DETAILED ACTION
This non-final action is responsive to application 18/622,929 as submitted on March 30, 2024.
Claims 1-20 are canceled.
Claims 21-48 are added.
Claim status is currently pending and under examination for claims 21-48 of which independent claims are 21, 34, 42 and 48.
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 Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitation(s) are:
In Claim 39:
“a data-acquisition module configured to…”
In Claim 40:
“a genetic-algorithm module configured to…”
In Claim 46:
“a genetic-algorithm module configured to …”
Because these claim limitation(s) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, they are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
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.
The following are the references relied upon in the rejections below:
Vasudevan (US 20210204152 A1)
Ganesan (US 20230198855 A1)
Claims 21-22, 24, 26-28, 30, 33-35, 38-39, 41-45 and 47-48 are rejected under 35 U.S.C. 103 as being unpatentable over Vasudevan / Ganesan.
Regarding Claim 21, Vasudevan teaches:
A method of classifying dataflows in a data network, the method comprising ([0003] “a communication system is configured to classify data traffic, including data that is encrypted. The classification determined can be used to assign a quality of service class or priority for service in the network.”):
using a network appliance having a hardware processor, for performing ([0111] “The processes and logic flows described in this specification can be performed by one or more programmable processors executing one or more computer programs to perform functions by operating on input data and generating output.”
[0038] “A system of one or more computers can be so configured by virtue of software, firmware, hardware, or a combination of them installed on the system that, in operation, cause the system to perform the actions.”):
in a training phase: acquiring a set of training [labeled] dataflows ([0056] “the client device 140 can collect data for training the machine learning models. The collected data can include traffic flow information, statistics, and estimates of classifications … when available, the client device 140 may collect and provide ground-truth information (e.g., information indicating the type of traffic corresponding to an encrypted data stream).”),
each [labeled] dataflow indicating values of specific characteristics of dataflows in a data network ([0056] “The collected data can include traffic flow information, statistics, and estimates of classifications (e.g., based on server name information, IP address, and so on)”);
obtaining data of a set of recognized classes of said dataflows … ([0056] “the client device 140 may collect and provide ground-truth information (e.g., information indicating the type of traffic corresponding to an encrypted data stream).”);
generating a set of labeled dataflows … based on said recognized classes (The types of traffic of encrypted data streams are ‘recognized classes’ that are used as ground-truth (labels) to train machine learning models.);
using said set of labeled dataflows for initial training of a machine learning (ML) model ([0056] “the client device 140 can collect data for training the machine learning models”
[0077] “The anomaly detector 174 and the ML-based traffic classifier 176 have been previously trained based on examples of traffic patterns similar to each of the data streams 202a-202d, e.g., interactive data, streaming data, bulk upload data, and bulk download data.”);
and initializing a set of classified dataflows as an empty set ([0081] “Traffic labeled as anomalous is also collected by data collection functionality 220, along with associated meta-data such as a relevant IP address, a server name indication, a DNS, a FQDN, etc. This data can be stored in a cache 175”
Traffic labeled as anomalous have their data collected and stored in a cache, therefore a cache configured to store collected traffic acts as an initialized empty set of classified (labeled) traffic.);
in an operation phase, during each epoch of successive epochs: activating said ML model to classify dataflows received from users ([0087] “The anomaly detector 174 is capable of identifying video traffic from different sources. The ability to identify video sources is important because the training data may not contain data from those particular sources, and the knowledge of this information is important for improving traffic classifier performance. As noted above, there is can be a feedback loop to utilize the identified anomalous traffic and improve the anomaly detector 174 and the traffic classifier 176.”
[0076] “six data streams 202a-202f to be processed, each composed of various objects. The data stream 202a represents interactive content, such as web traffic, web application data, or other data for an interactive user interface”
An anomaly detector processes video traffic to label traffic as anomalous and then anomalous traffic is used to initiate model improvement which can be achieved by fine-tuning model parameters (see [0050]). A feedback loop using anomalous traffic to improve models represents epochs of a continuous operational phase (continuous anomaly detection).),
each dataflow indicating respective values of said specific characteristics ([0078] “The traffic classification system includes a robust self-learning mechanism to adapt to new and/or changing data flow characteristics. The anomaly detector 174 has been trained to track the current distributions of known traffic classes. The anomaly detector 174 can automatically flag new traffic distributions, or flag changes in the distributions of existing classes.”);
appending said resulting classified dataflows to said set of classified dataflows ([0081] “Traffic labeled as anomalous is also collected … This data can be stored in a cache 175”);
and retraining said ML model, using said set of labeled dataflows and said set of classified dataflows, for use in a subsequent epoch ([0006] “a portion of the system can be configured to detect anomalous data flows. When traffic is identified that differs from previously established data flow types, the data can be labelled and collected, and then provided to one or more other devices for further training of machine learning models.”
[0106-0107] “Domain adaptation seeks to learn a representation that not only minimizes the training objective but also better generalizes to unseen data from different domains. The model is fine-tuned by showing training data with labels, and domain offset field data without labels. Using either of the above-mentioned training procedures, both ML-based anomaly detector 174 and the ML-based traffic classifier 176 would then be updated”
[0026] “the updated anomaly detector can be one that has been updated, based on data from one or more terminals, to identify one or more additional types of traffic that the anomaly detector before the update was not configured to identify.”
Anomalous traffic can be labelled and collected to update the anomaly detector. To enable domain adaptation for the anomaly detector, the anomaly detector is updated using training data with labels (‘labeled dataflows’) and unlabeled (unseen) data from different domains. Therefore, anomaly detector re-training is performed using labeled anomalous traffic (‘set of classified dataflows’) and labeled training data (from domain adaptation). The updated anomaly detector is then deployed to be used to identify additional types of traffic it could not identify before, therefore retraining the detector for use in a subsequent epoch.);
thereby classifying encrypted dataflows and facilitating enhanced security of the data network ([0004] “In many systems, different data flows are assigned priority levels (e.g., quality of service (QoS) classes), and the limited available communication bandwidth is allocated based on the priority levels”
[Abstract] “use the traffic classifier to predict QoS classes for traffic that the anomaly detector predicts to be similar to the observed traffic patterns”
[0021] “the anomaly detector and the traffic classifier are each configured to generate predictions for encrypted data traffic based on at least one of packet-level information, object-level information, or application layer information”
An anomaly detector can be used to predict if encrypted data flows contain similar traffic patterns as observed traffic patterns with corresponding QoS classes, thereby classifying encrypted data flows. By classifying encrypted data flows, data flows do not have to be decrypted and can remain securely encrypted, thereby facilitating enhanced security of a data network.).
However, Vasudevan does not teach acquiring a set of training unlabeled dataflows, data stating for each recognized class a range of values for each characteristic, reporting classified dataflows to users and pruning classified dataflows, which is taught by Ganesan:
in a training phase: acquiring a set of training unlabeled dataflows, each unlabeled dataflow indicating values of specific characteristics of dataflows in a data network ([0043] “the server 160 can also apply labels to the log data 134 that reflects user data through the terminals 130. The labeled data may indicate, for example, example data sets (e.g., sets of network traffic measures) that have been assigned to a first QoE class, … the server 160 can use clustering techniques or regression analysis to identify the characteristics that are common among the examples for each QoE class … the server 160 can evaluate the sets of network traffic measures in the log data 134 to assign them to the different clusters or groups. … The examples in the log data 134 are then labelled with the QoE class for the cluster to which they have been assigned”
A server performs clustering to identify a cluster that examples of network traffic belong to, therefore examples of network traffic are unlabeled dataflows that are clustered using their characteristics.);
obtaining data of a set of recognized classes of said dataflows stating, for each recognized class, a range of values for each said characteristic ([0043] “the server 160 can use clustering techniques or regression analysis to identify the characteristics that are common among the examples for each QoE class (e.g., the patterns, commonalities, ranges of values, etc. that distinguish examples in one class from the others).”);
generating a set of labeled dataflows from said set of training unlabeled dataflows based on said recognized classes ([0043] “The examples in the log data 134 are then labelled with the QoE class for the cluster to which they have been assigned.”);
using said set of labeled dataflows for … training of a machine learning (ML) model ([0044] “the server 160 updates the ML model 132 based on the log data 134 and 134′ to generate updated ML model 136.”);
reporting resulting classified dataflows to respective users ([0015] “The terminals 130 and the test terminals 130′ are configured at the edges of the network connection and close to the end users, and thus may be categorized as network edge devices.”
[0018] “edge devices may be configured to forward the feature values for model input (e.g., network traffic statistics or other measurements), to a central server, which in turn runs the model to generate predictions on behalf of the edge device then sends the results to the device.”);
appending said resulting classified dataflows to said set of classified dataflows ([0029] “the terminals 130 and 130′ repeatedly generate log data 134 and 134′ about network traffic and provide the log data to the server 160, which uses the log data as training data to generate an updated machine learning model”);
pruning said set of classified dataflows ([0071] “For training newer ML models 132, the system 200 collects input values for the types of input data that the ML models are designed to receive and process. Hence, the system 200 includes data loggers at the CPE/VSAT 130, 130′ to log the input features, … At the terminals 130, 130′, user traffic is filtered and post-processed and converted to structured data from which the input features for the ML models are derived.”);
and retraining said ML model, using … said set of classified dataflows, for use in a subsequent epoch ([0028] “To capture data for further training the models, the system includes a data collection or telemetry aspect, provided through a data logger in the terminals 130 and 130′ that provides data over the network to a data collector in the gateway 110 or a server 160. The server 160 provides an ML pipeline for continual learning and model optimization that is then distributed to the edge devices to improve the accuracy of the models in those devices in an ongoing basis. This provides the system high responsiveness to changing user behavior and the changing traffic generated by applications and websites.”
[0080] “the ML model training can be set up as online learning for repeated continual model updates, …The continual update process can help in keeping up accuracy despite gradual changes in underlying traffic patterns.”);
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the system of Vasudevan with the techniques disclosed by Ganesan to determine labels for unlabeled traffic by using a range of values associated with each label, and to filter received traffic. By determining labels for unlabeled traffic by using a range of values associated with each label, unlabeled traffic characteristics can be compared against ranges of values to assign labels, thereby allowing labeled training data to be created without requiring ground-truths to be known. By filtering received traffic, the quantity of traffic analyzed and pre-processed for model retraining can be reduced, thereby saving storage space and enabling faster model retraining.
Regarding Claim 22, the combined system of Vasudevan / Ganesan teaches:
The method of claim 21, further comprising: during said operational phase: acquiring an additional collection of training unlabeled dataflows (Ganesan [0026] “Edge devices that are deployed and in use by users (e.g., network service customers) can each generate data about the state of the device and the traffic characteristics observed and periodically upload the captured data to a server 160. This provides an ongoing stream of new observations that can be used to train new ML models based on traffic representative of current usage patterns. This type of data, generated by terminals about the network conditions observed, can be unlabeled data from the terminals 130 associated with client devices 140a-140c”);
generating a corresponding collection of labeled dataflows based on said recognized classes (Ganesan [0029] “the terminals 130 and 130′ repeatedly generate log data 134 and 134′ about network traffic and provide the log data to the server 160, which uses the log data as training data to generate an updated machine learning model”
[0043] “the server 160 can use clustering techniques or regression analysis to identify the characteristics that are common among the examples for each QoE class … the server 160 can evaluate the sets of network traffic measures in the log data 134 to assign them to the different clusters or groups. … The examples in the log data 134 are then labelled with the QoE class for the cluster to which they have been assigned”);
and appending said corresponding collection of labeled dataflows to said set of labeled dataflows prior to said retraining ([0044] “In stage (D), the server 160 updates the ML model 132 based on the log data 134 and 134′ to generate updated ML model 136. The server 160 also uses the labels applied, both the ground-truth labels of the network traffic management parameter obtained from the test devices 140d in the log data 135 and the labels generated for the test data 134.”).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined system of Vasudevan / Ganesan with the technique disclosed by Ganesan to generate new training data during model deployment. By generating new training data during model deployment, unlabeled traffic that is observed during deployment can be labeled and then used to retrain a model, thereby allowing the model to learn new traffic patterns and improve its predictions.
Regarding Claims 24, 38 and 45, the combined system of Vasudevan / Ganesan teaches:
The method of claim 21, wherein said successive epochs are defined according to one of: a specified time interval per epoch (Vasudevan [0057] “The terminal 130 also has an anomalous traffic cache 175 that stores information about traffic that the anomaly detector 174 determines to be unusual or different from expected distributions. Anomalous traffic information is stored in the cache 175 over a time window, and can then be used for further training the anomaly detector 174 and the ML-based traffic classifier 176.”
[0087] “there is can be a feedback loop to utilize the identified anomalous traffic and improve the anomaly detector 174 and the traffic classifier 176.”
In a feedback loop, each epoch (loop) is defined as collecting identified anomalous traffic and improving the anomaly detector. Anomalous traffic is cached during a time window, when the time window ends, the cached traffic is then used to retrain the anomaly detector, marking the end of the loop (epoch).);
and reaching a predefined number of classified dataflows per epoch (Vasudevan [0071] “To address irregular traffic, the terminal 130 can cache information about irregular traffic that occurs during a time window. … If the number of anomalous connections in a particular time window exceeds a pre-set threshold, the terminal 130 can save all the traffic and meta-data seen during that time window, including traffic not detected as anomalous. The saved data, whether for all traffic or only anomalous traffic, can be later used to update the training state of the machine learning models in the system, e.g., a real-time machine learning traffic classifier 176, a non-real-time machine learning traffic classifier, and/or the anomaly detector 174.”
In a feedback loop, each epoch (loop) is defined as collecting identified anomalous traffic and improving the anomaly detector (see [0087]). Anomalous traffic is cached during a time window. When the number of anomalous connections (‘classified dataflows’) exceeds a pre-set threshold (‘predefined number’), the cached traffic is then used to retrain the anomaly detector, marking the end of the loop (epoch).).
Regarding Claim 26, the combined system of Vasudevan / Ganesan teaches:
The method of claim 21, wherein the dataflows comprise encrypted dataflows (Vasudevan [0059] “the traffic classification system 170 is configured to classify encrypted traffic.”).
Regarding Claim 27, the combined system of Vasudevan / Ganesan teaches:
The method of claim 21, wherein said generating comprises determining proximity of said values of said specific characteristics, of said each unlabeled dataflow, to values of corresponding characteristics of each recognized class of said set of recognized classes (Ganesan [0023] “perform traffic classification (e.g., to assign packets or streams to appropriate QoS classes for prioritization) and user QoE estimation (e.g., estimating the level quality experienced for streaming video)”
Ganesan [0043] “the server 160 can also apply labels to the log data 134 that reflects user data through the terminals 130. … the server 160 can use clustering techniques or regression analysis to identify the characteristics that are common among the examples for each QoE class (e.g., the patterns, commonalities, ranges of values, etc. that distinguish examples in one class from the others). … Once the clusters or characteristics of each QoE class are determined, the server 160 can evaluate the sets of network traffic measures in the log data 134 to assign them to the different clusters or groups. Each particular example in the log data 134 can be assigned to the cluster that includes examples most similar to the particular example. The examples in the log data 134 are then labelled with the QoE class for the cluster to which they have been assigned. This enables the server 160 to extend the ground-truth labels known for the smaller set of log data 134′ to label the log data 134 with labels that are most likely to be correct, without requiring manual analysis and labeling.”
To label examples of traffic, a server uses clustering to cluster unlabeled examples of traffic based on characteristics of traffic that are common among each QoE class (‘set of recognized classes’). The cluster each unlabeled traffic is assigned to determines the label each traffic is assigned. Therefore, proximity of characteristics of unlabeled traffic to characteristics of each recognized class (QoE classes) is determined to cluster and label examples of unlabeled traffic.).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined system of Vasudevan / Ganesan with the technique disclosed by Ganesan to use clustering to assign labels to unlabeled traffic. By using clustering to assign labels to unlabeled traffic, collected traffic can be labeled without requiring manual analysis and labeling, thereby automating the training data labeling process and increasing the amount of available training data.
Regarding Claims 28 and 39, the combined system of Vasudevan / Ganesan teaches:
The method of claim 21, further comprising extracting metadata from said dataflows and determining said values of specific characteristics from said metadata (Vasudevan [0081] “Traffic labeled as anomalous is also collected by data collection functionality 220, along with associated meta-data such as a relevant IP address, a server name indication, a DNS, a FQDN, etc. This data can be stored in a cache 175 (see FIG. 1) that is periodically flushed by sending the data to the gateway 110 or other system configured to receive and process the collected data. The collected information can be recorded for later analysis and labelling or use as training data for further training the machine learning models 174, 176.”
Vasudevan [0007] “The anomalous traffic, or at least statistics and/or metadata for the traffic, is stored and can used to update the training of both the anomaly detector and the traffic classifier.”).
Regarding Claim 30, the combined system of Vasudevan / Ganesan teaches:
The method of claim 21, wherein said activating and said retraining are implemented concurrently (Vasudevan [0006] “there is a flow of communication between the traffic classifiers and model training system, allowing the classification models to be automatically updated and refined as new traffic patterns are encountered.”
Vasudevan [0062] “The traffic classifier system 170 can be configured to operate efficiently for various connections. For example, the classification technique can output an updated prediction about a connection at multiple times, for example, after looking at each estimated object in the connection. The system can make these predictions in parallel for all current open connections and can updates their QoS class appropriately in real-time.”).
Regarding Claim 33, the combined system of Vasudevan / Ganesan teaches:
The method of claim 21 wherein said operational phase is executed in real time (Vasudevan [0062] “The system can make these predictions in parallel for all current open connections and can updates their QoS class appropriately in real-time.”
Vasudevan [0003] “The communication system can be configured to classify encrypted and unencrypted traffic in real-time”).
Regarding Claim 42, Vasudevan teaches:
A system for classifying encrypted dataflows in a data network, comprising ([0003] “a communication system is configured to classify data traffic, including data that is encrypted. The classification determined can be used to assign a quality of service class or priority for service in the network.):
a training module, having a respective processor ([0057] “Anomalous traffic information is stored in the cache 175 over a time window, and can then be used for further training the anomaly detector 174 and the ML-based traffic classifier 176. For example, the anomalous traffic information can be provided to the gateway 110 or another device to perform the updated training,”
Gateway 110 can update machine learning models, therefore it is implied the gateway has a processor.);
a plurality of operational modules, having respective processors ([0011] “satellite terminals having classification models that operate and are updated as described herein may provide better user experience by more accurately servicing traffic according to the correct QoS class”
[0058] “The terminal 130 can include an outdoor unit (ODU) such as an antenna, and an indoor unit (IDU) performing signal processing and data processing. The traffic classification system 170 can be implemented at the IDU.”);
said training module being configured to: acquire a set of training [labeled] dataflows ([0056] “the client device 140 can collect data for training the machine learning models. The collected data can include traffic flow information, statistics, and estimates of classifications … when available, the client device 140 may collect and provide ground-truth information (e.g., information indicating the type of traffic corresponding to an encrypted data stream).”),
each [labeled] dataflow indicating values of specific characteristics of dataflows ([0056] “The collected data can include traffic flow information, statistics, and estimates of classifications (e.g., based on server name information, IP address, and so on)”);
obtain data of a set of recognized classes of said dataflows … ([0056] “the client device 140 may collect and provide ground-truth information (e.g., information indicating the type of traffic corresponding to an encrypted data stream).”);
generate a set of labeled dataflows … based on said recognized classes (The types of traffic of encrypted data streams are ‘recognized classes’ that are used as ground-truth (labels) to train machine learning models.);
use said set of labeled dataflows for initial training of a machine-learning (ML) model ([0056] “the client device 140 can collect data for training the machine learning models”
[0077] “The anomaly detector 174 and the ML-based traffic classifier 176 have been previously trained based on examples of traffic patterns similar to each of the data streams 202a-202d, e.g., interactive data, streaming data, bulk upload data, and bulk download data.”);
communicate data of said ML model to each operational module of said plurality of operational modules ([0053] “the terminal 130 includes a traffic classification system 170 that can classify traffic into QoS traffic classes.”
[0057] “The traffic classification system 170 includes a history-based classifier 172, a machine-learning-based (ML-based) traffic classifier 176 and an ML-based anomaly detector 174.”
See Figure 6 depicting each VSAT terminal contains an anomaly detector 174.
Each terminal (‘plurality of operational modules’) has an anomaly detector, therefore a trained anomaly detector (‘said ML model’) is communicated to each terminal. When the anomaly detector is communicated to the terminals, the detector’s parameters must also be communicated (therefore communicating data of said ML model).);
and initialize a set of classified dataflows as an empty set ([0081] “Traffic labeled as anomalous is also collected by data collection functionality 220, along with associated meta-data such as a relevant IP address, a server name indication, a DNS, a FQDN, etc. This data can be stored in a cache 175”
Traffic labeled as anomalous have their data collected and stored in a cache, therefore a cache configured to store collected traffic acts as an initialized empty set of classified (labeled) traffic.);
said each operational module being configured to, during each epoch of successive epochs of a continuous operational phase: activate said ML model to classify dataflows received from users ([0087] “The anomaly detector 174 is capable of identifying video traffic from different sources. The ability to identify video sources is important because the training data may not contain data from those particular sources, and the knowledge of this information is important for improving traffic classifier performance. As noted above, there is can be a feedback loop to utilize the identified anomalous traffic and improve the anomaly detector 174 and the traffic classifier 176.”
[0076] “six data streams 202a-202f to be processed, each composed of various objects. The data stream 202a represents interactive content, such as web traffic, web application data, or other data for an interactive user interface”
See Figure 6 depicting each VSAT terminal contains an anomaly detector 174.
An anomaly detector at each terminal processes video traffic to label traffic as anomalous and then anomalous traffic is used to initiate model improvement which can be achieved by fine-tuning model parameters (see [0050]). A feedback loop using anomalous traffic to improve models represents epochs of a continuous operational phase (continuous anomaly detection).),
each dataflow indicating respective values of said specific characteristics ([0078] “The traffic classification system includes a robust self-learning mechanism to adapt to new and/or changing data flow characteristics. The anomaly detector 174 has been trained to track the current distributions of known traffic classes. The anomaly detector 174 can automatically flag new traffic distributions, or flag changes in the distributions of existing classes.”);
and report resulting classified dataflows … to said training module ([0057] “The terminal 130 also has an anomalous traffic cache 175 that stores information about traffic that the anomaly detector 174 determines to be unusual or different from expected distributions. … the anomalous traffic information can be provided to the gateway 110 or another device to perform the updated training”
Traffic information labeled as anomalous is provided to the gateway (‘training module’).);
said training module being further configured to: append said classified dataflows from all operational modules to said set of classified dataflows ([0003] “terminals can collect information about unusual traffic patterns, provide the data to a gateway or other system, and then receive updated traffic classification models that have been refined using the collected information.”
Anomalous traffic and their corresponding data (see [0081]) are stored in a cache, therefore appending classified dataflows (anomalous traffic) from all terminals (‘operational modules’) to said set of classified dataflows (cache) to be provided to the gateway.);
retrain said ML model, using said set of labeled dataflows and said set of classified dataflows, for use in a subsequent epoch ([0006] “a portion of the system can be configured to detect anomalous data flows. When traffic is identified that differs from previously established data flow types, the data can be labelled and collected, and then provided to one or more other devices for further training of machine learning models.”
[0106-0107] “Domain adaptation seeks to learn a representation that not only minimizes the training objective but also better generalizes to unseen data from different domains. The model is fine-tuned by showing training data with labels, and domain offset field data without labels. Using either of the above-mentioned training procedures, both ML-based anomaly detector 174 and the ML-based traffic classifier 176 would then be updated”
[0026] “the updated anomaly detector can be one that has been updated, based on data from one or more terminals, to identify one or more additional types of traffic that the anomaly detector before the update was not configured to identify.”
Anomalous traffic can be labelled and collected to update the anomaly detector. To enable domain adaptation for the anomaly detector, the anomaly detector is updated using training data with labels (‘labeled dataflows’) and unlabeled (unseen) data from different domains. Therefore, anomaly detector re-training is performed using labeled anomalous traffic (‘set of classified dataflows’) and labeled training data (from domain adaptation). The updated anomaly detector is then deployed to be used to identify additional types of traffic it could not identify before, therefore retraining the detector for use in a subsequent epoch.);
and communicate data of said ML model to said each operational module ([0082] “Terminals receive these model updates and begin using the new models (e.g., models 174, 176 having an improved or refined training state, such as the set of internal parameters such as neural network weights).”);
thereby classifying encrypted dataflows and facilitating enhanced security of the data network ([0004] “In many systems, different data flows are assigned priority levels (e.g., quality of service (QoS) classes), and the limited available communication bandwidth is allocated based on the priority levels”
[Abstract] “use the traffic classifier to predict QoS classes for traffic that the anomaly detector predicts to be similar to the observed traffic patterns”
[0021] “the anomaly detector and the traffic classifier are each configured to generate predictions for encrypted data traffic based on at least one of packet-level information, object-level information, or application layer information”
An anomaly detector can be used to predict if encrypted data flows contain similar traffic patterns as observed traffic patterns with corresponding QoS classes, thereby classifying encrypted data flows. By classifying encrypted data flows, data flows do not have to be decrypted and can remain securely encrypted, thereby facilitating enhanced security of a data network.).
However, Vasudevan does not teach acquiring a set of training unlabeled dataflows, data stating for each recognized class a range of values for each characteristic, reporting classified dataflows to users and pruning classified dataflows, which is taught by Ganesan:
acquire a set of training unlabeled dataflows, each unlabeled dataflow indicating values of specific characteristics of dataflows ([0043] “the server 160 can also apply labels to the log data 134 that reflects user data through the terminals 130. The labeled data may indicate, for example, example data sets (e.g., sets of network traffic measures) that have been assigned to a first QoE class, … the server 160 can use clustering techniques or regression analysis to identify the characteristics that are common among the examples for each QoE class … the server 160 can evaluate the sets of network traffic measures in the log data 134 to assign them to the different clusters or groups. … The examples in the log data 134 are then labelled with the QoE class for the cluster to which they have been assigned”
A server performs clustering to identify a cluster that examples of network traffic belong to, therefore examples of network traffic are unlabeled dataflows that are clustered using their characteristics.);
obtain data of a set of recognized classes of said dataflows stating, for each recognized class, a range of values for each said characteristic ([0043] “the server 160 can use clustering techniques or regression analysis to identify the characteristics that are common among the examples for each QoE class (e.g., the patterns, commonalities, ranges of values, etc. that distinguish examples in one class from the others).”);
generate a set of labeled dataflows from said set of training unlabeled dataflows based on said recognized classes ([0043] “The examples in the log data 134 are then labelled with the QoE class for the cluster to which they have been assigned.”);
use said set of labeled dataflows for … training of a machine-learning (ML) model ([0044] “the server 160 updates the ML model 132 based on the log data 134 and 134′ to generate updated ML model 136.”);
and report resulting classified dataflows to respective users … ([0015] “The terminals 130 and the test terminals 130′ are configured at the edges of the network connection and close to the end users, and thus may be categorized as network edge devices.”
[0018] “edge devices may be configured to forward the feature values for model input (e.g., network traffic statistics or other measurements), to a central server, which in turn runs the model to generate predictions on behalf of the edge device then sends the results to the device.”);
append said classified dataflows from all operational modules to said set of classified dataflows ([0029] “the terminals 130 and 130′ repeatedly generate log data 134 and 134′ about network traffic and provide the log data to the server 160, which uses the log data as training data to generate an updated machine learning model”);
prune said set of classified dataflows ([0071] “For training newer ML models 132, the system 200 collects input values for the types of input data that the ML models are designed to receive and process. Hence, the system 200 includes data loggers at the CPE/VSAT 130, 130′ to log the input features, … At the terminals 130, 130′, user traffic is filtered and post-processed and converted to structured data from which the input features for the ML models are derived.”);
retrain said ML model, using ... said set of classified dataflows, for use in a subsequent epoch ([0028] “To capture data for further training the models, the system includes a data collection or telemetry aspect, provided through a data logger in the terminals 130 and 130′ that provides data over the network to a data collector in the gateway 110 or a server 160. The server 160 provides an ML pipeline for continual learning and model optimization that is then distributed to the edge devices to improve the accuracy of the models in those devices in an ongoing basis. This provides the system high responsiveness to changing user behavior and the changing traffic generated by applications and websites.”
[0080] “the ML model training can be set up as online learning for repeated continual model updates, …The continual update process can help in keeping up accuracy despite gradual changes in underlying traffic patterns.”);
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the system of Vasudevan with the techniques disclosed by Ganesan to determine labels for unlabeled traffic by using a range of values associated with each label, and to filter received traffic. By determining labels for unlabeled traffic by using a range of values associated with each label, unlabeled traffic characteristics can be compared against ranges of values to assign labels, thereby allowing labeled training data to be created without requiring ground-truths to be known. By filtering received traffic, the quantity of traffic analyzed and pre-processed for model retraining can be reduced, thereby saving storage space and enabling faster model retraining.
Regarding Claim 34, the rejection of claim 42 is incorporated. The difference in scope being:
A network appliance for classifying dataflows in a data network (Vasudevan [0003] “a communication system is configured to classify data traffic, including data that is encrypted. The classification determined can be used to assign a quality of service class or priority for service in the network.”
[0038] “A system of one or more computers can be so configured by virtue of software, firmware, hardware, or a combination of them installed on the system that, in operation, cause the system to perform the actions.”),
Regarding Claims 35 and 43, the combined system of Vasudevan / Ganesan teaches:
The network appliance of claim 34, wherein said training module is further configured to: acquire an additional collection of training unlabeled dataflows during said operational phase (Ganesan [0026] “Edge devices that are deployed and in use by users (e.g., network service customers) can each generate data about the state of the device and the traffic characteristics observed and periodically upload the captured data to a server 160. This provides an ongoing stream of new observations that can be used to train new ML models based on traffic representative of current usage patterns. This type of data, generated by terminals about the network conditions observed, can be unlabeled data from the terminals 130 associated with client devices 140a-140c”);
generate a corresponding collection of labeled dataflows based on said recognized classes (Ganesan [0029] “the terminals 130 and 130′ repeatedly generate log data 134 and 134′ about network traffic and provide the log data to the server 160, which uses the log data as training data to generate an updated machine learning model”
[0043] “the server 160 can use clustering techniques or regression analysis to identify the characteristics that are common among the examples for each QoE class … the server 160 can evaluate the sets of network traffic measures in the log data 134 to assign them to the different clusters or groups. … The examples in the log data 134 are then labelled with the QoE class for the cluster to which they have been assigned”);
and append said corresponding collection of labeled dataflows to said set of labeled dataflows ([0044] “In stage (D), the server 160 updates the ML model 132 based on the log data 134 and 134′ to generate updated ML model 136. The server 160 also uses the labels applied, both the ground-truth labels of the network traffic management parameter obtained from the test devices 140d in the log data 135 and the labels generated for the test data 134”).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined system of Vasudevan / Ganesan with the technique disclosed by Ganesan to generate new training data during model deployment. By generating new training data during model deployment, unlabeled traffic that is observed during deployment can be labeled and then used to retrain a model, thereby allowing the model to learn new traffic patterns and improve its predictions.
Regarding Claim 41, the combined system of Vasudevan / Ganesan teaches:
The network appliance of claim 34, wherein during each epoch of said operational phase, said operational module is configured to apply said ML model to classify said dataflows received from users while said training module is configured to retrain a duplicate of said ML model for use during a subsequent epoch (Vasudevan [0006] “the classification models to be automatically updated and refined as new traffic patterns are encountered.”
Vasudevan [0062] “The system can make these predictions in parallel for all current open connections and can updates their QoS class appropriately in real-time.”
In a feedback loop, each epoch (loop) is defined as collecting identified anomalous traffic and improving the anomaly detector (see Vasudevan [0087]). Predictions are made in parallel and in real-time. Model retraining of the anomaly detector is performed automatically and in real-time, therefore a terminal (‘operational module’) deploys an anomaly detector (see Vasudevan [0013]) to perform predictions while the anomaly detector is retrained for use during a subsequent loop (using the retrained detector to make updated predictions) by the gateway (‘training module’). See Vasudevan [0057] describing the gateway performs updated training of the anomaly detector.
Ganesan [0004] “the techniques discussed herein allow for frequent training of new models (whether as entirely new models or as updates to existing models) and distribution of the models to edge devices in a satellite-based ISP network.”
Ganesan [0026] “Edge devices that are deployed and in use by users (e.g., network service customers) can each generate data about the state of the device and the traffic characteristics observed and periodically upload the captured data to a server 160. This provides an ongoing stream of new observations that can be used to train new ML models based on traffic representative of current usage patterns”
Ganesan [0028] “transferring the new model through an SDL channel, e.g., through a protocol that manages file transfer and file version control, and which can push updates of files via multicast to multiple devices that each need an updated version of the file”
Incoming new traffic characteristics can be used to retrain a model that is distributed to edge devices. Instead of retraining the distributed model by performing updates on the model, an entirely new model can be generated, therefore retraining a duplicate of the model and enabling version control.).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined system of Vasudevan / Ganesan with the model retraining technique disclosed by Ganesan to generate a new model when performing model retraining. By generating a new model when performing model retraining, a copy of the original model that is being retrained can be saved as an original version and new models can be saved as newer versions, thereby enabling version control and allowing edge devices to use different model versions.
Regarding Claim 44, the combined system of Vasudevan / Ganesan teaches:
The system of claim 42, further configured to: connect said plurality of operational modules to said training module in a star configuration through the data network (Vasudevan [0082] “The gateway 110 or other system can include an analysis module 230 configured to analyze and label data. … The labeled data is provided to a model training module 240 that manages further training of the models 174, 176 and then periodically provides the updated models back to terminals”
Vasudevan discloses Figure 6 (reproduced below) depicting a communication system (‘data network’) comprised of VSAT terminals (‘operational modules’) connected to a gateway (‘training module’) comprised of analysis module 230 and model training module 240. The terminals send data to the gateway (indicated by solid arrows) and the gateway returns model updates (indicated by dotted line arrows). Therefore, the terminals are connected to the gateway in a star configuration.
PNG
media_image1.png
732
1021
media_image1.png
Greyscale
);
and connect said each operational module to respective users through the data network ([0052] “a client device 140 obtains network service through a connection with a satellite terminal 130. … The client device 140 can be any appropriate device that can transfer data over a network, for example, a phone, a laptop computer, a desktop computer, a wearable device”).
Regarding Claim 47, the combined system of Vasudevan / Ganesan teaches:
The system of claim 42, wherein during each epoch of said operational phase, said each operational module is configured to apply said ML model to classify said dataflows received from users while said training module is configured to retrain a duplicate of said ML model for use during a subsequent epoch (Vasudevan [0006] “the classification models to be automatically updated and refined as new traffic patterns are encountered.”
Vasudevan [0062] “The system can make these predictions in parallel for all current open connections and can updates their QoS class appropriately in real-time.”
See Figure 6 depicting each VSAT terminal contains an anomaly detector 174.
In a feedback loop, each epoch (loop) is defined as collecting identified anomalous traffic and improving the anomaly detector (see Vasudevan [0087]). Predictions are made in parallel and in real-time. Model retraining of the anomaly detector is performed automatically and in real-time, therefore each terminal (‘operational modules’) deploy an anomaly detector (see Vasudevan [0013]) to perform predictions while the anomaly detector is retrained for use during a subsequent loop (using the retrained detector to make updated predictions) by the gateway (‘training module’). See Vasudevan [0057] describing the gateway performs updated training of the anomaly detector.
Ganesan [0004] “the techniques discussed herein allow for frequent training of new models (whether as entirely new models or as updates to existing models) and distribution of the models to edge devices in a satellite-based ISP network.”
Ganesan [0026] “Edge devices that are deployed and in use by users (e.g., network service customers) can each generate data about the state of the device and the traffic characteristics observed and periodically upload the captured data to a server 160. This provides an ongoing stream of new observations that can be used to train new ML models based on traffic representative of current usage patterns”
Ganesan [0028] “transferring the new model through an SDL channel, e.g., through a protocol that manages file transfer and file version control, and which can push updates of files via multicast to multiple devices that each need an updated version of the file”
Incoming new traffic characteristics can be used to retrain a model that is distributed to edge devices. Instead of retraining the distributed model by performing updates on the model, an entirely new model can be generated, therefore retraining a duplicate of the model and enabling version control.).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined system of Vasudevan / Ganesan with the model retraining technique disclosed by Ganesan to generate a new model when performing model retraining. By generating a new model when performing model retraining, a copy of the original model that is being retrained can be saved as an original version and new models can be saved as newer versions, thereby enabling version control and allowing edge devices to use different model versions.
Regarding Claim 48, the rejection of Claim 42 is incorporated. The difference in scope being:
A communication network, comprising (Vasudevan [0052] “The example of FIG. 1 shows a satellite communication system … a client device 140 obtains network service through a connection with a satellite terminal 130. The terminal 130 communicates wirelessly with a satellite 120, which in turn communicates wirelessly with a gateway 110.”):
a system for classifying dataflows in the communication network, the system comprising (Vasudevan [0053] “In the system 100, the terminal 130 includes a traffic classification system 170 that can classify traffic into QoS traffic classes.”
See Figure 1 depicting a communication system 100 that includes a terminal for classifying traffic.).
The following are the references relied upon in the rejections below:
Brabec (US 20190102337 A1)
Claims 23 and 37 are rejected under 35 U.S.C. 103 as being unpatentable over Vasudevan / Ganesan / Brabec.
Regarding Claims 23 and 37, the combined system of Vasudevan / Ganesan teaches: The method of claim 21, however, the combination does not teach random sampling of dataflows, which is taught by Brabec:
wherein said pruning comprises random sampling of dataflows of said set of classified dataflows under a constraint of one of ([0052] “Operationally, the techniques herein address the situation in which there is more training data available to train a random decision forest than can be feasibly used for training. One potential approach to this situation would be to randomly select a subset of the training dataset and only use this subset for training. The remaining portion of the full training dataset, which is typically orders of magnitude larger than the subset, can then be ignored.”
[0033] “the training data may include sample telemetry data that is “normal,” or “malware-generated.””
[0031] “traffic analysis process 248 may assess captured telemetry data regarding one or more traffic flows, to determine whether a given traffic flow or set of flows are caused by malware in the network”
Training data is comprised of samples of captured telemetry data regarding traffic flows labeled as “normal” or “malware-generated”, therefore the training data is a set of classified dataflows. When a subset of training data is randomly selected, the rest of the unselected training data is ignored, therefore ‘pruning’ the training data based on random sampling.):
a permissible number of dataflows of said set of classified dataflows ([0058] “First, let D represent the whole training dataset that is available. Let us define following notations used in the rest of this text. A randomly sampled subset of training dataset D is denoted as
S
f
, where f represents the fraction of data from D that is in
S
f
. For example, the dataset
S
0.2
, includes 20% of the samples randomly sampled from D.”
When a subset of training data is randomly sampled, only 20 percent of the total number of samples are selected from the training dataset, therefore pruning and random sampling is constrained under a permissible number of dataflows (20 percent).);
and a specified number of dataflows per epoch ([0063] “Various stopping criteria are possible, to determine when to stop the stochastic iterative retraining of the model. … the stopping criterion may be a predefined, maximum number of iterations, so that all of the data in the training dataset has the chance to be selected. For example, if 20% of D is chosen as the training subset during any iteration (e.g., f=0.2), the maximum number of iterations may be chosen as 5.”).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined system of Vasudevan / Ganesan with the sampling technique disclosed by Brabec to randomly sample training data samples. By randomly sampling training data samples, a subset of samples can be chosen from a large dataset, thereby giving all samples of the large dataset a chance to be chosen and reducing model training time since less samples need to be learned.
The following are the references relied upon in the rejections below:
Daniels (US 11010640 B1)
Claims 25 and 36 are rejected under 35 U.S.C. 103 as being unpatentable over Vasudevan / Ganesan / Daniels.
Regarding Claims 25 and 36, the combined system of Vasudevan / Ganesan teaches:
The method of claim 22, further comprising … said set of labeled dataflows (Vasudevan [0056] “the client device 140 can collect data for training the machine learning models. The collected data can include traffic flow information, statistics, and estimates of classifications … when available, the client device 140 may collect and provide ground-truth information (e.g., information indicating the type of traffic corresponding to an encrypted data stream).”).
However, the combination does not teach reducing content of labeled data based on dates, which is taught by Daniels:
reducing content of said set of labeled [data] based on dates of generating individual labeled [data] ((Col. 4, Lines 22-24) “the filtered set of training data comprises the set of training data filtered based at least in part on whether date of training data is within a date range.”
(Col. 12, Lines 20-26) “event data is filtered for date range. For example, a set of events is selected for being recent (e.g., within the last year, month, quarter, day, etc.), within a range of dates, within a time window during the date (e.g., during daylight, during night time, during first shift, during swing shift, etc.), during holidays, during weekdays, during weekends”
(Abstract) “and determine an updated set of training data by relabeling the set of potential mislabeled data”).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined system of Vasudevan / Ganesan with the filtering technique disclosed by Daniels to filter a training dataset based on training data date. By filtering a training dataset based on training data date, a model can be trained exclusively on recent samples, thereby preventing the model from learning outdated patterns and increasing prediction accuracy on new data.
The following are the references relied upon in the rejections below:
Gay (US 20180013776 A1)
Claims 29, 40 and 46 are rejected under 35 U.S.C. 103 as being unpatentable over Vasudevan / Ganesan / Gay.
Regarding Claims 29, 40 and 46, the combined system of Vasudevan / Ganesan teaches: The method of claim 21, however, the combination does not teach using a genetic algorithm for selecting characteristics, which is taught by Gay:
further comprising using a genetic algorithm for selecting said specific characteristics from a superset of characteristics ([0051] “Classification approaches try to extract features of network flows and traffic that are characteristic of normal traffic or malicious traffic, constructing from these features a classifier that is able to differentiate between the two classes (normal and malicious).”
[0069] “leverage genetic programming to select the right features for a particular situation/class of anomalies. The “genome” of each individual may describe the enabled features, evaluation can be performed by running the anomaly detection process on a set of traffic traces”
[0077] “the system may use a genetic programming approach, as shown in FIG. 5D, to select the proper input features that will specialize the anomaly detectors towards a particular type of anomaly”
[0084] “the genome of each individual is a set of Boolean variables specifying whether a feature of the anomaly detector is activated. In other words, the system may evaluate the effects of using different input features to the anomaly detector”
To perform anomaly classification of traffic, a classifier is constructed by using extracted features of traffic. To select the right features for each class (normal and malicious), genetic programming is used. A genome represents the ‘superset of characteristics’ that are mutated to determine (select) a subset of proper features.).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined system of Vasudevan / Ganesan with the feature selection technique disclosed by Gay to use genetic programming to select an optimal subset of features for model training. By using genetic programming to select an optimal subset of features, noisy and irrelevant features can be excluded during model training, thereby simplifying the final model and speeding up model training.
The following are the references relied upon in the rejections below:
Sanders (US 20180013639 A1)
Claim 31 is rejected under 35 U.S.C. 103 as being unpatentable over Vasudevan / Ganesan / Gay / Sanders.
Regarding Claim 31, the combined system of Vasudevan / Ganesan / Gay teaches: The method of claim 29, however the combination does not teach bidirectional dataflows, which is taught by Sanders:
wherein said superset of characteristics pertain to bidirectional dataflows comprising, per dataflow, at least two of ([0055] “Access to the TCP/IP traffic traces allows extraction of many bidirectional traffic features”
[0057] “The success of classification models relies on the selection of informative, uncorrelated, and robust features”):
minimum, mean, maximum, and standard deviation of packet sizes;
minimum, mean, maximum, and standard deviation of packet inter-arrival time;
total number of packets ([0085] “The number of bidirectional packets is a feature that roughly approximates the amount of data transferred to render a web page. As expected, this feature can be used to identify mobile and video pages.”);
total data size in bytes ([0084-0085] “Total Number of Bytes Transferred: The number of bidirectional packets is a feature that roughly approximates the amount of data transferred to render a web page. As expected, this feature can be used to identify mobile and video pages. The minimum number of bytes transmitted by the client per TCP connection is informative for classifying mobile web pages—this makes sense because mobile sites are designed to be more efficient than traditional web pages.);
total number of payload packets; entropy of packet sizes and entropy of inter-packet arrival time; probability of state transition between consecutive packets with same and different packet sizes; probability of state transition between consecutive packets with same and different inter- packet arrival time; and probability of state transition for packet direction between consecutive packets.
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined system of Vasudevan / Ganesan / Gay with the technique disclosed by Sanders to use total number of bidirectional packets and total number of bytes transferred to classify traffic. By using total number of bidirectional packets and total number of bytes transferred to classify traffic, traffic can be classified without having to inspect the payload contents of the packets, thereby reducing computational overhead and avoiding traffic decryption.
The following are the references relied upon in the rejections below:
Xue (CN 102984131 A)
Claim 32 is rejected under 35 U.S.C. 103 as being unpatentable over Vasudevan / Ganesan / Xue.
Regarding Claim 32, the combined system of Vasudevan / Ganesan teaches: The method of claim 21, however the combination does not teach performing a two-step classification and binary classification of payload and non-payload dataflows, which is taught by Xue:
wherein said ML model is configured to perform a two-step classification: binary classification of payload dataflows versus non-payload dataflows ([0031] “using the two-dimensional normal distribution model can not only the encrypted data packet is protocol for identifying different applications can also be used for the data packet transmission protocol identification.”
[0047] “in step 101 can train to obtain a plurality of two-dimensional normal distribution model calculation (i.e., multiple sets of two-dimensional normal distribution model parameters), so that it can be adjusted according to a plurality of two-dimensional normal distribution model to obtain a plurality of classification features in this step.”
[0035] “when a transport layer protocol is Transmission Control Protocol (Transmission Control Protocol (TCP) three-way handshake to be eliminating data packet in TCP protocol, generally, the data packet will not be encrypted or other processes, does not carry information quantity. at this time, this step comprises: judging whether the training data set of data packet is TCP three-way handshake data packet; if the judging result is yes, then eliminating the three times handshake data packet of the TCP protocol of the training data set, the training data after eliminating the functions of calculating the two-dimensional normal distribution model parameters.”
A two-dimensional normal distribution model is used to eliminate data packets that are TCP three-way handshakes (data packet transmission protocol identification). The model can also be used for identifying different applications for encrypted data packets, therefore the model performs a two-step classification with a binary classification of determining whether data packets are three-way handshakes (non-payload dataflows) or non-handshake data packets (payload dataflows).);
and classification of the payload dataflows ([0031] “using the two-dimensional normal distribution model can not only the encrypted data packet is protocol for identifying different applications can also be used for the data packet transmission protocol identification.”
[0034] “the different applications including secure socket layer hypertext transfer protocol (Hypertext Transfer Protocol over Secure Socket Layer, HTTPS). opening a virtual private network Open Virtual Private Network (OpenVPN) and onion routing (THE Onion Router (TOR).”).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined system of Vasudevan / Ganesan with the technique disclosed by Xue to identify non-payload data packets. By identifying non-payload data packets, handshake data packets that do not contain payload information can be eliminated so that only payload data packets can be classified, thereby reducing the amount of data packets needed to be processed and classified.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Wang et al. (CN 114358177 A) teaches updating a flow characteristic extraction model and flow separation model using samples with known classes and unknown classes. Labels for the samples with unknown classes are found by performing K-means clustering. The updating of the models is repeated to continuously classify unknown encryption flows.
Pastor Perales et al. (US 20190294995 A1) teaches continuously training and validating machine learning algorithms that receive real traffic by generating synthetic data and labeling real traffic. Algorithms are iteratively trained by using different subsets of labeled training data.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PEDRO J MORALES whose telephone number is (571)272-6106. The examiner can normally be reached 8:30 AM - 6:00 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, MIRANDA M HUANG can be reached at (571)270-7092. 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.
/PEDRO J MORALES/Examiner, Art Unit 2124
/MIRANDA M HUANG/Supervisory Patent Examiner, Art Unit 2124