Deep learning pipeline for detecting malicious commands and control traffic
Through the deep learning model, the feature identification of the request header of the network traffic session is solved, and the problem of difficult to identify malware in real-time commands and control traffic in the existing technology is solved, efficient detection and identification of malicious traffic is achieved, and network security protection capabilities are improved. .
Patent Information
- Application Number
- CN202380028345.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-01-18
- Filing Date
- 2023-01-17
- Publication Date
- 2024-11-05
AI Technical Summary
The prior art is difficult to effectively identify and prevent malware's real-time commands and control traffic, especially when malware authors constantly evade detection technology.
Using a deep learning model, automatic feature recognition is performed based on the request header of the network traffic session. By extracting character word elements and word word elements, and feeding them to the embedding layer to find feature groups, combining a bidirectional long and short-term memory network and a fully connected layer. Predict tags for HTTP sessions to identify malicious traffic.
It realizes efficient detection of real-time malicious commands and control traffic, improves the identification ability of new malware variants, reduces false positive and false negative errors, and enhances network security protection.
Smart Images

Figure CN118901223A_ABST
Abstract
Description
Background Art
[0001] Malware is a general term that is often used to refer to malicious software (e.g., including a variety of hostile, intrusive and / or otherwise unwanted software). Example uses of malware include disrupting computer and / or computer network operations, stealing proprietary information (e.g., confidential information such as identity, financial and / or intellectual property related information), and / or gaining access to private / proprietary computer systems and / or computer networks. Malware can be in the form of code, scripts, active content and / or other software. Unfortunately, as technology has developed to help detect and mitigate malware, unscrupulous authors have found ways to circumvent such efforts. Therefore, there is a need to continuously improve the technology used to identify and mitigate malware. BRIEF DESCRIPTION OF THE DRAWINGS
[0002] Various embodiments of the invention are disclosed in the following detailed description and accompanying drawings.
[0003] Figure 1 An example of an environment in which malicious applications are detected and prevented from causing harm is illustrated.
[0004] Figure 2A An embodiment of a data appliance is illustrated.
[0005] Figure 2B is a functional diagram of the logical components of an embodiment of a data appliance.
[0006] Figure 3 Examples of logical components that may be included in a system for analyzing a sample are illustrated.
[0007] Figure 4A An embodiment of a process for performing detection of command and control malware is illustrated.
[0008] Figure 4B An embodiment of a process for performing automatic feature recognition is illustrated.
[0009] Figure 4C An embodiment of a process for feeding character-grams and word-grams into an embedding layer of a deep learning model to find feature groups is illustrated.
[0010] Figure 5 An example of a request header is illustrated.
[0011] Figure 6 An example of an embedding layer is illustrated.
[0012] Figure 7 Figure 2 shows an example of a one-dimensional convolution.
[0013] Figure 8 Illustration of an example model architecture with convolutions.
[0014] Fig. 9 Illustration of an example of bidirectional long short-term memory with attention.
[0015] Fig.10 Illustration of an example of a fully connected layer in a deep learning network.
[0016] Fig.11 An example of a cost function is illustrated.
[0017] Fig.12 An example of the results of model training is shown in the figure. DETAILED DESCRIPTION
[0018] The present invention may be implemented in a variety of ways, including as a process; an apparatus; a system; a component of matter; a computer program product embodied on a computer-readable storage medium; and / or a processor (such as a processor configured to execute instructions stored on and / or provided by a memory coupled to the processor). In this specification, these implementations or any other form that the present invention may take may be referred to as techniques. In general, the order of the steps of the disclosed processes may be changed within the scope of the present invention. Unless otherwise stated, a component described as being configured to perform a task (such as a processor or memory) may be implemented as a general component that is temporarily configured to perform the task at a given time or as a specific component manufactured to perform the task. As used herein, the term "processor" refers to one or more devices, circuits, and / or processing cores that are configured to process data (such as computer program instructions).
[0019] A detailed description of one or more embodiments of the present invention is provided below together with the accompanying drawings that illustrate the principles of the present invention. The present invention is described in conjunction with such embodiments, but the present invention is not limited to any embodiment. The scope of the present invention is limited only by the claims, and the present invention includes multiple alternatives, modifications and equivalents. In order to provide a comprehensive understanding of the present invention, many specific details are set forth in the following description. These details are provided for illustrative purposes, and the present invention can be implemented according to the claims without some or all of these specific details. For the sake of clarity, technical materials known in the technical field related to the present invention are not described in detail to avoid unnecessarily obscuring the present invention.
[0020] In some embodiments, a system / method / computer program product for performing detection of command and control malware includes: monitoring network traffic sessions; and using a deep learning model to perform automatic feature recognition for real-time malicious command and control traffic detection based on request headers of the monitored network traffic sessions.
[0021] In some embodiments, performing automatic feature recognition for real-time malicious command and control traffic detection includes extracting a request header from a network traffic session; performing tokenization of the request header to generate character tokens and word tokens; and feeding the character tokens and word tokens into an embedding layer of a deep learning model to find feature groups.
[0022] In some embodiments, feeding character tokens and word tokens into an embedding layer of a deep learning model to find feature groups includes multiplying the embedding layer with a filter matrix to obtain a convolutional layer; and max-pooling multiple elements of the convolutional layer to obtain the feature groups.
[0023] In some embodiments, multiplying the embedding layer with the filter matrix to obtain a convolutional layer includes multiplying the embedding layer with the filter matrix at a first position to obtain a first element of the convolutional layer; moving the filter matrix down one row along the embedding layer to obtain a second position of the filter matrix; and multiplying the embedding layer with the filter matrix at the second position to obtain a second element of the convolutional layer.
[0024] In some embodiments, performing automatic feature recognition for real-time malicious command and control traffic detection includes finding contextual relationships between features of a feature group using a bidirectional long short-term memory (Bi-LSTM) after performing max pooling of multiple elements.
[0025] In some embodiments, the system / method / computer program product further comprises predicting a tag of a Hypertext Transfer Protocol (HTTP) session based on the probability score using a fully connected layer of the deep learning model.
[0026] In some embodiments, the deep learning model corresponds to a neural network.
[0027] I. Overview
[0028] A firewall generally protects a network from unauthorized access while allowing authorized communications to pass through the firewall. A firewall is generally a device, a collection of devices, or software executed on a device that provides firewall functionality for network access. For example, a firewall can be integrated into an operating system of a device (e.g., a computer, a smart phone, or other type of device that supports network communications). A firewall can also be integrated into one or more software on various types of devices or executed as one or more software applications on various types of devices, such as computer servers, gateways, network / routing devices (e.g., network routers), and data appliances (e.g., security appliances or other types of specialized devices), and in various implementations, certain operations can be implemented in dedicated hardware (such as ASICs or FPGAs).
[0029] Firewalls typically deny or allow network transmissions based on a set of rules. These sets of rules are often referred to as policies (e.g., network policies or network security policies). For example, a firewall can filter inbound traffic by applying a set of rules or policies to prevent unwanted external traffic from reaching the protected device. A firewall can also filter outbound traffic by applying a set of rules or policies (e.g., allowing, blocking, monitoring, notifying, or logging, and / or other actions that can be specified in a firewall rule or firewall policy, which can be triggered based on a variety of criteria such as those described herein). A firewall can also filter local network (e.g., intranet) traffic by similarly applying a set of rules or policies.
[0030] Security devices (e.g., security appliances, security gateways, security services, and / or other security devices) may include various security functions (e.g., firewalls, anti-malware, intrusion prevention / detection, data loss prevention (DLP), and / or other security functions), network functions (e.g., routing, quality of service (QoS), workload balancing of network-related resources, and / or other network functions), and / or other functions. For example, routing functions may be based on source information (e.g., IP address and port), destination information (e.g., IP address and port), and protocol information.
[0031] Basic packet filtering firewalls filter network communication traffic by inspecting individual packets transmitted on the network (e.g., packet filtering firewalls or first generation firewalls, which are stateless packet filtering firewalls). Stateless packet filtering firewalls typically inspect the individual packets themselves and apply rules based on the inspected packets (e.g., using a combination of the packet's source and destination address information, protocol information, and port number).
[0032] Application firewalls can also perform application layer filtering (e.g., application layer filtering firewalls or second generation firewalls, which work at the application layer of the TCP / IP stack). Application layer filtering firewalls or application firewalls can generally identify certain applications and protocols (e.g., web browsing using Hypertext Transfer Protocol (HTTP), Domain Name System (DNS) requests, file transfers using File Transfer Protocol (FTP), and various other types of applications and other protocols, such as Telnet, DHCP, TCP, UDP, and TFTP (GSS)). For example, an application firewall can block unauthorized protocols that attempt to communicate over standard ports (e.g., protocols that attempt to sneak through using non-standard ports for unauthorized / non-compliant protocols can generally be identified using an application firewall).
[0033] A stateful firewall can also perform state-based packet inspection, in which each packet is inspected within the context of a series of packets associated with the packet stream transmitted by the network. This firewall technology is often called stateful packet inspection because it maintains a record of all connections passing through the firewall and can determine whether a packet is the start of a new connection, part of an existing connection, or an invalid packet. For example, the state of the connection itself can be one of the criteria for triggering a rule within a policy.
[0034] As discussed above, advanced or next-generation firewalls can perform stateless and stateful packet filtering as well as application layer filtering. Next-generation firewalls can also perform additional firewall technologies. For example, some newer firewalls, sometimes referred to as advanced or next-generation firewalls, can also identify users and content (e.g., next-generation firewalls). In particular, some next-generation firewalls are expanding the list of applications that these firewalls can automatically identify to thousands of applications. Examples of such next-generation firewalls are commercially available from Palo Alto Networks (e.g., Palo Alto Networks' PA Series firewalls). For example, Palo Alto Networks' next-generation firewalls enable enterprises to use a variety of identification technologies to identify and control applications, users, and content, not just ports, IP addresses, and packets, such as the following technologies: APP-ID for accurately identifying applications, User-ID for user identification (e.g., by user or user group), and Content-ID for real-time content scanning (e.g., controlling web surfing and limiting data and file transfers). These identification technologies allow enterprises to securely implement application usage using business-related concepts rather than following the traditional methods provided by traditional port blocking firewalls. Furthermore, specialized hardware for next-generation firewalls (e.g., implemented as dedicated appliances) typically provides higher levels of application inspection performance than software executed on general-purpose hardware (e.g., such as security appliances offered by Palo Alto Networks, which use dedicated, function-specific processing that is tightly integrated with a single-pass software engine to maximize network throughput while minimizing latency).
[0035] Advanced or next-generation firewalls may also be implemented using virtualized firewalls. Examples of such next-generation firewalls are commercially available from Palo Alto Networks (e.g., Palo Alto Networks' VM series firewalls, which support a variety of commercial virtualization environments, including, for example, ESXi TM and NSX TM , NetScaler SDXTM , KVM / OpenStack (Centos / RHEL, ) and Amazon Web Services (AWS). For example, virtualized firewalls can support similar or identical advanced threat prevention features available in next-generation firewalls and physical form factor appliances, allowing enterprises to securely enable applications to flow into and through their private, public, and hybrid cloud computing environments. Automation features such as VM monitoring, dynamic address groups, and REST-based APIs allow enterprises to proactively monitor VM changes, dynamically feed that context into security policies, eliminating policy lags that can occur when VMs change.
[0036] II. Sample Environment
[0037] Figure 1 An example of an environment in which malicious applications ("malware") are detected and prevented from causing harm is illustrated. As will be described in more detail below, malware classification (e.g., determined by security platform 122) may be used in Figure 1 The environment shown in FIG. 1 is different from the environment shown in FIG. 1 and the ...
[0038] The term "application" is used throughout this specification to refer to programs, packages, manifests, packaged software, etc., regardless of form / platform. An "application" (also referred to as a "sample" herein) can be a separate file (e.g., a calculator application with a file name "calculator.apk" or "calculator.exe"), and can also be an independent component of another application (e.g., a mobile advertising SDK or library embedded in a calculator application).
[0039] As used herein, "malware" refers to an application that participates in behaviors that the user does not approve / will not approve if fully informed, whether or not these behaviors are secret (and whether or not illegal). Examples of malware include Trojans, viruses, root toolkits, spyware, hacker tools, keyloggers, etc. An example of malware is a desktop application that collects and reports the location of the terminal user to a remote server (but does not provide location-based services, such as map services, to the user). Another example of malware is a malicious Android application package .apk (APK) file, which appears to be a free game to the terminal user, but secretly sends SMS paid messages (e.g., each costs $10), thereby increasing the terminal user's phone bill. Another example of malware is an Apple iOS flashlight application that secretly collects the user's contacts and sends these contacts to spammers. Other forms of malware (such as ransomware) can also be detected / thwarted using the technology described herein. Furthermore, while malware signatures are described herein as being generated for malicious applications, the techniques described herein may also be used in various embodiments to generate profiles for other kinds of applications (eg, adware profiles, goodware profiles, etc.).
[0040] The techniques described herein can be used in conjunction with various platforms (e.g., desktop computers, mobile devices, gaming platforms, embedded systems, etc.) and / or various types of applications (e.g., Android .apk files, iOS applications, Windows PE files, Adobe Acrobat PDF files, etc.). Figure 1 In the example environment shown in , client devices 104-108 (respectively) are a laptop computer, a desktop computer, and a tablet computer that reside within an enterprise network 140. Client device 110 is a laptop computer that resides outside of enterprise network 140.
[0041] Data appliance 102 is configured to implement policies regarding communications between client devices (such as client devices 104 and 106) and nodes external to enterprise network 140 (e.g., reachable via external network 118). Examples of such policies include policies that control traffic shaping, quality of service, and traffic routing. Other examples of policies include security policies, such as security policies that require scanning of incoming (and / or outgoing) email attachments, website content, files exchanged via instant messaging programs, and / or other file transfers for threats. In some embodiments, data appliance 102 is also configured to implement policies regarding traffic that resides within enterprise network 140.
[0042] Despite Figure 1Although illustrated as a single element in FIG. 1 , enterprise network 140 may include multiple networks, any / each of which may include one or more data appliances or other components embodying the techniques described herein. For example, the techniques described herein may be deployed by a large multinational corporation (or other entity) with multiple offices in multiple geographic locations. Furthermore, although client devices 104-108 may be deployed in Figure 1 102 , it is understood that one or more intermediate nodes (eg, routers, switches, and / or proxies) may be, and typically are, inserted between the various elements in the enterprise network 140 .
[0043] Figure 2A 102. An embodiment of a data appliance is shown in FIG. The example shown is a representation of the physical components included in the data appliance 102 in various embodiments. Specifically, the data appliance 102 includes a high-performance multi-core central processing unit (CPU) 202 and a random access memory (RAM) 204. The data appliance 102 also includes a storage device 210 (such as one or more hard disks or solid-state storage units). In various embodiments, the data appliance 102 stores (whether in RAM 204, storage device 210 and / or other appropriate locations) information for monitoring the enterprise network 140 and implementing the disclosed technology. Examples of such information include application identifiers, content identifiers, user identifiers, requested URLs, IP address mappings, policies and other configuration information, signatures, host name / URL classification information, malware profiles, and machine learning models. The data appliance 102 may also include one or more optional hardware accelerators. For example, the data appliance 102 may include a cryptographic engine 206 configured to perform encryption and decryption operations and one or more field programmable gate arrays (FPGAs) 208 configured to perform matching, act as a network processor, and / or perform other tasks.
[0044] The functions described herein as being performed by the data appliance 102 may be provided / implemented in a variety of ways. For example, the data appliance 102 may be a dedicated device or a collection of devices. The functions provided by the data appliance 102 may also be integrated into or executed as software on a general purpose computer, a computer server, a gateway, and / or a network / routing device. In some embodiments, at least some of the services described as being provided by the data appliance 102 are alternatively (or additionally) provided to a client device (e.g., client device 104 or client device 110) by software (e.g., endpoint protection application 132) executing on the client device.
[0045] Whenever the data appliance 102 is described as performing a task, a single component, a subset of components, or all components of the data appliance 102 may collaborate to perform the task. Similarly, whenever a component of the data appliance 102 is described as performing a task, a subcomponent may perform the task and / or the component may perform the task together with other components. In various embodiments, portions of the data appliance 102 are provided by one or more third parties. Depending on factors such as the amount of computing resources available to the data appliance 102, various logical components and / or features of the data appliance 102 may be omitted, and the techniques described herein are adapted accordingly. Similarly, when applicable, additional logical components / features may be included in the embodiments of the data appliance 102. In various embodiments, an example of a component included in the data appliance 102 is an application identification engine, which is configured to identify applications (e.g., using various application signatures for identifying applications based on packet flow analysis). For example, an application identification engine may determine what type of traffic a session involves, such as web browsing—social networking; web browsing—news; SSH, etc.
[0046] Figure 2B 102 is a functional diagram of the logical components of an embodiment of a data appliance. The examples shown are representations of the logical components that may be included in the data appliance 102 in various embodiments. Unless otherwise noted, the various logical components of the data appliance 102 may generally be implemented in various ways, including as a collection of one or more scripts (e.g., written in Java, Python, etc., when applicable).
[0047] As shown, the data appliance 102 includes a firewall, and includes a management plane 232 and a data plane 234. The management plane is responsible for managing user interactions, such as by providing a user interface for configuring policies and viewing log data. The data plane is responsible for managing data, such as by performing packet processing and session processing.
[0048] The network processor 236 is configured to receive packets from client devices (such as the client device 108) and provide them to the data plane 234 for processing. Whenever the flow module 238 identifies a packet as part of a new session, it creates a new session flow. Based on the flow lookup, subsequent packets will be identified as belonging to the session. If applicable, SSL decryption is applied by the SSL decryption engine 240. Otherwise, the processing by the SSL decryption engine 240 is omitted. The decryption engine 240 can help the data appliance 102 inspect and control SSL / TLS and SSH encrypted traffic, and thus help prevent threats that may still be hidden in the encrypted traffic. The decryption engine 240 can also help prevent sensitive content from leaving the enterprise network 140. Decryption can be selectively controlled (e.g., enabled or disabled) based on parameters such as URL categories, traffic sources, traffic destinations, users, user groups, and ports. In addition to decryption policies (e.g., decryption policies that specify which sessions are decrypted), decryption profiles can be assigned to control various options for sessions controlled by the policy. For example, it may be necessary to use a specific cipher suite and encryption protocol version.
[0049] The application identification (APP-ID) engine 242 is configured to determine what type of traffic a session involves. As an example, the application identification engine 242 can identify a GET request in the received data and infer that the session requires an HTTP decoder. In some cases, such as a web browsing session, the identified application can change, and this change will be noticed by the data appliance 102. For example, a user may initially browse to a company wiki (classified as "Web Browsing - Productivity" based on the URL visited), and then browse to a social networking site (classified as "Web Browsing - Social Networking" based on the URL visited). Different types of protocols have corresponding decoders 244.
[0050] Based on the determination made by the application identification engine 242, the data packet is sent to the appropriate decoder 244. The decoder 244 is configured to assemble the data packets (which may have been received out of order) into the correct order, perform tokenization, and extract the information. The decoder 244 also performs signature matching to determine what should happen to the data packet. The SSL encryption engine 246 can re-encrypt the decrypted data as needed. The forwarding module 248 is used to forward the data packet for transmission (e.g., to a destination).
[0051] As in Figure 2BAs shown in FIG. 2 , policies 252 are received and stored in the management plane 232. A policy may include one or more rules, which may be specified using domains and / or host / server names, and each rule may apply one or more signatures or other matching criteria or heuristic algorithms, such as for enforcing security policies on subscriber / IP flows based on various parameters / information extracted from monitored session traffic flows. An interface (I / F) communicator 250 is provided for management communications (e.g., via (REST) APIs, messages or network protocol communications, or other communication mechanisms).
[0052] III. Security Platform
[0053] Back to Figure 1 In various embodiments, the security platform 122 is configured to provide various services (including providing services to the data appliance 102), including analyzing malicious samples (e.g., samples of documents, applications, etc.), classifying applications, classifying domains / URLs / URIs, etc.
[0054] Assume that a malicious individual (using system 120) has created malware 130. The malicious individual hopes that a client device (such as client device 104) will execute a copy of malware 130, compromising the client device and causing the client device to become a bot in a botnet. The compromised client device can then be instructed to perform tasks (e.g., cryptocurrency mining, or engaging in a denial of service attack) and report information to an external entity (such as command and control (C&C) server 150), and, when applicable, receive instructions from C&C server 150.
[0055] Assume that data appliance 102 has intercepted an email sent (e.g., by system 120) to user "Alice" operating client device 104. A copy of malware 130 has been attached to the message by system 120. As an alternative but similar scenario, data appliance 102 may intercept an attempted download of malware 130 (e.g., from a website) by client device 104. In either scenario, data appliance 102 determines whether a signature for a file (e.g., an email attachment or a website download of malware 130) is present on data appliance 102. If a signature is present, the signature may indicate that the file is known to be safe (e.g., whitelisted), and may also indicate that the file is known to be malicious (e.g., blacklisted).
[0056] In various embodiments, the data appliance 102 is configured to work in cooperation with the security platform 122. As an example, the security platform 122 may provide the data appliance 102 with a signature set of known malicious files (e.g., as part of a subscription). If the signature of the malware 130 (e.g., an MD5 hash of the malware 130) is included in the signature set, the data appliance 102 may accordingly (e.g., by detecting that the MD5 hash of the email attachment sent to the client device 104 matches the MD5 hash of the malware 130) prevent the transmission of the malware 130 to the client device 104. The security platform 122 may also provide the data appliance 102 with a list of known malicious domains and / or IP addresses, thereby allowing the data appliance 102 to block traffic between the enterprise network 140 and the C&C server 150 (e.g., in the case where the C&C server 150 is known to be malicious). The list of malicious domains (and / or IP addresses) may also help the data appliance 102 determine when one of its nodes has been compromised. For example, if the client device 104 attempts to contact the C&C server 150, such an attempt is a strong indicator that the client 104 has been compromised by malware (and remedial action should be taken accordingly, such as isolating the client device 104 from communicating with other nodes within the enterprise network 140). The security platform 122 may also provide other types of information to the data appliance 102 (e.g., as part of a subscription), such as a collection of machine learning models that may be used by the data appliance 102 to perform online analysis of files.
[0057] In various embodiments, if the signature of the attachment is not found, the data appliance 102 can take various actions. As a first example, the data appliance 102 can achieve fail-safe by blocking the transmission of any attachment that is not whitelisted as a benign attachment (e.g., does not match the signature of a known good file). The potential disadvantage of this method is that there may be many legitimate attachments that are unnecessarily blocked as potential malware when they are actually benign. As a second example, the data appliance 102 can achieve fail-danger by allowing the transmission of any attachment that is not blacklisted as a malicious attachment (e.g., does not match the signature of a known bad file). The potential disadvantage of this method is that it will not be possible to prevent newly created malware (which the platform 122 has not seen before) from causing harm. As a third example, the data appliance 102 can be configured to provide a file (e.g., malware 130) to the security platform 122 for static / dynamic analysis to determine whether it is malicious and / or to classify it in other ways. The data appliance 102 can take various actions while performing analysis on the attachment (the signature of the attachment does not exist) by the security platform 122. As a first example, data appliance 102 may prevent an email (and attachment) from being delivered to Alice before a response is received from security platform 122. Assuming that platform 122 takes approximately 15 minutes to thoroughly analyze the sample, this means that incoming messages to Alice will be delayed by 15 minutes. In this example, since the attachment is malicious, this delay will not negatively affect Alice. In an alternative example, assume that someone has sent Alice a time-sensitive message with a benign attachment, and the signature of the message also does not exist. Delaying the delivery of the message to Alice by 15 minutes may be considered (e.g., by Alice) unacceptable. As will be described in more detail below, an alternative approach is to perform at least some real-time analysis of the attachment on data appliance 102 (e.g., while waiting for a determination from platform 122). If data appliance 102 can independently determine whether the attachment is malicious or benign, it can take initial action (e.g., block or allow delivery to Alice), and once a determination is received from security platform 122, additional actions can be adjusted / taken when applicable.
[0058] Security platform 122 stores a copy of the received sample in storage device 142 and begins analysis (or scheduling when applicable). An example of storage device 142 is an Apache Hadoop cluster (HDFS). The results of the analysis (and additional information about the application) are stored in database 146. In the event that the application is determined to be malicious, the data appliance can be configured to automatically block file downloads based on the analysis results. In addition, a signature can be generated for malware and distributed (e.g., to data appliances such as data appliances 102, 136, and 148) to automatically block future file transfer requests to download files determined to be malicious.
[0059] In various embodiments, the security platform 122 includes one or more dedicated commercially available hardware servers (e.g., with (one or more) multi-core processors, 32G+ of RAM, (one or more) gigabit network interface adapters, and (one or more) hard drives) running a typical server-class operating system (e.g., Linux). The security platform 122 can be implemented across a scalable infrastructure that includes multiple such servers, solid-state drives, and / or other applicable high-performance hardware. The security platform 122 can include several distributed components, including components provided by one or more third parties. For example, part or all of the security platform 122 can be implemented using Amazon Elastic Compute Cloud (EC2) and / or Amazon Simple Storage Service (S3). In addition, as with the data appliance 102, whenever the security platform 122 is referred to as performing a task (such as storing data or processing data), it is understood that a subcomponent or multiple subcomponents of the security platform 122 (whether individually or in collaboration with third-party components) can collaborate to perform the task. As an example, the security platform 122 can optionally collaborate with one or more virtual machine (VM) servers (such as VM server 124) to perform static / dynamic analysis.
[0060] An example of a virtual machine server is a physical machine that includes commercially available server-class hardware (e.g., a multi-core processor, 32+ gigabytes of RAM, and one or more gigabit network interface adapters) running open source and / or commercially available virtualization software (such as Linux Kernel-based Virtual Machine (KVM), VMware ESXi, Citrix XenServer, and / or Microsoft Hyper-V). In some examples, the virtual machine server is omitted. In addition, the virtual machine server can be under the control of the same entity that manages the security platform 122, but can also be provided by a third party. As an example, the virtual machine server can rely on EC2, where the remainder of the security platform 122 is provided by dedicated hardware, which is owned by the operator of the security platform 122 and under the control of the operator of the security platform 122. The VM server 124 is configured to provide one or more virtual machines 126-128 for emulating client devices. The virtual machine can execute various operating systems and / or versions of the various operating systems. The observed behavior generated by executing the application in the virtual machine (e.g., an indication that the application is malicious) is recorded and analyzed. In some embodiments, log analysis is performed by a VM server (e.g., VM server 124). In other embodiments, the analysis is performed at least in part by other components of the security platform 122 , such as the orchestrator 144 .
[0061] In various embodiments, as part of the subscription, the security platform 122 provides the data appliance 102 with the results of its sample analysis via a list of signatures (and / or other identifiers). For example, the security platform 122 may periodically send content packages that identify malware applications (e.g., daily, hourly, or at some other interval, and / or based on events configured by one or more policies). An example content package includes a list of identified malware applications, with information such as the package name, a hash value for uniquely identifying the application, and a malware name (and / or malware family name) for each identified malware application. The subscription may cover analysis of only those files that are intercepted by the data appliance 102 and sent to the security platform 122 by the data appliance 102, and may also cover signatures of all malware known to the security platform 122 (or a subset of all malware, such as only mobile malware, not other forms of malware (e.g., PDF malware)). The platform 122 may also provide other types of information, such as machine learning models that may help the data appliance 102 detect malware (e.g., through techniques other than hash-based signature matching).
[0062] In various embodiments, the security platform 122 is configured to provide security services to various entities in addition to (or, where applicable, in lieu of) the operator of the data appliance 102. For example, other enterprises with their own respective enterprise networks 114 and 116 and their own respective data appliances 136 and 148 may contract with the operator of the security platform 122. Other types of entities may also utilize the services of the security platform 122. For example, an Internet service provider (ISP) that provides Internet services to the client device 110 may contract with the security platform 122 to analyze applications that the client device 110 attempts to download. As another example, the owner of the client device 110 may install the endpoint protection software 134 on the client device 110 that communicates with the security platform 122 (e.g., to receive content packages from the security platform 122, use the received content packages to inspect attachments according to the techniques described herein, and transmit applications to the security platform 122 for analysis).
[0063] In various embodiments, the security platform 122 is configured to collaborate with one or more third-party services. As one example, the security platform 122 can provide malware scan results (and other information, when applicable) to a third-party scanner service (e.g., VirusTotal). The security platform 122 can similarly merge information obtained from a third-party scanner service (e.g., malicious determinations from entities other than the security platform 122) into its own information (e.g., information stored in the database 146 or another appropriate information repository).
[0064] 4. Analyze the sample using static / dynamic analysis
[0065] Figure 3 An example of logical components that may be included in a system for analyzing a sample is illustrated. Analysis system 300 may be implemented using a single device. For example, the functionality of analysis system 300 may be implemented in malware analysis module 112 incorporated into data appliance 102. Analysis system 300 may also be implemented across multiple different devices. For example, the functionality of analysis system 300 may be provided by security platform 122.
[0066] In various embodiments, the analysis system 300 utilizes a list, database, or other collection of known safe content and / or known objectionable content (in Figure 3Collection 314 may be obtained in various ways, including via a subscription service (e.g., provided by a third party) and / or as a result of other processing (e.g., performed by data appliance 102 and / or security platform 122). Examples of information included in collection 314 are: URLs, domain names, and / or IP addresses of known malicious servers; URLs, domain names, and / or IP addresses of known secure servers; URLs, domain names, and / or IP addresses of known command and control (C&C) domains; signatures, hashes, and / or other identifiers of known malicious applications; signatures, hashes, and / or other identifiers of known secure applications; signatures, hashes, and / or other identifiers of known malicious files (e.g., Android exploit files); signatures, hashes, and / or other identifiers of known secure libraries; and signatures, hashes, and / or other identifiers of known malicious libraries.
[0067] A. Absorption
[0068] In various embodiments, when a new sample is received for analysis (e.g., an existing signature associated with the sample does not exist in the analysis system 300), it is added to the queue 302. Figure 3 As shown in , application 130 is received by system 300 and added to queue 302 .
[0069] B. Static Analysis
[0070] Coordinator 304 monitors queue 302, and when resources (e.g., static analysis workers) become available, coordinator 304 takes samples from queue 302 for processing (e.g., taking out a copy of malware 130). In particular, coordinator 304 first provides the sample to static analysis engine 306 for static analysis. In some embodiments, one or more static analysis engines are included in analysis system 300, where analysis system 300 is a single device. In other embodiments, static analysis is performed by a separate static analysis server that includes multiple workers (i.e., multiple instances of static analysis engine 306).
[0071] The static analysis engine (which may be implemented via a collection of scripts written in an appropriate scripting language) obtains general information about the sample and includes it (along with heuristics and other information, when applicable) in a static analysis report 308. The report may be created by the static analysis engine, or by the coordinator 304 (or by another appropriate component), which may be configured to receive the information from the static analysis engine 306. In some embodiments, instead of or in addition to creating a separate static analysis report 308 (i.e., a portion of the database record forming the report 308), the collected information is stored in a database record for the sample (e.g., in the database 316). In some embodiments, the static analysis engine also forms a decision about the application (e.g., "safe", "suspicious", or "malicious"). As an example, if there is even one "malicious" static feature in the application (e.g., the application includes a hard link to a known malicious domain), the decision may be "malicious". As another example, each feature can be assigned a point (e.g., based on severity if found; based on how reliable the feature is for predicting maliciousness; etc.) and a determination can be assigned by static analysis engine 306 (or orchestrator 304, if applicable) based on the number of points associated with the static analysis results.
[0072] C. Dynamic Analysis
[0073] Once the static analysis is complete, the coordinator 304 locates an available dynamic analysis engine 310 to perform dynamic analysis on the application. As with the static analysis engine 306, the analysis system 300 may directly include one or more dynamic analysis engines. In other embodiments, the dynamic analysis is performed by a separate dynamic analysis server that includes multiple workers (i.e., multiple instances of the dynamic analysis engine 310).
[0074] Each dynamic analysis worker manages a virtual machine instance. In some embodiments, the results of the static analysis (e.g., performed by the static analysis engine 306), whether in the form of a report (308) and / or stored in a database 316, or stored in other ways, are provided as input to the dynamic analysis engine 310. For example, static report information can be used to help select / customize the virtual machine instance used by the dynamic analysis engine 310 (e.g., Microsoft Windows 7 SP2 vs. Microsoft Windows 10 Enterprise Edition, or iOS 11.0 vs. iOS 12.0). In the case of multiple virtual machine instances executing simultaneously, a single dynamic analysis engine can manage all instances, or multiple dynamic analysis engines can be used (e.g., each of which manages its own virtual machine instance). During the dynamic portion of the analysis, the actions taken by the application (including network activity) are analyzed.
[0075] In various embodiments, static analysis of the sample is omitted or performed by a separate entity, when applicable. As an example, conventional static and / or dynamic analysis can be performed on a file by a first entity. Once a determination is made (e.g., by a first entity) that a given file is suspicious or malicious, the file can be provided to a second entity (e.g., an operator of security platform 122) for additional analysis specifically regarding malware's use of network activity (e.g., by dynamic analysis engine 310).
[0076] The environment used by the analysis system 300 is instrumented / hooked so that behaviors observed while the application executes are logged as they occur (e.g., using a custom kernel that supports hooks and log capture (logcat)). Network traffic associated with the emulator is also captured (e.g., using pcap). The log / network data can be stored as temporary files on the analysis system 300, and can also be stored more persistently (e.g., using HDFS or another appropriate storage technology or combination of technologies, such as MongoDB). The dynamic analysis engine (or another appropriate component) can compare the connections established by the sample to a list (314) of domains, IP addresses, etc., and determine whether the sample has communicated (or attempted to communicate) with a malicious entity.
[0077] As with the static analysis engine, the dynamic analysis engine stores its analysis results in a record associated with the application being tested in database 316 (and / or includes the results in report 312, when applicable). In some embodiments, the dynamic analysis engine also forms a determination (e.g., "safe," "suspicious," or "malicious") about the application. As an example, if the application takes even just one "malicious" action (e.g., an attempt is made to contact a known malicious domain, or an attempt is observed to leak sensitive information), the determination may be "malicious." As another example, points may be assigned to the action taken (e.g., based on severity (if found); based on how reliable the action is for predicting maliciousness; etc.) and the determination may be assigned by dynamic analysis engine 310 (or coordinator 304, when applicable) based on the number of points associated with the dynamic analysis results. In some embodiments, a final determination associated with the sample is made based on a combination of report 308 and report 312 (e.g., by coordinator 304).
[0078] 5. Signature collection based on automatic fuzzy hashing
[0079] As mentioned above, malicious individuals are constantly looking for new ways to prevent security infrastructure from detecting the malicious nature of their malware. One often used approach is for malware authors to make various versions of a given piece of malware (usually with subtle differences) so that although a first signature (e.g., stored in database 146 and available to data appliance 102) may successfully mark the first piece of malware as malicious, the same signature will not match a new variant of the malware, and the new malware variant will not be detected (e.g., by data appliance 102). One way to combat this technique is for the security platform 122 to use a fuzzy hashing scheme. An example of a fuzzy hashing scheme is "ssdeep," which can efficiently calculate context-triggered piece-by-piece hashes (CTPH), also known as fuzzy hashing. Fuzzy hashing can be used to detect new instances of malware that are similar to, but not necessarily identical to, known malware. As malware authors continue to modify / release variants of their malware, fuzzy hashing can be very useful in identifying / mitigating the potential harm that this new malware may cause. Unfortunately, fuzzy hashing has disadvantages. In particular, they may result in both false positives (where benign samples are incorrectly labeled as malicious) and false negatives (where malicious samples are incorrectly classified as benign). In various embodiments, one task performed by the security platform 122 is malware signature generation. And, as will be described in more detail below, in various embodiments, the security platform 122 includes an automatic fuzzy hash signature generator 150 that can efficiently and automatically generate fuzzy hash-based signatures that minimize both false positives and false negatives.
[0080] Figure 4A An embodiment of a process for performing detection of command and control malware is illustrated. In some embodiments, process 400 is performed by Figure 3 The system 300 is implemented and includes:
[0081] At 410, the system monitors network traffic sessions.
[0082] At 420 , the system uses a deep learning model to perform automatic feature recognition for real-time malicious command and control traffic detection based on request headers of the monitored network traffic sessions.
[0083] In some embodiments, the system extracts a request header from a network traffic session. In some embodiments, the request header relates to an HTTP request. In some embodiments, the request header is tokenized to generate character tokens and word tokens. Character tokens can be described as single-letter words. Character tokens and word tokens can be "symbols" including letters and numbers, for example, as defined in the ASCII table. In some embodiments, tokenization includes tokenizing special characters to obtain words so that the words end with special characters. In some embodiments, no word is required to end with a special character. In some embodiments, special characters are removed during the tokenization operation. Examples of special characters include " / ", "\", "$", "#", ";", etc. In some embodiments, character tokens and word tokens are filtered or cleared (e.g., words and / or characters that appear twice in the request header are removed).
[0084] Figure 4B An embodiment of a process for performing automatic feature recognition is illustrated. In some embodiments, process 4200 is Figure 4A The implementation of operation 420 includes:
[0085] At 4210, the system extracts a request header from the network traffic session. In some embodiments, the system pads the request header to a fixed length (e.g., the fixed length is 500 characters, 600 characters, 700 characters, etc.). In some embodiments, the system determines the word from the number using a vocabulary index map.
[0086] At 4220, the system performs tokenization of the request header to generate character tokens and word tokens. In some embodiments, special characters of the request header are tokenized into word tokens.
[0087] In 4230, the system feeds the character word and word word into the embedding layer of the deep learning model to find the feature group. In some embodiments, the deep learning model corresponds to a neural network. In some embodiments, the embedding layer is multiplied by the filter matrix to obtain a one-dimensional convolutional neural network (1DCNN). In some embodiments, after obtaining the 1D CNN, the elements of the 1D CNN are reduced by the elements of the 1D CNN by maximum pooling. In some embodiments, the maximum pooled elements are fed into a bidirectional long short-term memory (Bi-LSTM) with attention to obtain the contextual relationship between the features. After obtaining the contextual relationship, the feature with the highest contextual relationship score is determined to obtain the feature group. In some embodiments, the feature group is fed into the fully connected layer of a deep learning model (such as a neural network), and the result is obtained from the deep learning model.
[0088] Figure 4CAn embodiment of a process for feeding character tokens and word tokens into an embedding layer of a deep learning model to find feature groups is illustrated. In some embodiments, process 42300 is an implementation of operation 4230 and includes:
[0089] At 42310, the system multiplies the embedding layer by the filter matrix to obtain the convolutional layer. In some embodiments, the embedding layer is multiplied by the filter matrix to obtain the first element of the convolutional layer, and then the filter matrix is shifted down one row to a different set of elements of the embedding layer to obtain the second element of the convolutional layer, and so on, to obtain all elements of the convolutional layer. In some embodiments, the filter matrix is a k×Y matrix, where k is an integer less than the number of rows in the embedding layer (e.g., 2, 3, 4, 5, etc.) and Y is the number of columns in the filter matrix (e.g., 300, 400, etc.). In some embodiments, each element of the filter matrix is a randomly generated number.
[0090] In 42320, the system max-pools multiple elements of the convolutional layer to obtain a feature set. In some embodiments, the system obtains a maximum value from the pool of elements. In some embodiments, the pool corresponds to a fixed number of elements of the convolutional layer (e.g., 3, 5, 10, 15, etc.). In some embodiments, the system obtains a minimum value from the pool of elements. In some embodiments, the system obtains a mean or average value from the pool of elements. In some embodiments, max-pooling, min-pooling, or average-pooling is used to reduce the feature set.
[0091] In some embodiments, after identifying the feature set, a cost function is determined.
[0092] Figure 5 An example of a request header is illustrated. In some embodiments, the request header is padded with characters so that the request header is a fixed length (e.g., 500 characters, 600 characters, etc.). In some embodiments, commonly used characters are located in the request header. In some embodiments, special characters in the request header are used for tokenization. In some embodiments, numbers and words that appear more than once are cleared so that repeated numbers and words appear only once after tokenization. For example, the words in the first line of the request header example include GET, sobaka, gif, 12db3cf, 98861835, and HTTP.
[0093] In some embodiments, when tokens of a request header are used to perform detection of command and control malware within a firewall, the tokens of the request header are converted into integer encodings and fed to an embedding layer.
[0094] Figure 6An example of an embedding layer is illustrated. In this example, words extracted from multiple request headers are listed in a column. As an example, the number of words in the column is 200, 500, 600, 700, etc. In some embodiments, the number of items in the column corresponds to all the different words that have been extracted from the various request headers. Furthermore, in this example, the number of elements in a row of the embedding layer is 300. In this example, the embedding layer is a 600×300 matrix. In some embodiments, the embedding layer is initialized so that each element of the embedding layer is a random number.
[0095] Figure 7 An example of a one-dimensional convolution is illustrated. In some embodiments, the filter matrix consists of random elements, where the number of elements in the column is k, where k is an integer of at least 1, and the number of elements in the row is N, where N is, for example, 300. For example, k is 2, 4, 5, 10, 11, etc. In some embodiments, the number of elements in the rows of the filter matrix is the same as the number of elements in the rows of the embedding layer. In this example, the size of the filter column is set to 5. In some embodiments, the number of elements in the row is 100, 200, etc.
[0096] The embedding layer multiplies the filter matrix to output the first element of the one-dimensional (1D) convolution (e.g., the elements of the embedding layer are multiplied by the corresponding elements of the filter matrix, and the products of the various multiplied elements are summed to obtain the elements of the 1D convolution). In addition, the filter matrix is shifted one column to the right, and the second element of the 1D convolution is calculated. The movement and calculation can be repeated until the entire 1D convolution is obtained. Please note that the second element can be obtained before the first element, after the first element, or at the same time as the first element. In other words, the order in which the first element and the second element (or any element of the 1D convolution) are obtained is not important. In some embodiments, the 1D convolution is max-pooled to reduce the number of elements output. In this example, the six elements of the 1D convolution are max-pooled to obtain the maximum value of the group of six elements. As an aspect, two groups with three elements can be used for max-pooling, so that the result is that the two elements are the maximum values of their corresponding groups. In this example, the max-pooling output is 0.8, which corresponds to the sobaka element.
[0097] Figure 8 Illustration of an example model architecture with convolutions.
[0098] In this example, the model architecture includes multiple request headers. After being tokenized, the multiple request headers can form a set of embedding layers, which are multiplied by filter matrices with various widths to obtain a set of 1D convolutions. In some embodiments, the set of 1D convolutions is max-pooled to obtain a reduced set of elements.
[0099] Fig. 9An example of a bidirectional long short-term memory with attention is illustrated. As an example, the reduced set of elements is input into a bidirectional long short-term memory (LSTM) to identify two elements related to each other. In some embodiments, the LSTM has a feedback connection, so when each element passes through the LSTM unit, the LSTM unit calculates a score based on the elements it has seen before (from front to back). In Bi-LSTM, these scores (first scores) can also be calculated based on the elements that the LSTM unit will see later (from back to front), and the first score can be stored in the unit state (memory of the LSTM unit). The LSTM unit can also only calculate the score of the instant element, and store the score of the instant element (second score) in the hidden state. These two scores (first score and second score) are multiplied to obtain a relationship score. After the reduced set of elements is input into the bidirectional LSTM, the bidirectional LSTM outputs a set of related elements, wherein the element pairs have a contextual relationship based on the relationship score. In some embodiments, the element pairs related to each other are used as the input of the deep learning network.
[0100] Fig.10 Illustration of an example of a fully connected layer in a deep learning network.
[0101] In the example, the related elements (x1, x2, ..., x m ) is input into a fully connected layer in a deep learning network, and the values of y and 1-y are output from the fully connected layer. As an example, the deep learning network is a neural network. In some embodiments, y represents the possibility that the sample is malicious and (1-y) represents the possibility that the sample is benign. For example, y is 0.99 and 1-y is 0.01. In some embodiments, the fully connected layer learns how to classify a collection of related elements. In this example, the closer y is to 1, the more likely the sample is malicious, and the closer y is to 0, the more likely the sample is benign. In this example, using a threshold of 0.99, the fully connected network indicates that the sample is malicious, and the true value (truth) found in the request header confirms that the sample is malicious.
[0102] Fig.11 An example of a cost function is illustrated.
[0103] In some embodiments, the cost function corresponds to the cross entropy trained by stochastic gradient descent and back propagation algorithms. n corresponds to the probability output from the deep learning network and t n represents the true value of the sample. In some embodiments, t nis a label of the request header, and the label can be malicious (1) or benign (0). Subsequently, the embedding layer and the filter matrix can be trained using a cost function. For example, the result of the cost function is multiplied by each element of the embedding layer to obtain a new embedding layer, and the result of the cost function is multiplied by each element of the filter matrix to obtain a new filter matrix, and the entire process is repeated using the new embedding layer and the new filter matrix. In some embodiments, during the training of the embedding layer and the filter matrix, the embedding layer and the filter matrix are updated multiple times to obtain increased model accuracy and reduced model loss.
[0104] After the model (including the embedding layer, the CNN layer, the bidirectional LSTM layer, and the fully connected layer) is trained, the model can be deployed to the cloud or the firewall itself, so that the cloud or the firewall can use the model to perform real-time malicious command and control traffic detection.
[0105] Fig.12 An example of the results from model training is illustrated. The accuracy of the model is evaluated using a set of real-world test data or using a set of training data. For the set of training data, the accuracy of the model increases steadily until the accuracy after each epoch or iteration is approximately 0.995. Conversely, for the set of training data, the loss of the model decreases steadily until the loss after each epoch or iteration reaches approximately 0.005. For the set of real-world test data, the accuracy of the model increases steadily until the accuracy after each epoch or iteration reaches a plateau at approximately 0.98. Conversely, for the set of training data, the loss of the model decreases steadily until the loss after each epoch or iteration reaches a plateau at approximately 0.02.
[0106] Although the foregoing embodiments have been described in specific detail for the purpose of clarity of understanding, the invention is not limited to the details provided. There are many alternative ways to implement the invention. The disclosed embodiments are illustrative rather than restrictive.
Claims
1. A system comprising: The processor is configured as: Monitor network traffic sessions; as well as Use deep learning models to perform automatic feature recognition for real-time malicious command and control traffic detection based on request headers of monitored network traffic sessions; as well as A memory is coupled to the processor and configured to provide instructions to the processor.
2. The system of claim 1, wherein performing automatic feature recognition for real-time malicious command and control traffic detection comprises: extracting the request header from the network traffic session; performing lemmatization of the request header to generate character lemmas and word lemmas; as well as The character tokens and the word tokens are fed into an embedding layer of the deep learning model to find feature groups.
3. The system of claim 1 , wherein performing automatic signature recognition for real-time malicious command and control traffic detection comprises: extracting the request header from the network traffic session; performing lemmatization of the request header to generate character lemmas and word lemmas; as well as Feeding the character tokens and the word tokens into an embedding layer of the deep learning model to find feature groups, including: Multiplying the embedding layer with a filter matrix to obtain a convolutional layer; as well as Multiple elements of the convolutional layer are max-pooled to obtain the feature group.
4. The system of claim 3, wherein multiplying the embedding layer with the filter matrix to obtain the convolutional layer comprises: At a first position, multiplying the embedding layer by the filter matrix to obtain a first element of the convolutional layer; shifting the filter matrix down one row along the embedding layer to obtain a second position of the filter matrix; as well as The embedding layer is multiplied by the filter matrix at the second position to obtain a second element of the convolutional layer.
5. The system of claim 3, wherein performing automatic signature recognition for real-time malicious command and control traffic detection comprises: After performing the maximum pooling on the plurality of elements, a bidirectional long short-term memory (Bi-LSTM) is used to find contextual relationships between features of the feature group.
6. The system of claim 1, wherein the processor is further configured to: Using the fully connected layers of the deep learning model, the labels of the Hypertext Transfer Protocol (HTTP) sessions are predicted based on the probability scores.
7. The system of claim 1, wherein the processor is further configured to: A label of a Hypertext Transfer Protocol (HTTP) session is predicted based on the probability score using a fully connected layer of the deep learning model, wherein the deep learning model corresponds to a neural network.
8. A method comprising: Monitor network traffic sessions; as well as Use deep learning models to perform automatic feature recognition for real-time malicious command and control traffic detection based on request headers of monitored network traffic sessions.
9. The method of claim 8, wherein performing automatic signature recognition for real-time malicious command and control traffic detection comprises: extracting the request header from the network traffic session; performing lemmatization of the request header to generate character lemmas and word lemmas; as well as The character tokens and the word tokens are fed into an embedding layer of the deep learning model to find feature groups.
10. The method according to claim 8, further comprising: Using the fully connected layers of the deep learning model, the labels of the Hypertext Transfer Protocol (HTTP) sessions are predicted based on the probability scores.
11. The method of claim 8, wherein the deep learning model corresponds to a neural network.
12. A computer program product embodied in a non-transitory computer readable medium and comprising computer instructions for: Monitor network traffic sessions; and Use deep learning models to perform automatic feature recognition for real-time malicious command and control traffic detection based on request headers of monitored network traffic sessions.
13. The computer program product of claim 12, wherein performing automatic feature recognition for real-time malicious command and control traffic detection comprises: extracting the request header from the network traffic session; performing lemmatization of the request header to generate character lemmas and word lemmas; as well as The character tokens and the word tokens are fed into an embedding layer of the deep learning model to find feature groups.
14. The computer program product of claim 12, further comprising computer instructions for: Using the fully connected layers of the deep learning model, the labels of the Hypertext Transfer Protocol (HTTP) sessions are predicted based on the probability scores.
15. The computer program product of claim 12, wherein the deep learning model corresponds to a neural network.
