DETAILED ACTION
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 . This action is made final.
This Office Action is in response to the amendments file on March 25, 2026.
Claims 1, 10, and 20 have been amended.
Claims 2, 5, 11, and 14 have been cancelled.
Claims 1, 3, 4, 6-10, 12, 13, and 15-22 remain pending.
Response to Amendment
The amendment filed March 25, 2026has been entered. Claims 1, 3, 4, 6-10, 12, 13, and 15-22, remain pending in the application.
Response to Arguments
Regarding the DP arguments on page 10
Applicant's arguments filed March 25, 2026 have been fully considered but they are not persuasive. The amended claims remain not patently distinct from the claims of copending application No. US 12/061,692 B2. In particular, the copending claims recite the same two stage classified training architecture, including training at course 1 class classifier and a benign/noise one class classifier, applying an ensemble of the classifiers to filter events from a first dataset to create a third training set, and training a final one class classifier using the third training set. Co pending claims further recite that the first data set includes a systemwide trace and a sequence of events ordered based on process and thread information. Accordingly, the amendment does not render the instant claims patently distinct, in the provisional nonstatutory double patenting rejection is maintained. .
Regarding the 101 arguments
Applicant’s arguments, see pages 10-14, filed March 25, 2026, with respect to claims 1, 3, 4, 6-10, 12, 13, and 15-22 have been fully considered and are persuasive. The rejection of December 30, 2025 has been withdrawn.
Regarding the 103 rejection
Applicant’s arguments with respect to claim(s) 1, 3, 4, 6-10, 12, 13, and 15-22 have been considered but are moot because the present rejection sets forth a new ground of rejection that relies on newly applied reference(s) for the newly added limitation(s) and/or amended claim language.
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.
Claim(s) 1, 3, 4, 6-10, 12, 13, and 15-21 is/are rejected under 35 U.S.C. 103 as being unpatentable over Givental et al. (US 20210281592 A1, referred to as Givental), in view of Brown (US 20190205530 A1, referred to as Brown), in view of Palani et al. (US 20210344695 A1, referred to as Palani).
Regarding claim 1, Givental teaches A computer-implemented method of training a neural network ([0005]: “a method, in a data processing system comprising at least one processor and at least one memory, the memory comprising instructions executed by the at least one processor to cause the at least one processor to be specifically configured to implement a hybrid machine learning (ML) anomaly detector”), the method comprising:
in a first stage of training:
training a coarse machine learning one-class classifier using a first training set including a signal and noise ([0047-0051]: Describes training machine learning models, including support vector machines, which function as one-class classifiers. These classifiers are trained on input data that includes both anomalous (signal) and non-anomalous (noise) data, corresponding to a first training set used for training a coarse one-class classifier.), wherein the signal comprises information of interest for detecting an anomaly, and wherein the information of interest for detecting the anomaly comprises anomalous data and non-anomalous data(FIG. 1, and [0043-0047]: Describes that the system gathers and logs information, after this is transmitted to the data cleaning and feature engineering engine 110, it parses the data into desired formats to send to the hybrid anomaly detection models in the machine learning model ensemble 120. This system gathers information that contains data of interest and not of interest, it then sends it to be formatted, and further to be processed to determine which of the data values contain anomalous and non-anomalous data.); and
Although Givental teaches training one-class classifiers on a first training set including signal and noise data, it does not teach the first training set comprising a pre- processed system-wide trace including a sequence of events ordered based on process and thread information.
Brown teaches the first training set comprising a pre- processed system-wide trace including a sequence of events ordered based on process and thread information ([0016]: Describes receiving or processing event streams or sequences indicating activities of system components such as processes or threads. ;[0030] and [0066-0072]: Describes monitored computing devices running security agents that provide event records for analysis and collectors/filters/event consumers that observe events on the computing device, filter event data, and generate event records. ;[0077-0081], and [0096-0099]: Describes that the detected events form a sequence of events ordered substantially in the order they occurred or were detected The event records may include process related and thread process context information, including PID’s, parent PID’s, process group ids, command lines, IPC mechanisms, stack entries, and related process information.)
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combined Givental’s training set with Brown’s preprocessed system wide event trace. Doing so would have enabled the system to Improve anomaly detection by capturing relationships between events generated by related processes and threads, thereby reducing the likelihood that malicious behavior would be missed when individual log events are analyzed without their execution context.
Although Givental in view of Brown, teaches training one-class classifiers on a first training set including signal and noise data, it does not teach training separate classifiers on a second training set which excludes the signal data. Palani teaches training a noise machine learning one-class classifier using a second training set excluding the signal ([0032]: Describes removing outlier (signal) data from a dataset, this creates a dataset that has been filtered of outliers/anomalies, to create clustered time series training data 118. Removing outliers from a dataset corresponds to creating a new dataset excluding signals, which is then used to train the ensemble of deep learning models 110, on a single class of normal data which excludes the noise. These models correspond to a one-class classifier as they are trained on the cleaned dataset which excludes anomalous data.; [0037]: Describes training deep learning models on the resulting time clustered training data, corresponding to a second training set excluding the signal data for training.); and
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine the teachings of Givental and Palani in order to improve the precision and robustness of anomaly detection. One would have been motivated to use the one-class classifier architectures to separate signal and noise data for training of the models. This combination would have been predictable resulting in improved classier performance, use of known methods (applying multiple classifiers to refine data) to achieve an expected result (improved detection performance), and would thus have been obvious.
Givental further teaches: applying an ensemble of models including the noise machine learning one-class classifier and the coarse machine learning one-class classifier to the first training set to create a third training set representing the signal for a second stage of training ([0052-0055]: Describes applying an ensemble of classifiers including models trained on different datasets, to noisy or unlabeled input data, and using ensemble outputs to generate a refined, labeled dataset. This corresponds to applying both coarse and noise classifiers to the original training set to generate a new training set which represents the signal data.), wherein the applying of the ensemble of models comprises adding a data point to the third training set([0020-0024] describes applying an ensemble of unsupervised machine learning models to input data to generate anomaly classification that distinguish signal from noise. The modes (one-class SVM, isolation forest, local outlier factor) applied to determine whether data points represent anomalous or non-anomalous behavior to separate signal of interest form noise. ; [0055-0059]: Describes selecting a subset of data points based on the outputs of the ensemble and using the selected data points to form a refined training dataset for a subsequent stage of training. The selected data points identified by the ensemble for inclusion in a partially labeled or refined dataset, which is then used to train a downstream machine learning model. )…; and
training a final machine learning one-class classifier in the second stage of training using the third training set representing the signal ([0062]: Describes that the refined, partially labeled training dataset, created through application of ensemble classifiers, to train a final anomaly detection model. The refined dataset 162, generated from ensemble-based scoring and filtering of prior log data is used to train a semi-supervised machine learning model 170.).
Palani further teaches… in response to the coarse machine learning one-class classifier detecting the data point and the noise machine learning one-class classifier not detecting the data point ([0026-0032] teaches training a noise machine learning one-class classifier using a training dataset excluding signal and using the noise classifier to identify and exclude noise data points from further processing. This removes anomalous or noise data from historical data and employs a classifier trained on noise to filter out such data prior to subsequent analysis or model use.)
Regarding “wherein the applying of the ensemble of models comprises adding a data point to the third training set in response to the coarse machine learning one-class classifier detecting the data point and the noise machine learning one-class classifier not detecting the data point”
It would have been obvious to one of ordinary skill in the art at the time of the claimed invention to incorporate Palani’s noise machine learning one-class classifier with Givental, in view of Brown’s, ensemble-based training data refinement. Doing so would have enabled the system to form a cleaner third training set by including only data points flagged as signal form the coarse classifier while excluding data points identified as noise. This would reduce false positives in the refined training data and improved accuracy of the final classifier.
Regarding claim 2, Cancelled
Regarding claim 3, Givental in view of Brown, in view of Palani teaches the method of claim 1. Palani further teaches wherein the final machine learning one-class classifier includes a long short-term memory auto-encoder-decoder ([0026]: “Long Short-Term Memory (LSTM) model with autoencoders”)
Givental further teaches wherein the anomaly is detected using the trained final machine learning one-class classifier([0051-0052]: The unsupervised ensemble first classifies log entries and assigns anomaly scores, the partially labeled dataset is created (hybrid of outputs form the ensemble and analyst feedback); [0060-0063]: Describes that then the semi-supervised model (final model) uses that partially labeled dataset to propagate levels, assign final anomaly scores and detect anomalies.)
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine the teachings of Givental, in view of Brown and Palani in order to implement the final classifier as an LSTM auto-encoder to model sequence patterns and enhance anomaly scoring accuracy. LSTM networks are well-suited for capturing temporal dependencies and, when combined with an autoencoder structure, allow for accurate sequence reconstruction and anomaly detection based on deviations from expected patterns.
Regarding claim 4 Givental in view of Brown, in view of Palani teaches the method of claim 1. Givental further teaches wherein the third training set representing the signal includes information detectable by the coarse machine learning one-class classifier but not detectable by the noise machine learning one-class classifier. ([0052-0055]: Describes generating a dataset by applying an ensemble of classifiers with differing sensitivity to signal and noise data. These classifiers are trained on distinct anomaly scores.; [0056-0058]: Describes using score thresholds and weighted outputs to select or exclude data points.; These descriptions indicate that the generation of the new dataset, certain data would be detectable to the coarse classifier and not by the noise classifier.)
Regarding claim 5, Cancelled
Regarding claim 6, Givental in view of Brown, in view of Palani teaches the method of claim 1, wherein the final machine learning one-class classifier is capable of detecting the signal in information collected using a first operating system different from a second operating system used to collect the second training set excluding the signal (Givental teaches training a final classifier using a refined training set generated from ensemble scoring and anomaly thresholding([0056], and [0062]).; Palani discloses using data from a variety of sources including systems, applications and network devices [0025], implying heterogeneity in the computing environments from which training and evaluation data are obtained).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine the teachings of Givental, in view of Brown and Palani to enable the final classifier to generalize across system environments, such as different operating systems, since doing so would improve the practical applicability and robustness of anomaly detection systems deployed across enterprise or multi-platform networks.
Regarding claim 7, Givental in view of Brown, in view of Palani teaches the method of claim 1. Givental further teaches further comprising: in the first stage of training, training each particular classifier in a plurality of coarse machine learning one-class classifiers using a respective training set in a plurality of training sets, each particular training set in the plurality of training sets including the signal and noise ([0046-0047]: Describes training multiple classifiers using data that includes both anomalous and non-anomalous events.), wherein the plurality of coarse machine learning one-class classifiers includes the coarse machine learning one-class classifier (FIG. 1, and [0048-0051]: Describes an ensemble of unsupervised machine learning models 122-128 that include multiple types of one-class classifiers, all trained using unsupervised machine learning algorithms.) and the plurality of training sets includes the first training set ([0051-0055]: Describes ensemble models trained across multiple datasets including the initial training set.), and wherein the ensemble of models includes the plurality of coarse machine learning one-class classifiers (; FIG. 1, and [0052-0055]: Describes ensemble 120, which includes the plurality of one-class classifiers and scores the different datasets, which includes the coarse classifiers.).
Regarding claim 8, Givental in view of Brown, in view of Palani teaches the method of claim 7. Givental further teaches further comprising applying the ensemble of models to the plurality of training sets to create the third training set representing the signal for the second stage of training (Fig. 1, and [0048-0051]: Describes that these models are trained on distinct datasets and comprise the ensemble 120.; [0046-0047]: Describes that the input features include both anomalous and non-anomalous events derived from system logs.; [0055-0056]: Describes that anomaly scores can be used to create the new training dataset by identifying entries most likely to be signal. ; [0062]: Describes that the new dataset is then used for a further stage of training), wherein applying the ensemble of models to the plurality of training sets includes applying the ensemble of models to the first training set (FIG. 1, and [0052-0062]: Describes that the ensemble is applied directly to log input data to form the new dataset, showing direct application to the original dataset.).
Regarding claim 9, Givental in view of Brown, in view of Palani teaches the method of claim 7. Givental further teaches wherein applying the ensemble of models to the plurality of training sets includes ([0046-0047]: Describes the input data used by the models includes features extracted from logs containing events (signal and noise), and that these are parsed and transformed, implying separate application across datasets.; [0052-0055]: Describes that outputs from each model are scored and combined into a unified anomaly score for input data, implying that each model is applied to the training data independently, supporting the application of each classifier to each training set.): applying each particular classifier in the plurality of coarse machine learning one-class classifiers to each particular training set in the plurality of training sets (FIG. 1, and [0048-0051]: Describes that ensemble 120 includes a plurality of unsupervised machine learning models 122-128 that include multiple types of one-class classifiers, each trained separately and producing individual outputs on input data.); and
Givental does not teach applying the noise machine learning one-class classifier to each particular training set in the plurality of training sets.
Palani teaches applying the noise machine learning one-class classifier to each particular training set in the plurality of training sets ([0032]: Describes removal of anomalous data to form clustered time series training sets, each representing a noise-only dataset; [0037]: Describes training deep learning models on each clustered training set, thereby teaching application of a noise-trained model to each dataset individually.).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine the teachings of Givental, in view of Brown and Palani in order to apply each of the coarse and noise classifiers across a plurality of distinct training sets. Doing so would improve the robustness and granularity of anomaly detection by leveraging multiple perspectives from specialized classifiers training on distinct datasets.
Regarding claims 10, 12, 13, and 15-18, recites substantially the same limitations as claims 1, 3, 4 and 6-9. Claims 10, 12, 13, and 15-18 further recite a system (Givental, [0005], and [0024], and Palani, [0024], and [0025]: Both describe systems comprising at least one processor and memory configured to perform a method.) to perform the method steps of claims 1, 3, 4 and 6-9, respectively, and are therefore rejected on the same premise.
Regarding claim 11, Cancelled
Regarding claim 14, Cancelled
Regarding claim 19, recites substantially the same limitations as claim 1. Claims 19 further the claims recite a non-transitory computer-readable storage medium storing processor-executable instructions to train a neural network (Givental, ([0085-0087]) and Palani, ([0089-0091]): Both describe storing instructions in a non-transitory computer-readable medium configured to cause a processor to execute the claimed method steps.) which perform the method steps of claim 1, respectively, and are therefore rejected on the same premise.
Regarding claim 20, Givental in view of Brown, in view of Palani teaches the method of claim 1. Palani further teaches wherein final machine learning one-class classifier includes an auto-encoder-decoder. ([0026]: Describes that in their ensemble of deep learning models, can comprise of one or more “model with autoencoders” to create a specific model type.)
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine the teachings of Givental, in view of Brown and Palani in order to incorporate such architecture as part of the final classifier would predictably enhance the ability to isolate subtle or complex anomalies during the final stage of detection.
Regarding claim 21, Givental in view of Brown, in view of Palani teaches the method of claim 1. Givental further teaches wherein the noise comprises information not of interest for detecting the anomaly (FIG. 1, and [0046-0047]: Describes that data is input into the system and then formatted by the data cleaning and feature engineering engine 110. This takes out information that cannot be identified later by the models to make a decision if that data is anomalous or not, thus this formatting corresponds to detecting information not of interest for anomaly detection.).
Claim(s) 22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Givental et al. (US 20210281592 A1, referred to as Givental), in view of Brown (US 20190205530 A1, referred to as Brown), in view of Palani et al. (US 20210344695 A1, referred to as Palani), in view of Sharifi et al. (US 20220115009 A1, referred to as Sharifi).
Regarding claim 22, Givental in view of Brown, in view of Palani teaches the method of claim 21. Palani further teaches wherein the second training set includes the noise ([0032]: As disclosed above, when the outliers are).
Although Givental in view of Brown, in view of Palani teaches wherein the second set of training set includes the noise, it does not teach wherein the information not of interest for detecting the anomaly comprises information gathered while a computing device idles.
Sharifi teaches, wherein the information not of interest for detecting the anomaly comprises information gathered while a computing device idles([0071]: Describes that while the system is inactive it can receive data from its attached devices, the examiner notes that computers have been configured to receive information while it is idle as part of their normal operations. The reference of note is merely to reiterate that this is known prior to the filing data of the claimed invention, but is added to show it was common place.)
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine the teachings of Givental in view of Brown, in view of Palani and Sharifi’s inactive status. Doing so would have enabled the system to maintain data acquisition while in a down state, and continuously receive information to process when it becomes active later on.
Double Patenting
The nonstatutory double patenting rejection is based on a judicially created doctrine grounded in public policy (a policy reflected in the statute) so as to prevent the unjustified or improper timewise extension of the “right to exclude” granted by a patent and to prevent possible harassment by multiple assignees. A nonstatutory double patenting rejection is appropriate where the conflicting claims are not identical, but at least one examined application claim is not patentably distinct from the reference claim(s) because the examined application claim is either anticipated by, or would have been obvious over, the reference claim(s). See, e.g., In re Berg, 140 F.3d 1428, 46 USPQ2d 1226 (Fed. Cir. 1998); In re Goodman, 11 F.3d 1046, 29 USPQ2d 2010 (Fed. Cir. 1993); In re Longi, 759 F.2d 887, 225 USPQ 645 (Fed. Cir. 1985); In re Van Ornum, 686 F.2d 937, 214 USPQ 761 (CCPA 1982); In re Vogel, 422 F.2d 438, 164 USPQ 619 (CCPA 1970); In re Thorington, 418 F.2d 528, 163 USPQ 644 (CCPA 1969).
A timely filed terminal disclaimer in compliance with 37 CFR 1.321(c) or 1.321(d) may be used to overcome an actual or provisional rejection based on nonstatutory double patenting provided the reference application or patent either is shown to be commonly owned with the examined application, or claims an invention made as a result of activities undertaken within the scope of a joint research agreement. See MPEP § 717.02 for applications subject to examination under the first inventor to file provisions of the AIA as explained in MPEP § 2159. See MPEP § 2146 et seq. for applications not subject to examination under the first inventor to file provisions of the AIA . A terminal disclaimer must be signed in compliance with 37 CFR 1.321(b).
The filing of a terminal disclaimer by itself is not a complete reply to a nonstatutory double patenting (NSDP) rejection. A complete reply requires that the terminal disclaimer be accompanied by a reply requesting reconsideration of the prior Office action. Even where the NSDP rejection is provisional the reply must be complete. See MPEP § 804, subsection I.B.1. For a reply to a non-final Office action, see 37 CFR 1.111(a). For a reply to final Office action, see 37 CFR 1.113(c). A request for reconsideration while not provided for in 37 CFR 1.113(c) may be filed after final for consideration. See MPEP §§ 706.07(e) and 714.13.
The USPTO Internet website contains terminal disclaimer forms which may be used. Please visit www.uspto.gov/patent/patents-forms. The actual filing date of the application in which the form is filed determines what form (e.g., PTO/SB/25, PTO/SB/26, PTO/AIA /25, or PTO/AIA /26) should be used. A web-based eTerminal Disclaimer may be filled out completely online using web-screens. An eTerminal Disclaimer that meets all requirements is auto-processed and approved immediately upon submission. For more information about eTerminal Disclaimers, refer to www.uspto.gov/patents/apply/applying-online/eterminal-disclaimer.
Claims 1, 3, 10, 12, 19, and 20 are provisionally rejected on the ground of nonstatutory double patenting as being unpatentable over claim 1, 5, 8, 11, 15, 18, and 20 of copending Application No. US 12061692 B2 (reference application). Although the claims at issue are not identical, they are not patentably distinct from each other because The claims of the co-pending application recite the same two stage classifier training architecture, including training a course one class classifier and a benign/noise one class classifier, applying an ensemble of the classifiers to filter events from a first dataset to create third training set, and training a final one class classifier using the third training set. The co pending claims further recite that the first data set includes a systemwide trace and a sequence of events ordered based on process and thread information. Thus, the instant claims merely recite an obvious variation of the same claim training process and do not define a patently distinct invention. .
This is a provisional nonstatutory double patenting rejection because the patentably indistinct claims have not in fact been patented.
Instant Application
US 12,061,692 B2
Claim 1
A computer-implemented method of training a neural network, the method comprising: in a first stage of training: training a coarse machine learning one-class classifier using a first training set including a signal and noise, the first training set comprising a pre- processed system-wide trace including a sequence of events ordered based on process and thread information, wherein the signal comprises information of interest for detecting an anomaly, and wherein the information of interest for detecting the anomaly comprises anomalous data and non-anomalous data; and training a noise machine learning one-class classifier using a second training set excluding the signal; applying an ensemble of models including the noise machine learning one- class classifier and the coarse machine learning one-class classifier to the first training set to create a third training set representing the signal for a second stage of training, wherein the applying of the ensemble of models comprises adding a data point to the third training set in response to the coarse machine learning one-class classifier detecting the data point and the noise machine learning one-class classifier not detecting the data point; and training a final machine learning one-class classifier in the second stage of training using the third training set representing the signal.
Claim 1
A computer-implemented method of fingerprinting a malicious behavior, the method comprising:
in a first stage of training:
training a coarse machine learning one-class classifier to detect a first dataset of events, the first dataset of events including a dataset of events representing a malicious behavior and a dataset of events representing non-malicious behavior; and
training a benign machine learning one-class classifier to detect a second dataset of events, the second dataset of events excluding the dataset of events representing the malicious behaviour;
applying an ensemble of models including the benign machine learning one-class classifier and the coarse machine learning one-class classifier to the first dataset of events to filter out one or more events in the second dataset of events from the first dataset of events to create a third training set representing the malicious behavior for a second stage of training; and
training a final machine learning one-class classifier in the second stage of training using the third training set representing the malicious behavior, the final machine learning one-class classifier representing a fingerprint of the malicious behavior.
Claim 3
The method of claim 1, wherein the final machine learning one- class classifier includes a long short-term memory auto-encoder-decoder, and wherein the anomaly is detected using the trained final machine learning one-class classifier.
Claim 5
The method of claim 1, wherein the first dataset of events includes a system-wide trace including data corresponding to a plurality of non-malicious processes.
Claim 8
The method of claim 1, wherein the first dataset of events includes a sequence of events ordered based on process and threads information.
Claim 10
A system for training a neural network, the system comprising: a processor; a memory storing processor executable instructions that, when executed by the processor, cause the processor to: in a first stage of training: train a coarse machine learning one-class classifier using a first training set including a signal and noise, the first training set comprising a pre- processed system-wide trace including a sequence of events ordered based on process and thread information, wherein the signal comprises information of interest for detecting an anomaly, and wherein the information of interest for detecting the anomaly comprises anomalous data and non-anomalous data; and train a noise machine learning one-class classifier using a second training set excluding the signal; apply an ensemble of models including the noise machine learning one-class classifier and the coarse machine learning one-class classifier to the first training set to create a third training set representing the signal for a second stage of training, wherein the applying of the ensemble of models comprises adding a data point to the third training set in response to the coarse machine learning one-class classifier detecting the data point and the noise machine learning one-class classifier not detecting the data point; and train a final machine learning one-class classifier in the second stage of training using the third training set representing the signal.
Claim 11
. A system for fingerprinting a malicious behavior, the system comprising:
a processor;
a memory storing processor executable instructions that, when executed by the processor, cause the processor to:
in a first stage of training:
train a coarse machine learning one-class classifier to detect a first dataset of events, the first dataset of events including a dataset of events representing a malicious behavior and a dataset of events representing non-malicious behavior; and
train a benign machine learning one-class classifier to detect a second dataset of events, the second dataset of events excluding the dataset of events representing the malicious behaviour;
apply an ensemble of models including the benign machine learning one-class classifier and the coarse machine learning one-class classifier to the first dataset of events to filter out one or more events in the second dataset of events from the first dataset of events to create a third training set representing the malicious behavior for a second stage of training; and
train a final machine learning one-class classifier in the second stage of training using the third training set representing the malicious behavior, the final machine learning one-class classifier representing a fingerprint of the malicious behavior.
Claim 12
The system of claim 10, wherein the final machine learning one-class classifier includes a long short-term memory auto-encoder-decoder, and wherein the anomaly is detected using the trained final machine learning one-class classifier.
Claim 15
The system of claim 11, wherein the first dataset of events includes a system-wide trace including data corresponding to a plurality of non-malicious processes.
Claim 18
The system of claim 11, wherein the first dataset of events includes a sequence of events ordered based on process and threads information.
Claim 19
A non-transitory computer-readable storage medium storing processor-executable instructions to train a neural network, wherein the processor- executable instructions, when executed by a processor, cause the processor to: in a first stage of training: train a coarse machine learning one-class classifier using a first training set including a signal and noise, the first training set comprising a pre- processed system-wide trace including a sequence of events ordered based on process and thread information, wherein the signal comprises information of interest for detecting an anomaly, and wherein the information of interest for detecting the anomaly comprises anomalous data and non-anomalous data; and train a noise machine learning one-class classifier using a second training set excluding the signal; apply an ensemble of models including the noise machine learning one-class classifier and the coarse machine learning one-class classifier to the first training set to create a third training set representing the signal for a second stage of training, wherein the applying of the ensemble of models comprises adding a data point to the third training set in response to the coarse machine learning one-class classifier detecting the data point and the noise machine learning one-class classifier not detecting the data point; and train a final machine learning one-class classifier in the second stage of training using the third training set representing the signal.
Claim 20
A non-transitory computer-readable storage medium storing processor-executable instructions to fingerprint a malicious behavior, wherein the processor-executable instructions, when executed by a processor, are to cause the processor to:
in a first stage of training:
train a coarse machine learning one-class classifier to detect a first dataset of events, the first dataset of events including a dataset of events representing a malicious behavior and a dataset of events representing non-malicious behavior; and
train a benign machine learning one-class classifier to detect a second dataset of events, the second dataset of events excluding the dataset of events representing the malicious behaviour;
apply an ensemble of models including the benign machine learning one-class classifier and the coarse machine learning one-class classifier to the first dataset of events to filter out one or more events in the second dataset of events from the first dataset of events to create a third training set representing the malicious behavior for a second stage of training; and
train a final machine learning one-class classifier in the second stage of training using the third training set representing the malicious behavior, the final machine learning one-class classifier representing a fingerprint of the malicious behavior.
Claim 20
The non-transitory computer-readable storage medium of claim 19, wherein the final machine learning one-class classifier includes an auto-encoder- decoder, and wherein the anomaly is detected using the trained final machine learning one- class classifier.
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 DONALD T RODEN whose telephone number is (571)272-6441. The examiner can normally be reached Mon-Thur 8:00-5:00 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, Omar Fernandez Rivas can be reached at (571) 272-2589. 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.T.R./Examiner, Art Unit 2128
/OMAR F FERNANDEZ RIVAS/Supervisory Patent Examiner, Art Unit 2128
Prosecution Insights