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 .
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, 2, 4, 8, 9, 11, 15, 16, 18 are rejected under 35 U.S.C. 103 as being unpatentable over Mathews (US 20210097176, hereafter referred to as Mathews), in view of Kaiming He (Identity Mappings in Deep Residual Networks, hereafter referred to as He), Zhang (US 20210397965 A1, hereafter referred to as Zhang), and Elkerdawy (To filter prune, or to layer prune, that is the question, hereafter referred to as Elkerdawy).
Regarding claim 1, Mathews discloses “A method for detecting software attack, comprising: receiving, from a first machine learning model…” (Paragraph 0026; The DL model is the first machine learning model)
“… one or more neurons of the first machine learning model” (Paragraph 0024; shows that the neural network is made up of neurons)
“… is obtained when the first machine learning model processes a production data sample to generate a prediction outcome” (Paragraph 0026; This denotes the DL model producing output based on production data sample inputs, which include adversarial or drift data samples).
“wherein the production data sample is a software code and the prediction outcome indicates whether the software code has risk of malware…;” (Paragraph 0025, 0026; DL model 108 obtains the production data sample as an executable (software code) from the DL training server 102 and processes the production data sample to determine whether the executable has a risk of malware)
“using, a second machine learning model to process” (Paragraph 0019; This denotes the Adversarial attack detector performing classifications (i.e. a second machine learning model)) “to generate a distribution assessment” (Paragraph 0015; This denotes looking for concept drift, which looks for changes in a distribution).
“determining, based on the distribution assessment, whether the production data sample is an adversarial data sample or a drift data sample” (Paragraph 0026; The adversarial attack detector determines whether the sample is evidence of drift data or adversarial data.)
“and in response to detecting an attack by the production data sample, triggering an incident response” (Paragraph 0022, 0026; After the adversarial attack detector processes the production sample, which is the results of the DL model 108, triggers an incident response)
Mathews fails to explicitly disclose, “pre-activation data, wherein the pre-activation data comprises pre-activation”, and “the pre-activation data”.
However, He discloses, “pre-activation data, wherein the pre-activation data comprises pre-activation”, and “the pre-activation data” (Pg. 11, section 4.2; These paragraphs show the use of pre-activation data in training and optimizing machine learning models).
Therefore, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Mathews by the implementation of pre-activation models that is taught by He, to make the invention specify using pre-activation data for improved machine learning model training; thus, one of ordinary skill in the arts would be motivated to combine the references since it provides better performance and stabilized training results to their machine learning model due to improved optimization and regularization, as He teaches that “We find the impact of pre-activation is twofold. First, the optimization is further eased (comparing with the baseline ResNet) because f is an identity mapping. Second, using BN as pre-activation improves regularization of the models” (He, Page 11, Section 4.2, paragraph 1).
Mathews fails to explicitly disclose, “and wherein the one or more neurons belong to at least one layer of the first machine learning model, and the at least one layer is selected based on an importance level associated with each layer of the first machine learning model and a configured threshold importance level;”.
Zhang discloses “and wherein the one or more neurons belong to at least one layer of the first machine learning model,” (Paragraph 0091; neurons are present in each layer)
“…an importance level associated with each layer of the first machine learning model and a configured threshold importance level;” (Paragraph 0114, 0116; importance of a layer is calculated and this calculation can be done for each layer)
Zhang fails to explicitly disclose, “the at least one layer is selected based on an importance level”.
Elkerdawy discloses “the at least one layer is selected based on an importance level” (Section 2, Layer pruning, paragraph 2-3, Page 6; LayerPrune is a method of pruning the least important layers of a Convolutional Neural Network to leave only the most important layers based on an importance level)
Therefore, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention having Mathews, Zhang, and Elkerdawy to modify Mathews by making neurons present in each layer of the machine learning model and associating each layer of the machine learning model with an importance level, as well as selecting at least one layer based on an importance level. One would be motivated to do so in order to select at least one or more layers in the machine learning model that have information on neurons that the user is searching for.
Regarding claim 2, Mathews discloses “in response to determining that the production data sample is the adversarial data sample,” (Paragraph 0025; the executable is the production data sample and is determined to be either adversarial or drift) “determining whether an attack has been detected based on a configured policy” (Paragraph 0026; based off of the results of the adversarial attack detector, an attack has been detected if an adversarial data sample has been determined) “and storing the adversarial data sample” (Paragraph 0064; The adversarial data sample, which is an input file, is transmitted to a DL training server for storage (518 of FIG. 5)) “for a retraining of the first machine learning model” (Paragraph 0078; The example report from the results of the detector are used to retrain the first machine learning model).
Regarding claim 4, Matthews discloses “in response to determining that the production data sample is the drift data sample,” (Paragraph 0025; the executable is the production data sample and is determined to be either adversarial or drift) “determining whether to retrain the first machine learning model based on a configured policy” (Paragraph 0064; A report is transmitted to the first machine learning model and determines that the model should be retrained based on the information in the report) “and storing the drift data sample” (Paragraph 0064; The drift data sample, which is an input file, is transmitted to a DL training server for storage (518 of FIG. 5)).
Regarding claims 8 and 15, these claims are similar in scope to claim 1.
Regarding claims 9 and 16, these claims are similar in scope to claim 2.
Regarding claims 11 and 18, these claims are similar in scope to claim 4.
Claim Rejections - 35 USC § 103
Claims 7, 14, 21 are rejected under 35 U.S.C. 103 as being unpatentable over Mathews (US 20210097176, hereafter referred to as Mathews), in view of Kaiming He (Identity Mappings in Deep Residual Networks, hereafter referred to as He), Zhang (US 20210397965 A1, hereafter referred to as Zhang), and Elkerdawy (To filter prune, or to layer prune, that is the question, hereafter referred to as Elkerdawy), and further in view of Dawkins (Paul's Online Notes, hereafter referred to as Dawkins) and Hunter (US 20230004800 A1, hereafter referred to as Hunter).
Regarding claim 7, Mathews discloses “one or more neurons” (Paragraph 0024; This shows the neural network is made up of neurons).
Mathews fails to explicitly disclose, “the pre-activation data is a vector that includes the flattened pre-activation tensors”.
Dawkins discloses “the pre-activation data is a vector” (Section 11.1, 11.2; Vectors are useful for storing things and are resizable).
Dawkins fails to explicitly disclose, “flattened pre-activation tensors”.
However, Hunter discloses, “flattened pre-activation tensors” (Paragraph 0164; the process of combining tensors to associate them with vectors is done by flattening them into a one-dimensional column, which generates a vector).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify how Mathews manipulates neurons with pre-activation data by formatting pre-activation data as a vector to offer the advantages listed by Dawkins which include the ability to store information as well as being able to resize the vector, which is taught in “Vectors are used to represent quantities that have both a magnitude and a direction.” (Dawkins, Page 1, Paragraph 1) and “we can see that if c is positive all scalar multiplication will do is stretch (if c > 1) or shrink (if c < 1) the original vector, but it won’t change the direction” (Dawkins, Page 8, Paragraph 2). Additionally, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify Dawkins with the addition of flattened pre-activation tensors that is taught by Hunter, in order to compress the tensors that are from the neurons, into vectors. This transforms them into a format that is usable by the second machine learning model, and is taught as follows, “The combined dense tensors are associated with state vectors that include sparse tensor identifiers for the active weight values. The collection of the combined dense tensors and the state vectors may be denoted as augmented weight tensors (AWT). The complementary sparse tensors 810 are combined into a smaller number (L) of dense complementary sparse filter blocks (CSFBs) 820. The dense CSFBs 820 are examples of combined dense tensors. Each of these dense CSFBs is flattened into a one-dimensional column. The collection of the one-dimensional columns are concatenated horizontally into an AWT 830 that has K ports.” (Hunter, Paragraph 164). Thus; one of ordinary skill in the arts would be motivated to combine the references since it provides a scalable method of storing pre-activation data that is created from flattening tensors into a format that can be accepted by the second machine learning model.
Regarding claim 14, this claim is similar in scope to claim 7.
Response to Arguments
The objections to the drawing have been withdrawn in view of applicant’s amendments.
The objections to the specification have been withdrawn in view of applicant’s amendments.
The objections to claims 1, 7, 8, and 15 have been withdrawn in view of applicant’s amendments.
The 35 USC 101 rejections for claims 8-14 have been withdrawn in view of applicant’s amendments.
The 35 USC 101 rejections for claims 1-20 have been withdrawn in view of applicant’s amendments.
Applicant's arguments regarding the 35 USC 103 rejection are moot in view of the new grounds of rejection necessitated by applicant's amendments.
Zhang teaches that one or more neurons belong to a layer of a machine learning model in paragraph 0091 and also teaches that the importance level of a layer is associated with each layer of a machine learning model in paragraphs 0114 and 0116.
However, Zhang does not teach that at least one layer is selected by an importance level. Elkerdawy teaches this in Section 2, page 6 in “Layer pruning” where it mentions a method called LayerPrune, a method of selecting the layers by importance level for the purpose of pruning the least important layers according to their importance level.
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 DAVID KIM whose telephone number is (571)272-4331. The examiner can normally be reached 7:30 AM - 4:30 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, Matthew Ell can be reached at (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 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.
/D.K./Examiner, Art Unit 2141
/MATTHEW ELL/Supervisory Patent Examiner, Art Unit 2141
Prosecution Insights