the model can include one or more neural networks (e.g., deep neural networks, recurrent neural networks, convolutional neural networks, etc.) or other machine-learned models; para. [0027]).
Claim 10: McMahan, Chen, and Gu teach the method according to claim 1. McMahan further teaches comprising providing the updated global ML model to the plurality of local computing devices (i.e. the server aggregates the updates (e.g., by averaging the updates) and improves the global model; and the server redistributes the global model to all the clients; para. [0022]).
Claim 15 is similar in scope to Claims 1 and is rejected under a similar rationale.
McMahan teaches a central computing device comprising: a memory; and a processor coupled to the memory, wherein the processor is configured to (i.e. The one or more memory devices 214 can store information accessible by the one or more processors 212, including computer-readable instructions 216 that can be executed by the one or more processors 212; para. [0065]).
Claims 16-20 and 24 are similar in scope to Claims 2-6, 10 and are rejected under a similar rationale.
Claim 30: McMahan teaches a computer program product comprising a non-transitory computer readable medium storing a computer program comprising instructions which, when executed by processing circuitry, causes the processing circuitry to perform the method (i.e. The client device includes at least one processor and at least one non-transitory computer-readable medium that stores instructions that, when executed by the at least one processor, cause the client computing device to perform operations; para. [0005]) according to claim 1 (see rejection of claim 1 above).
6. Claims 7, 11, 21, and 25 are rejected under 35 U.S.C. 103 as being unpatentable over McMahan in view of Chen, Gu, and further in view of Cote et al. (U.S. Patent Application Pub. No. US 20190303726 A1).
Claim 7: McMahan, Chen, and Gu teach the method according to claim 1. McMahan does not explicitly teach wherein the plurality of parameters and weights correspond to an alarm dataset for a telecommunications operator, and the global ML model is a classifier-type model that classifies alarms as either a true alarm or a false alarm.
However, Cote teaches wherein the plurality of parameters and weights correspond to an alarm dataset (i.e. obtain network data including first data of devices and services in the network, Performance Monitoring (PM) data associated with the devices and services and with associated timestamps, and second data including any of tickets, alarms, and events; para. [0011]) for a telecommunications operator (i.e. The software application can be in a Networks Operations Center (NOC) or the like and can continuously operate to provide the actionable insights; para. [0041]), and the global ML model is a classifier-type model (i.e. For applying ML for networking applications, tasks that can be performed as “read-only” operations on the network, namely: classification, anomaly detection and regression (trends); para. [0066]) that classifies alarms as either a true alarm or a false alarm (i.e. The output distribution is continuous between 0 and 1, with “normal” outputs towards zero and “abnormal” outputs towards one. This can be turned into a binary classifier with a cut-off threshold illustrated by the dashed line; para. [0064]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of McMahan, Chen, and Gu to include the feature of Cote. One would have been motivated to make this modification because it improves network health prediction and reduces false positive alarms.
Claim 11: McMahan, Chen, and Gu teach the method according to claim 10. McMahan does not explicitly teach a plurality of radio network nodes which are configured to classify an alarm type.
However, Cote teaches a plurality of radio network nodes which are configured to classify an alarm type (i.e. obtain network data including first data of devices and services in the network, Performance Monitoring (PM) data associated with the devices and services and with associated timestamps, and second data including any of tickets, alarms, and events … The output distribution is continuous between 0 and 1, with “normal” outputs towards zero and “abnormal” outputs towards one. This can be turned into a binary classifier with a cut-off threshold illustrated by the dashed line; para. [0011, 0050, 0064]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of McMahan, Chen, and Gu to include the feature of Cote. One would have been motivated to make this modification because it improves network health prediction and reduces false positive alarms.
Claims 21 and 25 are similar in scope to Claims 7, 11 and are rejected under a similar rationale.
7. Claims 8, 9, 22, and 23 are rejected under 35 U.S.C. 103 as being unpatentable over McMahan in view of Chen, Gu, and further in view of Pichara et al. (U.S. Patent Application Pub. No. US 20210037863 A1).
Claim 8: McMahan, Chen, and Gu teach the method according to claim 1. McMahan does not explicitly teach wherein the plurality of parameters and weights correspond to an internet of senses dataset for one of: sight, sound, taste, smell and touch sensations, and the global ML model is a classifier-type model that classifies sensations.
However, Pichara teaches wherein the plurality of parameters and weights correspond to an internet of senses dataset for one of: sight, sound, taste, smell and touch sensations, and the global ML model is a classifier-type model that classifies sensations (i.e. a set of taste level classifiers (and/or other suitable taste models; etc.) may be used for determining taste level (e.g., from 1 to 5) for a set of taste descriptors (e.g., salty, bitter, umami, acid, and sweet, etc.); a multi-label flavor classifier model can be used in determining flavor categories; and a set of flavor predictor classifier models can be used for determining flavor descriptors; where any suitable combination of outputs from any suitable combination of models can be used in generating a flavor profile describing any suitable combination of ingredients (e.g., a recipe, etc.); para. [0062]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of McMahan, Chen, and Gu to include the feature of Pichara. One would have been motivated to make this modification because it improves taste classification using data from multiple local sources while retaining the benefits of distributed model training.
Claim 9: McMahan, Chen, Gu, and Pichara teach the method according to claim 8. McMahan does not explicitly teach wherein the plurality of parameters and weights correspond to an internet of senses dataset for taste sensations and the global ML model is a classifier-type model that classifies taste sensations as being from one to all five of the basic taste sensations.
However, Pichara further teaches wherein the plurality of parameters and weights correspond to an internet of senses dataset for taste sensations and the global ML model is a classifier-type model that classifies taste sensations as being from one to all five of the basic taste sensations (i.e. a set of taste level classifiers (and/or other suitable taste models; etc.) may be used for determining taste level (e.g., from 1 to 5) for a set of taste descriptors (e.g., salty, bitter, umami, acid, and sweet, etc.); a multi-label flavor classifier model can be used in determining flavor categories; and a set of flavor predictor classifier models can be used for determining flavor descriptors; where any suitable combination of outputs from any suitable combination of models can be used in generating a flavor profile describing any suitable combination of ingredients (e.g., a recipe, etc.); para. [0062]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of McMahan, Chen, and Gu to include the feature of Pichara. One would have been motivated to make this modification because it improves local anomaly classification.
Claims 22 and 23 are similar in scope to Claims 8, 9 and are rejected under a similar rationale.
8. Claims 12 and 26 are rejected under 35 U.S.C. 103 as being unpatentable over McMahan in view of Chen, Gu, and further in view of Stubbs et al. (U.S. Patent Application Pub. No. US 20210133607 A1).
Claim 12: McMahan, Chen, and Gu teach the method according to claim 10. McMahan does not explicitly teach a plurality of wireless sensor devices which are configured to classify an alarm type.
However, Stubbs teaches wherein the plurality of local computing devices comprises a plurality of wireless sensor devices which are configured to classify an alarm type (i.e. the one or more edge devices are configured to determine using the machine learning inference engine if an anomaly is detected in data received from the one or more sensors, and if so: communicate with the one or more signal processors to provide at least some of the data; the one or more cloud-based processors configured to receive at least some of the data and attempt to classify the anomaly using the narrow-band classifier and the machine learning-classifier; if the one or more signal processors cannot classify the anomaly, communicating with the work order system to generate a work order for a technician visit; if the one or more cloud-based processors classifies the anomaly, determine alerts, corrective actions, and/or maintenance requests associated with the anomaly; para. [0012]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of McMahan, Chen, and Gu to include the feature of Stubbs. One would have been motivated to make this modification because it improves taste classification using data from multiple local sources while retaining the benefits of distributed model training.
Claim 26 is similar in scope to Claims 12 and is rejected under a similar rationale.
9. Claims 13 and 28 are rejected under 35 U.S.C. 103 as being unpatentable over McMahan in view of Chen, Gu, Pichara, and further in view of Govea et al. (U.S. Patent Application Pub. No. US 20200327397 A1).
Claim 13: McMahan, Chen, Gu, and Pichara teach the method according to claim 10. McMahan does not explicitly teach a plurality of radio network nodes which are configured to classify internet of senses sensations.
However, Pichara teaches classify internet of senses sensations (i.e. a set of taste level classifiers (and/or other suitable taste models; etc.) may be used for determining taste level (e.g., from 1 to 5) for a set of taste descriptors (e.g., salty, bitter, umami, acid, and sweet, etc.); a multi-label flavor classifier model can be used in determining flavor categories; and a set of flavor predictor classifier models can be used for determining flavor descriptors; where any suitable combination of outputs from any suitable combination of models can be used in generating a flavor profile describing any suitable combination of ingredients (e.g., a recipe, etc.); para. [0062]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of McMahan, Chen, and Gu to include the feature of Pichara. One would have been motivated to make this modification because it improves local anomaly classification.
However, Govea teaches wherein the plurality of local computing devices comprises a plurality of radio network nodes which are configured to classify (i.e. the techniques described herein for classifying baseband signals derived from radio frequency signals with respect to modulation type using multiple data representations or for training a consolidated neural network to perform such classifications may be implemented by a modulation classification processing unit of a base station or a mobile station; para. [0076]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of McMahan, Chen, Gu, and Pichara to include the feature of Govea. One would have been motivated to make this modification because it supplies the radio network node implementation for classification problem.
Claim 28 is similar in scope to Claims 13 and is rejected under a similar rationale.
10. Claims 14 and 29 are rejected under 35 U.S.C. 103 as being unpatentable over McMahan in view of Chen, Gu, Pichara, and further in view of Stubbs et al. (U.S. Patent Application Pub. No. US 20210133607 A1).
Claim 14: McMahan, Chen, Gu, and Pichara teach the method according to claim 10. McMahan does not explicitly teach wherein the plurality of local computing devices comprises a plurality of wireless sensor devices which are configured to classify internet of senses sensations.
However, Pichara teaches to classify internet of senses sensations (i.e. a set of taste level classifiers (and/or other suitable taste models; etc.) may be used for determining taste level (e.g., from 1 to 5) for a set of taste descriptors (e.g., salty, bitter, umami, acid, and sweet, etc.); a multi-label flavor classifier model can be used in determining flavor categories; and a set of flavor predictor classifier models can be used for determining flavor descriptors; where any suitable combination of outputs from any suitable combination of models can be used in generating a flavor profile describing any suitable combination of ingredients (e.g., a recipe, etc.); para. [0062]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of McMahan, Chen, and Gu to include the feature of Pichara. One would have been motivated to make this modification because it improves local anomaly classification.
However, Stubbs teaches wherein the plurality of local computing devices comprises a plurality of wireless sensor devices which are configured to classify (i.e. the one or more edge devices are configured to determine using the machine learning inference engine if an anomaly is detected in data received from the one or more sensors, and if so: communicate with the one or more signal processors to provide at least some of the data; the one or more cloud-based processors configured to receive at least some of the data and attempt to classify the anomaly using the narrow-band classifier and the machine learning-classifier; if the one or more signal processors cannot classify the anomaly, communicating with the work order system to generate a work order for a technician visit; if the one or more cloud-based processors classifies the anomaly, determine alerts, corrective actions, and/or maintenance requests associated with the anomaly; para. [0012]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of McMahan, Chen, Gu, and Pichara to include the feature of Stubbs. One would have been motivated to make this modification because it improves taste classification using data from multiple local sources while retaining the benefits of distributed model training.
Claim 29 is similar in scope to Claims 14 and is rejected under a similar rationale.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure.
Samek et al. (Pub. No. US 20220108177 A1), A concept for Federated Learning which is more efficient and/or robust is presented. Beyond this, concepts for specifying clients and/or measuring training data similarities in a manner more suitable for being applied in Federated Learning environments, are described.
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. A reference is relevant for all it contains and may be relied upon for all that it would have reasonably suggested to one having ordinary skill in the art. In re Heck, 699 F.2d 1331, 1332-33, 216 U.S.P.Q. 1038, 1039 (Fed. Cir. 1983) (quoting In re Lemelson, 397 F.2d 1006, 1009, 158 U.S.P.Q. 275, 277 (C.C.P.A. 1968)).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TAN TRAN whose telephone number is (303)297-4266. The examiner can normally be reached on Monday - Thursday - 8:00 am - 5:00 pm MT.
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, Matt 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://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.
/TAN H TRAN/Primary Examiner, Art Unit 2141
Prosecution Insights