U.S. patent application number 15/399081 was filed with the patent office on 2018-07-05 for training a machine learning-based traffic analyzer using a prototype dataset.
The applicant listed for this patent is Cisco Technology, Inc.. Invention is credited to Blake Harrell Anderson, David McGrew.
Application Number | 20180189677 15/399081 |
Document ID | / |
Family ID | 60915435 |
Filed Date | 2018-07-05 |
United States Patent
Application |
20180189677 |
Kind Code |
A1 |
Anderson; Blake Harrell ; et
al. |
July 5, 2018 |
TRAINING A MACHINE LEARNING-BASED TRAFFIC ANALYZER USING A
PROTOTYPE DATASET
Abstract
In one embodiment, a device in a network generates a feature
vector based on traffic flow data regarding one or more traffic
flows in the network. The device makes a determination as to
whether the generated feature vector is already represented in a
training dataset dictionary by one or more feature vectors in the
dictionary. The device updates the training dataset dictionary
based on the determination by one of: adding the generated feature
vector to the dictionary when the generated feature vector is not
already represented by one or more feature vectors in the
dictionary, or incrementing a count associated with a particular
feature vector in the dictionary when the generated feature vector
is already represented by the particular feature vector in the
dictionary. The device generates a training dataset based on the
training dataset dictionary for training a machine learning-based
traffic flow analyzer.
Inventors: |
Anderson; Blake Harrell;
(San Jose, CA) ; McGrew; David; (Poolesville,
MD) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Cisco Technology, Inc. |
San Jose |
CA |
US |
|
|
Family ID: |
60915435 |
Appl. No.: |
15/399081 |
Filed: |
January 5, 2017 |
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
H04L 63/1408 20130101;
H04L 63/0428 20130101; G06N 20/00 20190101; H04L 63/1433 20130101;
H04L 63/1425 20130101; H04L 63/1441 20130101 |
International
Class: |
G06N 99/00 20060101
G06N099/00; H04L 29/06 20060101 H04L029/06 |
Claims
1. A method comprising: generating, by a device in a network, a
feature vector based on traffic flow data regarding one or more
traffic flows in the network; making, by the device, a
determination as to whether the generated feature vector is already
represented in a training dataset dictionary by one or more feature
vectors in the dictionary; updating, by the device, the training
dataset dictionary based on the determination as to whether the
generated feature vector is already represented in the training
dataset dictionary by one or more feature vectors in the
dictionary, wherein updating the training dataset dictionary
comprises one of: adding the generated feature vector to the
dictionary when the generated feature vector is not already
represented by one or more feature vectors in the dictionary, or
incrementing a count associated with a particular feature vector in
the is dictionary when the generated feature vector is already
represented by the particular feature vector in the dictionary; and
generating, by the device, a training dataset based on the training
dataset dictionary for training a machine learning-based traffic
flow analyzer.
2. The method as in claim 1, wherein the training dataset comprises
a plurality of labels, and wherein the machine learning-based
traffic flow analyzer comprises a machine learning-based traffic
flow classifier.
3. The method as in claim 1, wherein the traffic flow data
comprises header information for one or more encrypted traffic
flows.
4. The method as in claim 1, wherein adding the generated feature
vector to the dictionary comprises: initializing, by the device, a
count associated with the generated feature vector.
5. The method as in claim 1, wherein making the determination as to
whether the generated feature vector is already represented in the
training dataset dictionary comprises: computing, by the device,
function values between the generated feature vector and feature
vectors in the training dataset dictionary.
6. The method as in claim 5, wherein the function values are
computed using a squared exponential function.
7. The method as in claim 1, wherein generating the training
dataset based on the training dataset dictionary comprises:
receiving, at the device, one or more parameters indicative of a
particular traffic type for a target network to which the traffic
flow analyzer is to be deployed; identifying, by the device, one or
more feature vectors in the dictionary that are associated with the
particular traffic type; and determining, by the device, a
representation of the identified one or more feature vectors in the
generated training dataset based on the received one or more
parameters.
8. The method as in claim 7, wherein the representation of the
identified one or more feature vectors in the generated training
dataset excludes the identified one or more feature vectors in the
dictionary associated with the particular traffic type from the
generated training dataset.
9. The method as in claim 1, wherein the machine learning-based
traffic flow analyzer is configured to detect malicious traffic
flows.
10. An apparatus, comprising: one or more network interfaces to
communicate with a network; a processor coupled to the network
interfaces and configured to execute one or more processes; and a
memory configured to store a process executable by the processor,
the process when executed operable to: generate a feature vector
based on traffic flow data regarding one or more traffic flows in
the network; make a determination as to whether the generated
feature vector is already represented in a training dataset
dictionary by one or more feature vectors in the dictionary; update
the training dataset dictionary based on the determination as to
whether the generated feature vector is already represented in the
training dataset dictionary by one or more feature vectors in the
dictionary, wherein the apparatus updates the training dataset
dictionary by one of: adding the generated feature vector to the
dictionary when the generated feature vector is not already
represented by one or more feature vectors in the dictionary, or
incrementing a count associated with a particular feature vector in
the dictionary when the generated feature vector is already
represented by the particular feature vector in the dictionary; and
generate a training dataset based on the training dataset
dictionary for training a machine learning-based traffic flow
analyzer.
11. The apparatus as in claim 10, wherein the training dataset
comprises a plurality of labels, and wherein the machine
learning-based traffic flow analyzer comprises a machine
learning-based traffic flow classifier.
12. The apparatus as in claim 10, wherein the traffic flow data
comprises header information for one or more encrypted traffic
flows.
13. The apparatus as in claim 10, wherein the apparatus adds the
generated feature vector to the dictionary by: initializing a count
associated with the generated feature vector.
14. The apparatus as in claim 10, wherein the apparatus makes the
determination as to whether the generated feature vector is already
represented in the training dataset dictionary by: computing
function values between the generated feature vector and feature
vectors in the training dataset dictionary.
15. The apparatus as in claim 14, wherein the function values are
computed using a squared exponential function.
16. The apparatus as in claim 10, wherein the apparatus generates
the training dataset based on the training dataset dictionary by:
receiving one or more parameters indicative of a particular traffic
type for a target network to which the traffic flow analyzer is to
be deployed; identifying one or more feature vectors in the
dictionary that are associated with the particular traffic type;
and determining a representation of the identified one or more
feature vectors in the generated training dataset based on the
received one or more parameters.
17. The apparatus as in claim 16, wherein the representation of the
identified one or more feature vectors in the generated training
dataset excludes the identified one or more feature vectors in the
dictionary associated with the particular traffic type from the
generated training dataset.
18. The apparatus as in claim 10, wherein the machine
learning-based traffic flow analyzer is configured to detect
malicious traffic flows.
19. A tangible, non-transitory, computer-readable medium storing
program instructions that cause a device in a network to execute a
process comprising: generating, by the device, a feature vector
based on traffic flow data regarding one or more traffic flows in
the network; making, by the device, a determination as to whether
the generated feature vector is already represented in a training
dataset dictionary by one or more feature vectors in the
dictionary; updating, by the device, the training dataset
dictionary based on the determination as to whether the generated
feature vector is already represented in the training dataset
dictionary by one or more feature vectors in the dictionary,
wherein updating the training dataset dictionary comprises one of:
adding the generated feature vector to the dictionary when the
generated feature vector is not already represented by one or more
feature vectors in the dictionary, or incrementing a count
associated with a particular feature vector in the dictionary when
the generated feature vector is already represented by the
particular feature vector in the dictionary; and generating, by the
device, a training dataset based on the training dataset dictionary
for training a machine learning-based traffic flow analyzer.
20. The computer-readable medium as in claim 19, wherein making the
determination as to whether the generated feature vector is already
represented in the training dataset dictionary comprises:
computing, by the device, function values between the generated
feature vector and feature vectors in the training dataset
dictionary.
Description
TECHNICAL FIELD
[0001] The present disclosure relates generally to computer
networks, and, more particularly, to the training network traffic
flow classifiers using a prototype training dataset.
BACKGROUND
[0002] Enterprise networks are carrying a very fast growing volume
of both business and non-business critical traffic. Often, business
applications such as video collaboration, cloud applications, etc.,
use the same hypertext transfer protocol (HTTP) and/or HTTP secure
(HTTPS) techniques that are used by non-business critical web
traffic. This complicates the task of optimizing network
performance for specific applications, as many applications use the
same protocols, thus making it difficult to distinguish and select
traffic flows for optimization.
[0003] Beyond the various types of legitimate application traffic
in a network, some network traffic may also be malicious. For
example, some traffic may seek to overwhelm a service by sending a
large number of requests to the service. Such attacks are also
sometimes known as denial of service (DoS) attacks. Other forms of
malicious traffic may seek to exfiltrate sensitive information from
a network, such as credit card numbers, trade secrets, and the
like. Typically, such traffic is generated by a client that has
been infected with malware. Thus, further types of malicious
network traffic include network traffic that propagate the malware
itself and network traffic that passes control commands to already
infected devices.
[0004] To further complicate the analysis of network traffic to
discern between different types of traffic, the use of encryption
is also steadily increasing. Notably, many websites are now using
encryption, thereby protecting the payloads of their corresponding
traffic from inspection. Malicious entities are also leveraging
this fact to conceal malicious traffic using encryption.
BRIEF DESCRIPTION OF THE DRAWINGS
[0005] The embodiments herein may be better understood by referring
to the following description in conjunction with the accompanying
drawings in which like reference numerals indicate identically or
functionally similar elements, of which:
[0006] FIGS. 1A-1B illustrate an example communication network;
[0007] FIG. 2 illustrates an example network device/node;
[0008] FIG. 3 illustrates an example of the capture of traffic flow
data;
[0009] FIG. 4 illustrates an example architecture for generating a
training dataset for a traffic flow analyzer; and
[0010] FIG. 5 illustrates an example simplified procedure for
generating a training dataset for a traffic flow analyzer.
DESCRIPTION OF EXAMPLE EMBODIMENTS
Overview
[0011] According to one or more embodiments of the disclosure, a
device in a network generates a feature vector based on traffic
flow data regarding one or more traffic flows in the network. The
device makes a determination as to whether the generated feature
vector is already represented in a training dataset dictionary by
one or more feature vectors in the dictionary. The device updates
the training dataset dictionary based on the determination by one
of: adding the generated feature vector to the dictionary when the
generated feature vector is not already represented by one or more
feature vectors in the dictionary, or incrementing a count
associated with a particular feature vector in the dictionary when
the generated feature vector is already represented by the
particular feature vector in the dictionary. The device generates a
training dataset based on the training dataset dictionary for
training a machine learning-based traffic flow analyzer.
Description
[0012] A computer network is a geographically distributed
collection of nodes interconnected by communication links and
segments for transporting data between end nodes, such as personal
computers and workstations, or other devices, such as sensors, etc.
Many types of networks are available, with the types ranging from
local area networks (LANs) to wide area networks (WANs). LANs
typically connect the nodes over dedicated private communications
links located in the same general physical location, such as a
building or campus. WANs, on the other hand, typically connect
geographically dispersed nodes over long-distance communications
links, such as common carrier telephone lines, optical lightpaths,
synchronous optical networks (SONET), or synchronous digital
hierarchy (SDH) links, or Powerline Communications (PLC) such as
IEEE 61334, IEEE P1901.2, and others. The Internet is an example of
a WAN that connects disparate networks throughout the world,
providing global communication between nodes on various networks.
The nodes typically communicate over the network by exchanging
discrete frames or packets of data according to predefined
protocols, such as the Transmission Control Protocol/Internet
Protocol (TCP/IP). In this context, a protocol consists of a set of
rules defining how the nodes interact with each other. Computer
networks may be further interconnected by an intermediate network
node, such as a router, to extend the effective "size" of each
network.
[0013] Smart object networks, such as sensor networks, in
particular, are a specific type of network having spatially
distributed autonomous devices such as sensors, actuators, etc.,
that cooperatively monitor physical or environmental conditions at
different locations, such as, e.g., energy/power consumption,
resource consumption (e.g., water/gas/etc. for advanced metering
infrastructure or "AMI" applications) temperature, pressure,
vibration, sound, radiation, motion, pollutants, etc. Other types
of smart objects include actuators, e.g., responsible for turning
on/off an engine or perform any other actions. Sensor networks, a
type of smart object network, are typically shared-media networks,
such as wireless or PLC networks. That is, in addition to one or
more sensors, each sensor device (node) in a sensor network may
generally be equipped with a radio transceiver or other
communication port such as PLC, a microcontroller, and an energy
source, such as a battery. Often, smart object networks are
considered field area networks (FANs), neighborhood area networks
(NANs), personal area networks (PANs), etc. Generally, size and
cost constraints on smart object nodes (e.g., sensors) result in
corresponding constraints on resources such as energy, memory,
computational speed and bandwidth.
[0014] FIG. 1A is a schematic block diagram of an example computer
network 100 illustratively comprising nodes/devices, such as a
plurality of routers/devices interconnected by links or networks,
as shown. For example, customer edge (CE) routers 110 may be
interconnected with provider edge (PE) routers 120 (e.g., PE-1,
PE-2, and PE-3) in order to communicate across a core network, such
as an illustrative network backbone 130. For example, routers 110,
120 may be interconnected by the public Internet, a multiprotocol
label switching (MPLS) virtual private network (VPN), or the like.
Data packets 140 (e.g., traffic/messages) may be exchanged among
the nodes/devices of the computer network 100 over links using
predefined network communication protocols such as the Transmission
Control Protocol/Internet Protocol (TCP/IP), User Datagram Protocol
(UDP), Asynchronous Transfer Mode (ATM) protocol, Frame Relay
protocol, or any other suitable protocol. Those skilled in the art
will understand that any number of nodes, devices, links, etc. may
be used in the computer network, and that the view shown herein is
for simplicity.
[0015] In some implementations, a router or a set of routers may be
connected to a private network (e.g., dedicated leased lines, an
optical network, etc.) or a virtual private network (VPN), such as
an MPLS VPN thanks to a carrier network, via one or more links
exhibiting very different network and service level agreement
characteristics. For the sake of illustration, a given customer
site may fall under any of the following categories:
[0016] 1.) Site Type A: a site connected to the network (e.g., via
a private or VPN link) using a single CE router and a single link,
with potentially a backup link (e.g., a 3G/4G/LTE backup
connection). For example, a particular CE router 110 shown in
network 100 may support a given customer site, potentially also
with a backup link, such as a wireless connection.
[0017] 2.) Site Type B: a site connected to the network using two
MPLS VPN links (e.g., from different Service Providers), with
potentially a backup link (e.g., a 3G/4G/LTE connection). A site of
type B may itself be of different types:
[0018] 2a.) Site Type B1: a site connected to the network using two
MPLS VPN links (e.g., from different Service Providers), with
potentially a backup link (e.g., a 3G/4G/LTE connection).
[0019] 2b.) Site Type B2: a site connected to the network using one
MPLS VPN link and one link connected to the public Internet, with
potentially a backup link (e.g., a 3G/4G/LTE connection). For
example, a particular customer site may be connected to network 100
via PE-3 and via a separate Internet connection, potentially also
with a wireless backup link.
[0020] 2c.) Site Type B3: a site connected to the network using two
links connected to the public Internet, with potentially a backup
link (e.g., a 3G/4G/LTE connection).
[0021] Notably, MPLS VPN links are usually tied to a committed
service level agreement, whereas Internet links may either have no
service level agreement at all or a loose service level agreement
(e.g., a "Gold Package" Internet service connection that guarantees
a certain level of performance to a customer site).
[0022] 3.) Site Type C: a site of type B (e.g., types B1, B2 or B3)
but with more than one CE router (e.g., a first CE router connected
to one link while a second CE router is connected to the other
link), and potentially a backup link (e.g., a wireless 3G/4G/LTE
backup link). For example, a particular customer site may include a
first CE router 110 connected to PE-2 and a second CE router 110
connected to PE-3.
[0023] FIG. 1B illustrates an example of network 100 in greater
detail, according to various embodiments. As shown, network
backbone 130 may provide connectivity between devices located in
different geographical areas and/or different types of local
networks. For example, network 100 may comprise local/branch
networks 160, 162 that include devices/nodes 10-16 and
devices/nodes 18-20, respectively, as well as a data center/cloud
environment 150 that includes servers 152-154. Notably, local
networks 160-162 and data center/cloud environment 150 may be
located in different geographic locations.
[0024] Servers 152-154 may include, in various embodiments, a
network management server (NMS), a dynamic host configuration
protocol (DHCP) server, a constrained application protocol (CoAP)
server, an outage management system (OMS), an application policy
infrastructure controller (APIC), an application server, etc. As
would be appreciated, network 100 may include any number of local
networks, data centers, cloud environments, devices/nodes, servers,
etc.
[0025] In some embodiments, the techniques herein may be applied to
other network topologies and configurations. For example, the
techniques herein may be applied to peering points with high-speed
links, data centers, etc.
[0026] In various embodiments, network 100 may include one or more
mesh networks, such as an Internet of Things network. Loosely, the
term "Internet of Things" or "IoT" refers to uniquely identifiable
objects (things) and their virtual representations in a
network-based architecture. In particular, the next frontier in the
evolution of the Internet is the ability to connect more than just
computers and communications devices, but rather the ability to
connect "objects" in general, such as lights, appliances, vehicles,
heating, ventilating, and air-conditioning (HVAC), windows and
window shades and blinds, doors, locks, etc. The "Internet of
Things" thus generally refers to the interconnection of objects
(e.g., smart objects), such as sensors and actuators, over a
computer network (e.g., via IP), which may be the public Internet
or a private network.
[0027] Notably, shared-media mesh networks, such as wireless or PLC
networks, etc., are often on what is referred to as Low-Power and
Lossy Networks (LLNs), which are a class of network in which both
the routers and their interconnect are constrained: LLN routers
typically operate with constraints, e.g., processing power, memory,
and/or energy (battery), and their interconnects are characterized
by, illustratively, high loss rates, low data rates, and/or
instability. LLNs are comprised of anything from a few dozen to
thousands or even millions of LLN routers, and support
point-to-point traffic (between devices inside the LLN),
point-to-multipoint traffic (from a central control point such at
the root node to a subset of devices inside the LLN), and
multipoint-to-point traffic (from devices inside the LLN towards a
central control point). Often, an IoT network is implemented with
an LLN-like architecture. For example, as shown, local network 160
may be an LLN in which CE-2 operates as a root node for
nodes/devices 10-16 in the local mesh, in some embodiments.
[0028] In contrast to traditional networks, LLNs face a number of
communication challenges. First, LLNs communicate over a physical
medium that is strongly affected by environmental conditions that
change over time. Some examples include temporal changes in
interference (e.g., other wireless networks or electrical
appliances), physical obstructions (e.g., doors opening/closing,
seasonal changes such as the foliage density of trees, etc.), and
propagation characteristics of the physical media (e.g.,
temperature or humidity changes, etc.). The time scales of such
temporal changes can range between milliseconds (e.g.,
transmissions from other transceivers) to months (e.g., seasonal
changes of an outdoor environment). In addition, LLN devices
typically use low-cost and low-power designs that limit the
capabilities of their transceivers. In particular, LLN transceivers
typically provide low throughput. Furthermore, LLN transceivers
typically support limited link margin, making the effects of
interference and environmental changes visible to link and network
protocols. The high number of nodes in LLNs in comparison to
traditional networks also makes routing, quality of service (QoS),
security, network management, and traffic engineering extremely
challenging, to mention a few.
[0029] FIG. 2 is a schematic block diagram of an example
node/device 200 that may be used with one or more embodiments
described herein, e.g., as any of the computing devices shown in
FIGS. 1A-1B, particularly the PE routers 120, CE routers 110,
nodes/device 10-20, servers 152-154 (e.g., a network controller
located in a data center, etc.), any other computing device that
supports the operations of network 100 (e.g., switches, etc.), or
any of the other devices referenced below. The device 200 may also
be any other suitable type of device depending upon the type of
network architecture in place, such as IoT nodes, etc. Device 200
comprises one or more network interfaces 210, one or more
processors 220, and a memory 240 interconnected by a system bus
250, and is powered by a power supply 260.
[0030] The network interfaces 210 include the mechanical,
electrical, and signaling circuitry for communicating data over
physical links coupled to the network 100. The network interfaces
may be configured to transmit and/or receive data using a variety
of different communication protocols. Notably, a physical network
interface 210 may also be used to implement one or more virtual
network interfaces, such as for virtual private network (VPN)
access, known to those skilled in the art.
[0031] The memory 240 comprises a plurality of storage locations
that are addressable by the processor(s) 220 and the network
interfaces 210 for storing software programs and data structures
associated with the embodiments described herein. The processor 220
may comprise necessary elements or logic adapted to execute the
software programs and manipulate the data structures 245. An
operating system 242 (e.g., the Internetworking Operating System,
or IOS.RTM., of Cisco Systems, Inc., another operating system,
etc.), portions of which are typically resident in memory 240 and
executed by the processor(s), functionally organizes the node by,
inter alia, invoking network operations in support of software
processors and/or services executing on the device. These software
processors and/or services may comprise a training dataset
generator process 248 and/or a traffic flow analyzer process
249.
[0032] It will be apparent to those skilled in the art that other
processor and memory types, including various computer-readable
media, may be used to store and execute program instructions
pertaining to the techniques described herein. Also, while the
description illustrates various processes, it is expressly
contemplated that various processes may be embodied as modules
configured to operate in accordance with the techniques herein
(e.g., according to the functionality of a similar process).
Further, while processes may be shown and/or described separately,
those skilled in the art will appreciate that processes may be
routines or modules within other processes.
[0033] Traffic flow analyzer process 249 includes computer
executable instructions that, when executed by processor(s) 220,
cause device 200 to analyze available information about a traffic
flow to discern the traffic flow type of the flow under analysis.
In some embodiments, traffic flow analyzer process 249 may discern
between different types of benign traffic flows, such as the
various applications associated with the flows. In further
embodiments, traffic flow analyzer process 249 may discern between
benign and malicious traffic flows and may even identify the
specific type of a malicious flow (e.g., the specific family of
malware associated with the flow). Example forms of traffic that
can be caused by malware may include, but are not limited to,
traffic flows reporting exfiltrated data to a remote entity,
spyware or ransomware-related flows, command and control (C2)
traffic that oversees the operation of the deployed malware,
traffic that is part of a network attack, such as a zero day attack
or denial of service (DoS) attack, combinations thereof, or the
like. In further embodiments, traffic flow analyzer process 249 may
analyze traffic flow data to detect anomalous or otherwise
undesirable behaviors (e.g., malfunctioning devices, misconfigured
devices, etc.), traffic pattern changes (e.g., a group of hosts
begin sending significantly more or less traffic), or the like.
[0034] According to various embodiments, traffic flow analyzer
process 249 may employ any number of machine learning techniques,
to assess a given traffic flow in the network. In general, machine
learning is concerned with the design and the development of
techniques that receive empirical data as input (e.g., traffic data
regarding traffic in the network) and recognize complex patterns in
the input data. For example, some machine learning techniques use
an underlying model M, whose parameters are optimized for
minimizing the cost function associated to M, given the input data.
For instance, in the context of classification, the model M may be
a straight line that separates the data into two classes (e.g.,
labels) such that M=a*x+b*y+c and the cost function is a function
of the number of misclassified points. The learning process then
operates by adjusting the parameters a,b,c such that the number of
misclassified points is minimal. After this optimization/learning
phase, traffic flow analyzer process 249 can use the model M to
classify new data points, such as information regarding new traffic
flows in the network. Often, M is a statistical model, and the cost
function is inversely proportional to the likelihood of M, given
the input data.
[0035] In various embodiments, traffic flow analyzer process 249
may employ one or more supervised, unsupervised, or semi-supervised
machine learning models to analyze traffic flow data. Generally,
supervised learning entails the use of a training dataset, such as
a training dataset generated by training dataset generator process
248, which is used to train the model to apply labels to the input
data. For example, the training data may include sample traffic
data that is "normal," or "malware-generated." On the other end of
the spectrum are unsupervised techniques that do not require a
training set of labels. Notably, while a supervised learning model
may look for previously seen attack patterns that have been labeled
as such, an unsupervised model may instead look to whether there
are sudden changes in the behavior of the network traffic.
Semi-supervised learning models take a middle ground approach that
uses a greatly reduced set of labeled training data.
[0036] Example machine learning techniques that traffic flow
analyzer process 249 can employ may include, but are not limited
to, nearest neighbor (NN) techniques (e.g., k-NN models, replicator
NN models, etc.), statistical techniques (e.g., Bayesian networks,
etc.), clustering techniques (e.g., k-means, mean-shift, etc.),
neural networks (e.g., reservoir networks, artificial neural
networks, etc.), support vector machines (SVMs), logistic or other
regression, Markov models or chains, principal component analysis
(PCA) (e.g., for linear models), multi-layer perceptron (MLP) ANNs
(e.g., for non-linear models), replicating reservoir networks
(e.g., for non-linear models, typically for time series), random
forest classification, or the like.
[0037] The performance of a machine learning model can be evaluated
in a number of ways based on the number of true positives, false
positives, true negatives, and/or false negatives of the model. For
example, the false positives of the model may refer to the number
of traffic flows that are incorrectly classified as
malware-generated, anomalous, etc. Conversely, the false negatives
of the model may refer to the number of traffic flows that the
model incorrectly classifies as normal, when actually
malware-generated, anomalous, etc. True negatives and positives may
refer to the number of traffic flows that the model correctly
classifies as normal or malware-generated, etc., respectively.
Related to these measurements are the concepts of recall and
precision. Generally, recall refers to the ratio of true positives
to the sum of true positives and false negatives, which quantifies
the sensitivity of the model. Similarly, precision refers to the
ratio of true positives the sum of true and false positives.
[0038] In some cases, traffic flow analyzer process 249 may assess
the captured traffic data on a per-flow basis. In other
embodiments, traffic flow analyzer process 249 may assess traffic
data for a plurality of traffic flows based on any number of
different conditions. For example, traffic flows may be grouped
based on their sources, destinations, temporal characteristics
(e.g., flows that occur around the same time, etc.), combinations
thereof, or based on any other set of flow characteristics.
[0039] As shown in FIG. 3, various mechanisms can be leveraged to
capture information about traffic in a network. For example,
consider the case in which client node 10 initiates a traffic flow
with remote server 154 that includes any number of packets 302. Any
number of networking devices along the path of the flow may analyze
and assess packet 302, to capture traffic data regarding the
traffic flow. For example, as shown, consider the case of edge
router CE-2 through which the traffic between node 10 and server
154 flows. While the packets 302 flow through edge router CE-2,
router CE-2 may capture traffic data regarding the flow. Notably,
traffic flows can be monitored in many cases through the use of a
tap or Switch Port Analyzer (SPAN).
[0040] In some embodiments, a networking device may analyze packet
headers, to capture information about the traffic flow. For
example, router CE-2 may capture the source address and/or port of
host node 10, the destination address and/or port of server 154,
the protocol(s) used by packet 302, or other header information by
analyzing the header of a packet 302. In further embodiments, the
device may also assess the payload of the packet to capture
information about the traffic flow. For example, router CE-2 or
another device may perform deep packet inspection (DPI) on one or
more of packets 302, to assess the contents of the packet. Doing so
may, for example, yield additional information that can be used to
determine the application associated with the traffic flow (e.g.,
packets 302 were sent by a web browser of node 10, packets 302 were
sent by a videoconferencing application, etc.).
[0041] The networking device that captures the traffic data may
also compute any number of statistics or metrics regarding the
traffic flow. For example, CE-2 may determine the start time, end
time, duration, packet size(s), the distribution of bytes within a
flow, etc., associated with the traffic flow by observing packets
302.
[0042] In the specific case of encrypted traffic, the networking
device that captures the traffic data may also be a proxy device,
in some embodiments. For example, CE-2 or another intermediary
device may act as a man-in-the-middle between endpoints, to
intercept the encryption credentials used and simulate responses
from each respective node. In doing so, the device may be able to
decrypt and analyze the payloads of the packets. Alternatively, in
further embodiments, the device may simply capture header
information from encrypted traffic, such as Transport Layer
Security (TLS) header information.
[0043] As noted above, captured traffic flow data can be used to
form a training dataset for a machine learning-based traffic flow
analyzer, such as traffic flow analyzer process 249. However, it is
often the case that a training dataset based on the full set of
captured traffic data is too unwieldy for many systems. Notably, in
many cases, the training dataset must be small enough to reside in
memory during the training of the machine learning process.
[0044] It has also been discovered through preliminary testing that
captured traffic flow data is often redundant for purposes of
training a machine learning-based traffic analyzer. For example, an
experiment was conducted over a four day period and collected
traffic flow data regarding approximately 50 million observed
flows. However, due to the redundant and/or similar information in
the captured traffic flow data, it was discovered that a subset of
the traffic flow data for approximately only 150,000 flows
sufficiently represents the entirety of the .about.50 million flows
when generating a training dataset.
Training a Machine Learning-Based Network Traffic Analyzer Using a
Prototype Dataset
[0045] The techniques herein allow for the generation of a
"prototype" training dataset for a machine learning-based traffic
analyzer by reducing redundancies in the captured traffic flow data
used to form the training dataset. Doing so allows for the more
efficient training of machine learning processes, creation of more
robust traffic analyzers, reduction of storage requirements, and
more efficient transport of the datasets via a network. In some
aspects, the techniques herein compute a similarity score between a
newly observed feature vector and those already stored in a
training dataset dictionary, to determine whether or not to add the
new feature vector to the dictionary. In further aspects, the
techniques may also maintain counts for the feature vectors in the
dictionary, to better represent the traffic load in the network.
These counts can also be leveraged to generate a training dataset
that is adapted for a particular target network (e.g., if a certain
type of traffic is not present in the target network, feature
vectors for this traffic type can be excluded from the training
dataset, etc.).
[0046] Specifically, according to one or more embodiments of the
disclosure as described in detail below, a device in a network
generates a feature vector based on traffic flow data regarding one
or more traffic flows in the network. The device makes a
determination as to whether the generated feature vector is already
represented in a training dataset dictionary by one or more feature
vectors in the dictionary. The device updates the training dataset
dictionary based on the determination by one of: adding the
generated feature vector to the dictionary when the generated
feature vector is not already represented by one or more feature
vectors in the dictionary, or incrementing a count associated with
a particular feature vector in the dictionary when the generated
feature vector is already represented by the particular feature
vector in the dictionary. The device generates a training dataset
based on the training dataset dictionary for training a machine
learning-based traffic flow analyzer.
[0047] Illustratively, the techniques described herein may be
performed by hardware, software, and/or firmware, such as in
accordance with the training dataset generator process 248, which
may include computer executable instructions executed by the
processor 220 (or independent processor of interfaces 210) to
perform functions relating to the techniques described herein,
e.g., in conjunction with traffic flow analyzer process 249.
[0048] Operationally, FIG. 4 illustrates an example architecture
400 for generating a training dataset for a traffic flow analyzer,
according to various embodiments. As shown, training dataset
generator process 248 may include any number of sub-processes
and/or access any number of storage locations, to generate a
training dataset 424 for use to train traffic flow analyzer process
249. As would be appreciated, the various elements of architecture
400 may be implemented on a single device (e.g., a single device
200) or, alternatively, in a distributed manner across multiple
devices. For example, in some cases, the device executing training
dataset generator process 248 may send training dataset 424 via a
network to another device configured to train the machine learning
model(s) of traffic flow analyzer process 249. Further, while
specific sub-processes and memory locations are shown in
architecture 400, their respective functionalities can be combined
or incorporated into other elements, as desired.
[0049] Training dataset generator process 248 may receive as input
captured traffic flow data 402. In some cases, the device executing
training dataset generator process 248 may also operate to capture
traffic flow data 402. In other cases, the device executing
training dataset generator process 248 may receive traffic flow
data 402 from one or more other capturing devices in the network.
As noted previously, traffic flow data 402 may include any or all
information available regarding the traffic flows present in the
network. For example, traffic flow data 402 may indicate the
source/destination addresses of a flow, the ports in use, the
protocols in use, header information such as TLS information,
statistics or other computed metrics (e.g., in terms of packet
timing, packet sizes, etc.), payload information if available, or
any other information available from the observation of traffic
flows in the network. Traffic flow data 402 may also comprise
labels for the various flows, if known, such as benign vs.
malicious, an application type associated with a given flow,
etc.
[0050] Training dataset generator process 248 may execute a feature
vector constructor 404 configured to construct an observed feature
vector 406 for a particular observed traffic flow based on the
information in captured traffic flow data 402. In general, observed
feature vector 406 may be an n-dimensional set of
measurements/observations from traffic flow data 402 regarding a
particular observed traffic flow. For example, one dimension of
feature vector 406 may indicate whether or not the particular
traffic flow uses TCP or UDP, another dimension of feature vector
406 may indicate the ciphersuite associated with the traffic flow,
etc.
[0051] In various embodiments, training dataset generator process
248 may maintain a training dataset dictionary 410. In general,
training dataset dictionary 410 comprises a set of one or more
feature vectors 412 constructed from captured traffic flow data 402
(e.g., by feature vector constructor 404). In addition, in some
embodiments, training dataset dictionary 410 may include counts 414
that are associated with the feature vectors 412 in dictionary 410.
Thus, training dataset dictionary 410 may comprise not only the
feature vectors that are representative of the observed traffic in
the network, but also information regarding the volume of the
traffic represented by the various feature vectors. Said
differently, counts 414 may provide a measure of how often or
frequently traffic represented by a given feature vector 412 was
observed in the network.
[0052] To determine whether observed feature vector 406 is already
represented by one or more of feature vectors 412 in training
dataset dictionary 410, training dataset generator process 248 may
execute a similarity analyzer 408. In some embodiments, similarity
analyzer 408 may compute one or more similarity scores 416 that
represent how similar observed feature vector 406 is to any of the
feature vectors 412 in training dataset dictionary 410. For
example, a similarity score 416 may be a standard squared
exponential or other measurement of the similarity between observed
feature vector 406 and one of the feature vectors 412 in training
dataset dictionary 410.
[0053] In further embodiments, similarity analyzer 408 may compute
any form of function value (e.g., similarity score, analyzed rule,
etc.) to control whether observed feature vector 406 is added to
training dataset dictionary 410. Such a function value may be
computed in addition to, or in lieu of, similarity scores 416,
which are shown for illustrative purposes only. For example, in
further embodiments, similarity analyzer 408 may compute one or
more function values based on one or more features of feature
vector 406. Doing so may allow training dataset generator process
248 to purposely add feature vector 406 to training dataset
dictionary 410 when feature vector 406 has a predefined subset of
desired features (e.g., a specific operating system, a specific
crypto library, etc.). Conversely, such a function value may also
be used to filter out feature vectors that do not have the desired
characteristics for the training dataset.
[0054] In some embodiments, training dataset generator process 248
may also execute a dictionary updater 418 to update training
dataset dictionary 410 based on observed feature vector 406 and its
corresponding similarity score(s) 416 or other function values. For
example, if any of similarity scores 416 for observed feature
vector 406 are above a predefined threshold, dictionary updater 418
may determine that observed feature vector 406 is already
represented in training dataset dictionary 410. Conversely, if the
similarity scores 416 for observed feature vector 406 are below a
defined threshold, dictionary updater 418 may determine that
observed feature vector 406 is not represented in training dataset
dictionary 410.
[0055] If dictionary updater 418 determines that observed feature
vector 406 is already represented in training dataset dictionary
410, dictionary updater 418 may simply update the counts 414 for
one or more of the feature vectors 412 that are deemed similar to
observed feature vector 406. In some cases, dictionary updater 418
may increment the count 414 associated with the most similar
feature vector 412 to observed feature vector 406 (e.g., the
feature vector 412 with the highest similarity score 416). In other
cases, dictionary updater 418 may increment the count 414
associated with each of the feature vectors 412 that have
corresponding similarity scores 416 above the predefined similarity
threshold or a subset thereof (e.g., the top two most similar, the
top three most similar, etc.). In doing so, training dataset
dictionary 410 not only maintains a reduced representation of all
of the observed traffic flows, but also a measure of how often a
particular feature vector 412 represented an observed traffic flow
in the network.
[0056] If dictionary updater 418 determines that observed feature
vector 406 is not already represented by any of feature vectors 412
in training dataset dictionary 410, dictionary updater 418 may add
observed feature vector 406 to feature vectors 412. In addition, in
some embodiments, dictionary updater 418 may also initialize a
corresponding count 414 for added feature vector 406 (e.g., by
setting the count to `1`, etc.).
[0057] In various embodiments, training dataset generator process
248 may execute a training dataset customizer 420 configured to
generate a training dataset 424 for training the machine learning
model of traffic flow analyzer process 249. In general, given the
network flow feature vectors and their prevalence from training
dataset dictionary 410, training dataset customizer 420 can
efficiently train a classifier or other machine learning mechanism
by assigning weights to the feature vectors in training dataset 424
based on counts 414. In addition, if any labels are available in
training dataset dictionary 410 (e.g., if feature vectors 412 are
known to be benign or malicious, etc.), these labels may also be
included in training dataset 424.
[0058] In some embodiments, training dataset customizer 420 may
adjust the weights/counts of the feature vectors in training
dataset 424 based on target system parameters 422, allowing for
training dataset 424 to vary from the characteristics actually
observed in captured traffic flow data 402. In some embodiments,
parameters 422 may indicate the type of classifier or other machine
learning mechanism to be trained for traffic flow analyzer process
249. Notably, some machine learning mechanisms, such as stochastic
gradient descent (SGD) classifiers, can overfit to data samples
that have high prevalence. In such cases, training dataset
customizer 420 may enforce a maximum weight and/or count for any of
the feature vectors 412 included in training dataset 424. For
example, training a classifier on 10 million flows to google.com
would not be productive for a simple perceptron model.
[0059] Target system parameters 422 may also indicate the types
and/or volumes of traffic observed in the target network to which
traffic flow analyzer 249 is to be deployed. In such cases,
training dataset customizer 420 may adjust the weights/counts of
the feature vectors 412 included in training dataset 424,
accordingly. For example, assume that traffic flow analyzer process
249 is to be deployed to a network in which traffic to facebook.com
is blocked, but that certain feature vectors 412 in training
dataset dictionary 410 relate to facebook.com traffic. In such a
case, training dataset customizer 420 may treat these feature
vectors as having a count and/or weight of zero, thereby
effectively excluding these feature vectors 412 in training dataset
dictionary 410 from the generated training dataset 424.
[0060] To test the efficacy of the techniques herein, traffic flow
data for 1 million TLS encrypted flows from an enterprise network
and 250,000 malicious TLS flows from ThreatGRID.TM. by Cisco
Systems, Inc. were captured. Example data features that were
collected included sequence of packet length and packet
inter-arrival time (SPLT) information, byte distribution
information, unencrypted TLS header information (e.g., offered
ciphersuites, TLS extensions, etc.), and the like. The size of the
compressed JavaScript Object Notation (JSON) for the resulting
training dataset from this traffic flow data was 4.5 Gigabytes.
[0061] To determine flow similarity, a standard squared exponential
with lambda=1.0 and a threshold of 0.99 was used. As a result, the
filtered dataset included feature vectors for approximately 17,000
enterprise flows and approximately 11,000 malicious flows. The size
of the compressed JSON for the resulting "filtered" training data
was only 103 Megabytes, which is orders of magnitude smaller than
that of the full set.
[0062] The filtered and unfiltered training datasets were then used
to train a random forest classifier and a validation dataset
comprising 100,000 malicious TLS flows and 1,000,000 enterprise TLS
flows was used to compare the performance of the two classifiers.
In terms of training time, it was noted that training the
classifier using the unfiltered dataset was approximately 300
seconds compared to approximately 2 seconds for the filtered
dataset. It is important to note that random forests are
particularly efficient and this improvement in training times will
be more pronounced for other machine learning processes, such as
those that are kernel-based or based on backpropagation.
[0063] Total accuracy for the classifier trained using the
unfiltered training dataset was observed to be 99.7% on the
validation set and at a 1-in-10,000 false detection rate (FDR) was
89.3%. In contrast, total accuracy for the classifier trained using
the filtered training dataset was observed to be 99.6% on the
validation set, and the accuracy at a given 1-in-10,000 FDR was
91.3%, also an improvement over the unfiltered dataset. This is
believed to be due to the original unfiltered dataset having TLS
encrypted flows to content delivery networks (CDNs) that were also
in the enterprise dataset. By combining these flows as part of the
filtering, the resulting classifier was more robust to these
outliers.
[0064] FIG. 5 illustrates an example simplified procedure for
generating a training dataset for a traffic flow analyzer, in
accordance with one or more embodiments described herein. For
example, a non-generic, specifically configured device (e.g.,
device 200) may perform procedure 500 by executing stored
instructions (e.g., process 248 and/or 249). The procedure 500 may
start at step 505, and continues to step 510, where, as described
in greater detail above, the device may generate a feature vector
based on traffic flow data regarding one or more traffic flows in
the network.
[0065] At step 515, as detailed above, the device may make a
determination as to whether the generated feature vector is already
represented in a training dataset dictionary by one or more feature
vectors in the dictionary. In some embodiments, the device may
compute one or more similarity scores between the generated feature
vector and those in the dictionary. For example, the device may use
a squared exponential function with a predefined threshold, to
determine whether the generated feature vector is similar enough to
a feature vector in the dictionary so as to be considered already
represented.
[0066] At step 520, the device may update the training dataset
dictionary based on the determination from step 515, as described
in greater detail above. In some embodiments, if the feature vector
generated in step 510 is not represented in the dictionary, the
device may add the feature vector to the dictionary. In addition,
the device may initialize a count for the added feature vector.
However, in further embodiments, if the device determines that the
feature vector is already represented by a particular feature
vector in the dictionary, the device may simply increment a count
associated with the particular feature vector in the
dictionary.
[0067] At step 525, as detailed above, the device may generate a
training dataset based on the training dataset dictionary for
training a machine learning-based traffic flow analyzer. In some
embodiments, the training dataset may include weightings for the
included feature vectors from the dictionary that are based on
their corresponding counts in the dictionary and/or one or more
parameters for the target system. For example, the device may
adjust the counts/weights for a particular traffic type based on
the expected volume of traffic of that type in the target network
(e.g., to exclude certain feature vectors from the training dataset
if that type of traffic is not expected to exist, etc.). In further
cases, the device may impose a maximum weighting based on the type
of traffic flow analyzer, so as not to over-represent a given
sample in the training dataset. Procedure 500 then ends at step
530.
[0068] It should be noted that while certain steps within procedure
500 may be optional as described above, the steps shown in FIG. 5
are merely examples for illustration, and certain other steps may
be included or excluded as desired. Further, while a particular
order of the steps is shown, this ordering is merely illustrative,
and any suitable arrangement of the steps may be utilized without
departing from the scope of the embodiments herein.
[0069] The techniques described herein, therefore, allow for
various efficiency gains to be made in the learning phase of a
traffic flow analyzer. In further aspects, the techniques herein
allow for the adjustment of diversity metrics, such as by reducing
sample weights for over-represented samples in epoch-based
processes. Additionally, the techniques herein allow for less
expensive storage and transmission of training datasets in terms of
resource consumption. Further, with the data reduction techniques
presented herein, this also allows for the possibility of deploying
machine learning processes that require the training dataset to be
present on-box, such as k-NN processes or dual-SVM processes on a
router or other networking device.
[0070] While there have been shown and described illustrative
embodiments that provide for the training of a traffic flow
analyzer using a prototype training dataset, it is to be understood
that various other adaptations and modifications may be made within
the spirit and scope of the embodiments herein. For example, while
certain embodiments are described herein with respect to using
certain machine learning models, the models are not limited as such
and may be used for other functions, in other embodiments. In
addition, while certain protocols are shown, other suitable
protocols may be used, accordingly.
[0071] The foregoing description has been directed to specific
embodiments. It will be apparent, however, that other variations
and modifications may be made to the described embodiments, with
the attainment of some or all of their advantages. For instance, it
is expressly contemplated that the components and/or elements
described herein can be implemented as software being stored on a
tangible (non-transitory) computer-readable medium (e.g.,
disks/CDs/RAM/EEPROM/etc.) having program instructions executing on
a computer, hardware, firmware, or a combination thereof.
Accordingly this description is to be taken only by way of example
and not to otherwise limit the scope of the embodiments herein.
Therefore, it is the object of the appended claims to cover all
such variations and modifications as come within the true spirit
and scope of the embodiments herein.
* * * * *