Cisco Technology, Inc.

Karel Bartos

Michal Sofka

Martin Rehak

Vojtech Franc

Lukas Machlica

David A McGrew

Tomas Pevny

Martin Grill

Petr Somol

Ivan Nikolaev

Martin Vejman

Jiri Havelka

Jirí Havelka

Vit Zlamal

Martin Neznal

Gustav Sourek

Filip Zelezny

In one embodiment, a device obtains telemetry data for a plurality of encrypted traffic flows observed in a network. The device clusters the flows into observed flow clusters, based on one or more flow-level features of the obtained telemetry data, as well as malware-related traffic telemetry data into malware-related flow clusters. The observed and malware-related telemetry data are indicative of sequence of packet lengths and times (SPLT) information for the traffic flows. The device samples sets of flows from the observed and malware-related flow clusters, with each set including at least one flow from an observed flow cluster and at least one flow from a malware-related flow cluster. The device trains a deep learning neural network to determine whether a particular encrypted traffic flow is malware-related, by using the SPLT information for the sampled sets of traffic flows as input to an input layer of neurons of the deep network.

Detecting encrypted malware with SPLT-based deep networks

In one embodiment, a security device in a computer network detects potential domain generation algorithm (DGA) searching activity using a domain name service (DNS) model to detect abnormally high DNS requests made by a host attempting to locate a command and control (C&C) server in the computer network. The server device also detects potential DGA communications activity based on applying a hostname-based classifier for DGA domains associated with any server internet protocol (IP) address in a data stream from the host. The security device may then correlate the potential DGA searching activity with the potential DGA communications activity, and identifies DGA performing malware based on the correlating, accordingly.

Autonomous domain generation algorithm (DGA) detector

In one embodiment, a security device in a computer network determines a plurality of values for a plurality of features from samples of known malware, and computes one or more significant values out of the plurality of values, where each of the one or more significant values occurs across greater than a significance threshold of the samples. The security device may then determine feature values for samples of unlabeled traffic, and declares one or more particular samples of unlabeled traffic as synthetic malicious flow samples in response to all feature values for each synthetic malicious flow sample matching a respective one of the significant values for each corresponding respective feature. The security device may then use the samples of known malware and the synthetic malicious flow samples for model-based malware detection.

Refining synthetic malicious samples with unlabeled data

In one embodiment, a device generates one or more time series of characteristics of client-server communications observed in a network for a particular client in the network. The device partitions the one or more time series into sets of time windows based on patterns present in the characteristics of the client-server communications. The device compares the characteristics of the client-server communications from the partitioned time windows to determine measures of behavioral similarity between the compared time windows. The device provides the measures of behavioral similarity between the compared time windows as input to a machine learning-based malware detector. The device causes performance of a mitigation action in the network when the machine learning-based malware detector determines that the particular client in the network is infected with malware.

Using repetitive behavioral patterns to detect malware

In one embodiment, a learning machine device initializes thresholds of a data representation of one or more data features, the thresholds specifying a first number of pre-defined bins (e.g., uniform and equidistant bins). Next, adjacent bins of the pre-defined bins having substantially similar weights may be reciprocally merged, the merging resulting in a second number of refined bins that is less than the first number. Notably, while merging, the device also learns weights of a linear decision rule associated with the one or more data features. Accordingly, a data-driven representation for a data-driven classifier may be established based on the refined bins and learned weights.

Refined learning data representation for classifiers

Identifying malicious communications by generating data representative of network traffic based on adaptive sampling includes, at a computing device having connectivity to a network, obtaining a set of data flows representing network traffic between one or more nodes in the network and one or more domains outside of the network, wherein each data flow in the set of data flows includes a plurality of data packets. One or more features are extracted from the set of data flows based on statistical measurements of the set of data flows. The set of data flows are adaptively sampled based on at least the one or more features. Then, data representative of the network traffic is generated based on the adaptively sampling to identify malicious communication channels in the network traffic.

Identifying malicious communication channels in network traffic by generating data based on adaptive sampling

Identifying malicious network traffic based on distributed, collaborative sampling includes, at a computing device having connectivity to a network, obtaining a first set of data flows, based on sampling criteria, that represents network traffic between one or more nodes in the network and one or more domains outside of the network, each data flow in the first set of data flows including a plurality of data packets. The first set of data flows is forwarded for correlation with a plurality of other sets of data flows from other networks to generate global intelligence data. Adjusted sampling criteria is generated based on the global intelligence data and a second set of data flows is obtained based on the adjusted sampling criteria.

Identifying malicious network traffic based on collaborative sampling

Techniques are presented that identify malware network communications between a computing device and a server based on a cumulative feature vector generated from a group of network traffic records associated with communications between computing devices and servers. Feature vectors are generated, each vector including features extracted from the network traffic records in the group. A self-similarity matrix is computed for each feature which is a representation of the feature that is invariant to an increase or a decrease of feature values across all feature vectors in the group. Each self-similarity matrix is transformed into corresponding histograms to be invariant to a number of network traffic records in the group. The cumulative feature vector is a cumulative representation of the predefined set of features of all network traffic records included in the at least one group of network traffic records and is generated based on the corresponding histograms.

Robust representation of network traffic for detecting malware variations

In one embodiment, a method includes receiving at a security analysis device a plurality of indicators of compromise (IOCs) associated with an entity, sorting at the security analysis device, the IOCs based on a time of occurrence of each of the IOCs, creating a representation of transitions between the IOCs at the security analysis device, and generating at the security analysis device, a feature vector based on the representation of transitions. The feature vector is configured for use by a classifier in identifying malicious entities. An apparatus and logic are also disclosed herein.

Method and apparatus for aggregating indicators of compromise for use in network security

Techniques are presented that identify malware network communications between a computing device and a server utilizing a detector process. Network traffic records are classified as either malware or legitimate network traffic records and divided into groups of classified network traffic records associated with network communications between the computing device and the server for a predetermined period of time. A group of classified network traffic records is labeled as malicious when at least one of the classified network traffic records in the group is malicious and as legitimate when none of the classified network traffic records in the group is malicious to obtain a labeled group of classified network traffic records. A detector process is trained on individual classified network traffic records in the labeled group of classified network traffic records and network communication between the computing device and the server is identified as malware network communication utilizing the detector process.

Learning detector of malicious network traffic from weak labels

In an embodiment, a method, performed by processors of a computing device for creating and storing clusters of incident data records based on behavioral characteristic values in the records and origin characteristic values in the records, the method comprising: receiving a plurality of input incident data records comprising sets of attribute values; identifying two or more first incident data records that have a particular behavioral characteristic value; using a malicious incident behavioral data table that maps sets of behavioral characteristic values to identifiers of malicious acts in the network, and a plurality of comparison operations using the malicious incident behavioral data table and the two or more first incident data records, determining whether any of the two or more first incident data records are malicious; and if so, creating a similarity behavioral cluster record that includes the two or more first incident data records.

Global clustering of incidents based on malware similarity and online trustfulness

A system and a method are disclosed for identifying network threats based on hierarchical classification. The system receives packet flows from a data network and determines flow features for the received packet flows based on data from the packet flows. The system also classifies each packet flow into a flow class based on flow features of the packet flow. Based on a criterion, the system selects packet flows from the received packet flows and places the selected packet flows into an event set that represents an event on the network. The system determines event set features for the event set based on the flow features of the selected packet flows. The system then classifies the event set into a set class based on the determined event set features. Based on the set class, the computer system may report a threat incident on an internetworking device that originated the selected packet flows.

Identifying threats based on hierarchical classification

Techniques are presented to identify malware communication with domain generation algorithm (DGA) generated domains. Sample domain names are obtained and labeled as DGA domains, non-DGA domains or suspicious domains. A classifier is trained in a first stage based on the sample domain names. Sample proxy logs including proxy logs of DGA domains and proxy logs of non-DGA domains are obtained to train the classifier in a second stage based on the plurality of sample domain names and the plurality of sample proxy logs. Live traffic proxy logs are obtained and the classifier is tested by classifying the live traffic proxy logs as DGA proxy logs, and the classifier is forwarded to a second computing device to identify network communication of a third computing device as malware network communication with DGA domains via a network interface unit of the third computing device based on the trained and tested classifier.

Growing community of inventors

Prague, Czechia

Karel Bartos