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
The cancellation of claim 15 is acknowledged, and its corresponding rejections are rendered moot.
Applicant’s arguments, see REMARKS, filed , with respect to the rejection of claims 12 – 19 and 21 under 35 U.S.C. 102 and 103 have been fully considered and are persuasive. The method of calculating the global threshold taught by the cited Zhang reference does not calculate “solely from the transmitted local threshold values” as recited in the amended independent claims. Therefore, the rejection has been withdrawn. However, upon further consideration, a new grounds of rejection is made in view of new art found in a search of the prior art prompted by the narrowed scope of the claims.
Claim Objections
Claims 12 and 18 are objected to because of the following informalities:
The phrase “transmit locally determined local threshold values as parameters independent from the first and the at least one second client” in b) should be corrected to “transmit locally determined local threshold values as independent parameters from the first and the at least one second client”. The current phrasing implies the parameters are independent from the clients, rather than from one another.
The phrase “the global model comprising global model weightings and a local model output variable” in c) should be corrected to “the global model comprising global model weightings and a global model output variable”. The Examiner apologizes for failing to note this informality in previous rejections as it did not affect antecedent basis.
Appropriate correction is required.
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 12 – 14, 18, 19 and 21 are rejected under 35 U.S.C. 103 as being unpatentable over Rey et al. (NPL, Federated Learning for Malware Detection in IoT Devices, archived version from April 2021), hereinafter Rey, in view of Mehr (US Patent Application Publication 2018/0341248, cited in previous action).
Regarding claim 12, Rey teaches a computer-implemented method for anomaly detection (Page 7 section 4.1.4, the disclosed system is an anomaly detector) in sensor data (Page 3 section 2.3, the datasets for IoT networks include device information. Specifically, the N-BaIoT dataset is used as taught in page 5 section 4. The N-BaIoT dataset includes data from multiple devices as shown in Page 5 Table 2), the method comprising:
a) generating and training a first and at least one second local model (Page 5 Fig. 1 shows that there are multiple devices/clients, each with their own local model) based on an autoencoder (Page 7 Unsupervised situation, an autoencoder is used as the model at each client to perform federated anomaly detection), each model comprising local model weightings (Page 7, the model parameters of client k, w-k) and a local model output variable (Page 7, the mean squared reconstruction errors of the client k’s autoencoder for each sample are the output of the autoencoder, with a high error indicating an anomaly), and determining a local threshold value for a respective local model output variable aided by at least one of (i) a mean value and (ii) a standard deviation of local threshold values in via a first or at least one second client (Page 7 Equation 1, the formula for the threshold of client k, thr-k, is computed by using the mean and standard deviation of the mean squared reconstruction errors of the training samples for that client; Page 8 section 4.3.3, each client computes a local threshold);
b) transmitting the local model weightings and transmitting locally determined local threshold values as parameters independent from the first and the at least one second client to a server (Page 5 Fig. 2 Step 3, the updated individual model is transmitted to the server. Page 8 section 4.2.2 notes that the server receives updated model parameters from each client; Page 8 section 4.3.3, each client transmits their local threshold to the server);
c) generating and training a global model based on the autoencoder utilizing the local model weightings, the global model comprising global model weightings and a global model output variable (Page 8 section 4.2.2, the global model weights are labelled as w without a subscript. The global output variable is the MSE from the global weights which has its accuracy tested in section 6), and determining a global threshold value for a global model output variable by computing, at the server, the global threshold value solely from the transmitted local threshold values utilizing at least one of (i) the mean value and (ii) the standard deviation of the local threshold values by the server (Page 8 section 4.3.3, the global threshold is computed as the average of the local thresholds. Another method of aggregating the global threshold is explained, but the main method described is that the average of threshold is taken, without using any other data);
d) transmitting the global model weightings and global threshold value for the global model output variable to the first client, and adopting the global model weightings for the first local model of the first client (Page 5 Fig. 2, the aggregated global model is transmitted back to the clients to be further trained and evaluated);
e) capturing first sensor data by a first sensor possessed by the first client (Page 5 Fig. 1, each device has a data acquisition step, which produces the dataset for the corresponding device; Page 7 section 4.1.5, part of the test data used for the final evaluation of the models is from data owned by the clients, i.e. sensor data from sensors possessed by clients);
f) applying the first sensor data to the first local model and determining a local model output variable of the first client (Page 11 Unsupervised situation, the final trained model is evaluated in the final evaluation step. Note that Page 5 Fig. 2 states that this final evaluation (Step 5) happens after the aggregated global model is shared to the clients (Step 4). Therefore at the point of final evaluation, the local models and the global models are identical, and the evaluation of the final model is effectively evaluation of all local models at the clients.); and
g) detecting an anomaly for the sensor data obtained by the first client, if the local model output variable of the first client is outside a range (Page 7 Unsupervised situation, if the sample has a higher MSE than the threshold, it is detected as anomalous) which is fixed by the global threshold value for the global model output variable (As a continuation of the above final evaluation using test data collected from clients, page 11 Unsupervised situation teaches that the threshold is calculated as before. Because, as stated with respect to step f, the clients are all using an identical model, and because the clients are evaluating on the same test sets as one another, the threshold calculated here is the same as the global threshold.).
Rey does not expressly teach:
that the sensor data is obtained during an industrial manufacturing process;
that the sensor data and training data are formed as image data of manufactured components.
Mehr teaches using federated learning (Paragraphs 0003, 0034, 0126, 0153 and 1064, there may be a plurality of distributed learning systems sharing data) using an autoencoder (Paragraph 0132) in an industrial manufacturing process (Paragraph 0035), and that the sensor data and training data are formed as image data of manufactured components (Paragraphs 0039, 0121 – 0123, 0164 and Figs. 7A – 7C, the systems described in Mehr may use image date). Mehr is also directed to detecting defects in objects (Paragraph 0006).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention that the federated learning system of Rey could be applied to an industrial manufacturing process, and that the sensor data and training data would be formed as image data of manufactured components, as taught by Mehr. It would be obvious because machine learning methods are well-known in the art for image analysis tasks including automated object defect classification in manufacturing (Mehr paragraph 0033, the variety of machine learning algorithms discussed are known to those of ordinary skill in the art). These known methods provide advantages in a number of industries (Mehr paragraph 0035). Mehr demonstrates that such known learning methods, which include federated learning as described in paragraph 0034 and autoencoders as described in paragraph 0132, are applicable to detecting defects in image data of manufactured components. It would be clear to one of ordinary skill in the art that the federated learning system of Rey would, similarly to the other learning systems discussed in Mehr, be applicable to detecting defects in image data of manufactured components.
Regarding claim 13, Rey in view of Mehr teaches the method as claimed in claim 12, wherein said training of a respective local model is performed with training data which is assignable to an anomaly-free state in the sensor data (Rey page 6, the training dataset includes benign traffic).
Regarding claim 14, Rey in view of Mehr teaches the method as claimed in claim 12, wherein a respective model output variable is formed by at least one parameter value (Rey page 7, the mean squared reconstruction error is computed using model parameters), and a respective threshold value is defined by at least one corresponding assignable value or a range limit of a range (Rey page 7, the threshold in Rey is an assignable value thr).
Regarding claim 18, Rey teaches:
a first and at least one second client each having a client processor and a client memory (Fig. 1, the clients each contain a local dataset, local model, and perform processing);
a sensor (Fig. 1, the devices collecting data); and
a connected server having a server processor and a server memory (Fig. 1, the server performs processing in the form of the model aggregation; Fig. 1, the server contains memory as it stores the initial and global models).
Claim 18 otherwise recites similar language to claim 12, and is similarly rejected.
Regarding claim 19, Rey in view of Mehr teaches:
the system as claimed in claim 18, wherein the sensor is an imaging sensor (Rey Table 2, the devices include cameras; Mehr paragraphs 0039, 0121 – 0123, 0164 and Figs. 7A – 7C, the systems described in Mehr may use image date).
Regarding claim 21, Rey in view of Mehr teaches the method as claimed in claim 12.
Rey in view of Mehr as applied to claim 12 does not explicitly teach a non-transitory electronically readable data carrier encoded with readable control information comprising at least a computer program which, when using the data carrier in a computing facility, implements the method as claimed in claim 12.
Mehr teaches a non-transitory electronically readable data carrier encoded with readable control information comprising at least a computer program which, when using the data carrier in a computing facility, implements a method (Paragraphs 0171 and 0172).
It would have been obvious to one of ordinary skill in the art that Rey’s method could be implemented in such a carrier as taught by Mehr. It would be obvious because it is a well-known method of performing computer methods, and this phrasing is especially common in patents.
Claim 16 is rejected under 35 U.S.C. 103 as being unpatentable over Rey in view of Mehr, further in view of Ouyang (NPL, ClusterFL: A Similarity-Aware Federated Learning System for Human Activity Recognition, cited in previous action).
Regarding claim 16, Rey in view of Mehr teaches the method as claimed in claim 12, herein the global model weightings and the global threshold value are transmitted from the server to the at least one second client (Rey Page 5 Fig. 2, the aggregated global model is transmitted back to the clients to be further trained and evaluated), which has an autoencoder with a further local model (Rey page 7 Unsupervised situation, an autoencoder is used as the model at each client to perform federated anomaly detection).
Rey in view of Mehr does not explicitly teach that similarity is determined via predefined ranges for the local model weightings between the first and at least one second clients (No determination of similarity is explicitly made).
Ouyang teaches a method for federated learning wherein similarity is determined via predefined ranged for the local model weightings between the first and at least one second clients (Page 59 column 1, the similarity is measured by a predefined distance equation for the local model weightings, measuring the distance between the weightings of each client).
It would have been obvious to one of ordinary kill in the art before the effective filing date of the invention to use a distance-based similarity measure to determine whether clients are similar. It would have been obvious because clustering clients based on similarity advantageously mitigates issues of data heterogeneity (Ouyang page 56) which is an issue Rey similarly aims to overcome (Rey page 1 section 1, the data in IoT applications is typically heterogeneous; Rey page 10 s-Resampling aims to reduce heterogeneity prior to aggregation). It would be clear to one of ordinary skill in the art that overcoming this heterogeneity is a core problem in the field of federated learning, and as such, clustering solutions such as Ouyang are a well-explored avenue of modifying federation schemes.
Claim 17 is rejected under 35 U.S.C. 103 as being unpatentable over Rey in view of Mehr, and further in view of Li (NPL, base.py, cited in previous action).
Regarding claim 17, Rey in view of Mehr teaches the method as claimed in claim 12, wherein for a respective local threshold value, metadata with respect to the local threshold value and the first and at least one second local models is also acquired, and when generating and training the global model, the acquired metadata is applied when weighting individual model weightings (Rey page 8 section 4.2.2, a weighted average could be used when aggregating the individual model parameters if the number of samples varied among clients).
Rey in view of Mehr does not explicitly state that the metadata is acquired by the first or at least one second client, and transmitted to the server.
Li teaches an implementation of federated learning in which the clients transmit the metadata applied when weighting individual model weightings to the server (Function local_train, the variables in solns are each returned from the client local_train functions, solns including the number of samples of the client; Function aggregate, the weighted sum is taken using the number of samples in each item of solns).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention that the clients provide the metadata for weighting the individual model weightings. It would be obvious because there are a finite number of possible means to acquire the metadata: either by the clients or by the server. It would be clear to one of ordinary skill in the art that an implementation utilizing the client to acquire metadata would be an effective option to try.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 BRYAN PAI SONG HUANG whose telephone number is (571)272-0510. The examiner can normally be reached Monday - Friday 11:30 AM - 8:30 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, ASHISH THOMAS can be reached at (571) 272-0631. 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.
/B.P.H./ Examiner, Art Unit 2114
/ASHISH THOMAS/ Supervisory Patent Examiner, Art Unit 2114
Prosecution Insights