Artificial Intelligence and Machine Learning 101: White Paper
Artificial Intelligence and Machine Learning 101: White Paper
Let’s break this down into a system diagram. In the figure below, the three general steps
of human intelligence from left to right: input, processing, and output. In the human brain,
input takes place in the form of sensing and perceiving things. Your eyes, nose, ears, etc.,
take in raw input on the left, such as photons of light or the smell of pine trees, and then
process it. On the system’s right side is output. This includes speech and actions, both of
which are dependent on how we process the raw input that our brain is receiving.
The processing happens in the middle, where knowledge or memories are formed
Human Intelligence
and retrieved, decisions and inferences and made, and learning occurs.
Knowledge
&
Memory
Decision
&
Sensing & Inference Output &
Perception Interaction
Learning
1
Artificial Intelligence and Machine Learning 101
Picture stopping at a roadway intersection. Your eyes see that the traffic light in front of To intelligently
you has just turned green. Based on what you have learned from experience (and driver’s navigate the world
education), you know that a green light indicates that you should drive forward. So, you hit around us—answering
the gas pedal. The green light is the raw input, your acceleration is the output; everything in the phone, baking
between is processing. chocolate chip cookies,
or obeying traffic
To intelligently navigate the world around us—answering the phone, baking chocolate chip lights—we need to
cookies, or obeying traffic lights—we need to process the input that we receive. This is process the input
the core of human intelligence processing, and it is ultimately broken down into three that we receive.
distinct aspects:
1. K
nowledge and memory. We build up knowledge as we ingest facts (i.e., the Battle of
Hastings took place in 1066) and social norms (i.e., saying “Please” and “Thank you” is
considered polite). Additionally, memory enables us to recall and apply information from
the past to present situations. For example, Edward remembers that Jane did not thank
him for her birthday present, so he does not expect her to thank him when he gives her
a Christmas present.
2. Decision and inference. Decisions and inferences are made based on raw input combined
with knowledge and/or memory. For example, Edward ate a jalapeno pepper last year and
did not like it. When Johnny offers a pepper to Edward, he decides not to eat it.
3. Learning. Humans can learn by example, observation, or algorithm. In learning by example,
we are told that one animal is a dog, the other is a cat. In learning by observation, we figure
out on our own that dogs bark and that cats meow. The third learning method—algorithm—
enables us to complete a task by following a series of steps or a specific algorithm (e.g.,
performing long division).
These aspects of human intelligence parallel artificial intelligence. Just as we take in
information, process it, and share output, so can machines. Let’s take a look at the figure
Artificial Intelligence
below to see how this maps out.
Knowledge
Representation,
Knowledge
Ontologies, Graph
&
Databases, etc.
Memory
NLP Robotics
Prescriptive
Speech Analytics, Decision Navigation
Recognition Optimization, & Systems
Decision Inference
Making, etc.
Visual Speech
Recognition Generation
Data Threat
Machine Learning
Sources (supervised, unsupervised) Learning Leads
2
Artificial Intelligence and Machine Learning 101
In machines, the input part of artificial intelligence is exemplified by natural language Ideally, the most
processing, speech recognition, visual recognition, and more. You see such technologies and accurate and efficient
algorithms everywhere, from self-driving cars that need to sense the roadways and obstacles, artificial intelligence
to Alexa or Siri when it recognizes your speech. The output that follows are ways in which results require a
machines interact with the world around us. This might take the form of robotics, navigation combination of
systems (to guide those self-driving cars), speech generation (e.g., Siri), etc. In between, learning methods.
we have various forms of processing that takes place. Both supervised
and unsupervised
Similar to our accrual of knowledge and memories, machines can create knowledge machine learning
representations (e.g., graph databases, ontologies) that help them store information about the are useful methods—
world. Just as humans make decisions or draw inferences, machines can make a prediction, it’s all about applying
optimize for a target or outcome, and determine the best next steps or decisions to meet a the right approach
specific goal. or approaches to
the right use case.
Finally, just as we learn by example, observation, or algorithm, machines can be taught
using analogous methods. Supervised machine learning is much like learning by example:
the computer is given a dataset with “labels” within the data set that act as answers, and
eventually learns to tell the difference between different labels (e.g., this dataset contains
photos labeled as either “dog” or “cat”, and with enough examples, the computer will notice
that dogs generally have longer tails and less pointy ears than cats).
Ideally, the most accurate and efficient artificial intelligence results require a combination of
learning methods. Both supervised and unsupervised machine learning are useful methods—
it’s all about applying the right approach or approaches to the right use case.
Next, we’ll put machine learning under the microscope to understand how this part of AI
mirrors the neurons in our brain to turn input into to optimal output.
3
Artificial Intelligence and Machine Learning 101
Neural networks—a programming paradigm in which we train machines to “learn”—are Much of what we think
inspired by neurons, or specialized cells in the human body that form the foundation of our of as human learning
nervous system, and brains in particular. These cells transmit signals throughout our bodies can be described
trigger nervous system responses and processes. Neurons are what enable us to see, hear, by how strong the
smell, etc. connection between
two neurons in our
Dendrites
brain is, along with
the strength of the
Synapse
firing of our synapses.
Nucleus
Axon
In part one of this guide, we discussed the basic process of human intelligence: input on
the left, and output on the right. The neuron (pictured above) plays a critical role in this.
On the left side of the neuron, the cell body collects “input.” Once it receives enough input
or stimulation, the axon fires, transmitting the information to the right side—the synapse.
The “output” is then sent to other neurons.
At any given moment, our neurons are passing messages between each other. These cells
are responsible for our ability to perceive our surroundings. And when we learn, our neurons
become very active. In fact, much of what we think of as human learning can be described by
how strong the connection between two neurons in our brain is, along with the strength of
the firing of our synapses.
Each circular node represents an artificial, biologically inspired “neuron.” The lines represent
a connection from the output of one artificial neuron on the left to the input of another on the
right. Signals between these neurons flow along the lines from left to right. In these networks,
input—such as pixel data—flows from the input layer, through the middle “hidden” layers,
and ultimately to the output layer in a manner described by mathematical equations loosely
inspired by the electrical activity in actual biological neurons.
4
Artificial Intelligence and Machine Learning 101
Input Output
These “deeper” neural networks can do much more complex predictions. There can be
thousands of nodes and hundreds of layers, which means thousands of different calculations.
Deep learning models have become very good at specific problems, such as speech or
image recognition.
5
Artificial Intelligence and Machine Learning 101
It’s worth noting, however, that deep learning is not a silver bullet for machine learning— Neural networks might
especially not in cybersecurity, where sometimes there is not the large volume of clean seem like the stuff
data that is ideal for deep learning methods. It is important to pick the right algorithm, data, of the future, but it’s
and principles for the job. This is the best way for machines to gather evidence, connect the been around for a
dots, and draw a conclusion. while. In fact, neural
networks are based
Neural networks might seem like the stuff of the future, but it’s been around for a while. on ideas that started
In fact, neural networks are based on ideas that started circulating back in the 1940s. In the circulating back
next section, we will take a short trip back in time to understand how neural networks and in the 1940s.
machine learning have come to permeate many parts of modern life.
Although not referred to as such, the idea of artificial intelligence can be traced back to
antiquity (i.e., Greek god Hephaestus’s talking mechanical handmaidens).¹ Since the 1930s,
scientists and mathematicians alike have been eager to explore creating true intelligence
separate from humans.
AI’s defining moment in the mid-20th century was a happy confluence of math and biology,
with researchers like Norbert Wiener, Claude Shannon, and Alan Turing having already chipped
away at the intersection of electrical signals and computation. By 1943, Warren McCulloch
and Walter Pitts had created a model for neural networks. Neural networks paved the way
for a brave new world of computing with greater horsepower, and, in 1956, the field of AI
research was officially established as an academic discipline.
The latter half of the century was an exciting age for AI research and progress, interrupted
occasionally by “AI winters” in the mid-70s and late 80s where AI failed to meet public
expectations, and investment in the field was reduced. But despite setbacks, different appli
cations for AI and machine learning were appearing left and right. One particular anecdote
of such an application has become a popular parable within the scientific community,
speaking quite effectively to the trials and tribulations of AI research and implementation.
In the 1980s, the Pentagon decided to use a neural network to identify camouflaged tanks.
Working with just one mainframe (from the 1980s, keep in mind), the neural net was trained
with 200 pictures—100 tanks and 100 trees. Despite the relatively small neural network
(due to 1980’s limitations on computation and memory), the lab training resulted in 100%
accuracy. With such success, the team decides to give it a go out in the field. The results
were not great.
6
Artificial Intelligence and Machine Learning 101
Eventually, though, visual recognition via deep learning—facilitated by neural networks that
are much more complex than the Pentagon’s 1980s mainframe would have been able to
handle—became a reality. In 2012, Stanford professor Andrew Ng and Google fellow Jeff
Dean created one of the first deep neural networks using 1000 computers with 16 cores
each. The task: analyze 10 million YouTube videos. The result: it found cats.² Thanks to its
“deep learning” algorithm, the network was able to recognize cats over time, and with very
good accuracy.
With the availability of vast computing resources that were undreamed of back in the 1980’s,
deep neural networks have quickly become a popular area for research. Deep learning
gives a system the ability to automatically “learn” through billions of combinations and
observations, reducing the dependency on human resources. Within the cybersecurity
domain, the method has become particularly promising for detecting malware—scenarios in
which we have large datasets with many examples of malware from which the network
can learn.
Unfortunately, deep learning methods are currently less effective when it comes to certain
use cases, like insider threat, because we simply don’t have the right kind of data on these
types of attacks, in the volumes required. Most often, the information we have on insider
threats are anecdotal, which cannot be used efficiently by these types of neural networks.
7
Artificial Intelligence and Machine Learning 101
Until we can gather more effective datasets (and reduce the cost and complexity of deep We have seen
learning systems), deep learning is not the right choice for all use cases. And that’s okay. immense potential
Deep learning is just one of many machine learning algorithms, and these approaches can of AI technologies in
be just as if not more valuable—it all depends on the job at hand. the six decades since
its official “birth,” and
We have seen immense potential of AI technologies in the six decades since its official we have only just
“birth,” and we have only just scratched the surface, especially in security. Next, we will take scratched the surface,
a deeper dive into the potential applications for AI and analytics to change the way that we especially in security.
identify and respond to security threats.
Predictive analytics is just one piece of a much larger puzzle that can give us
much more useful insight for security teams.
We have seen what analytics can do for other industries, and there is potential for analytics
to have a profound impact on cybersecurity, too. We see this taking shape in a new field that
we refer to as security analytics, which essentially takes the battle-tested algorithms and
methodologies that we have discussed (and more) and applies them help solve the really
difficult problems in security.
8
Artificial Intelligence and Machine Learning 101
The most common analytics we see in security today involves predictive models, which allow We have more data
us to identify where risks might be within large amounts of data (this is where anomaly today than ever
detection fits in). In a nutshell, predictive modeling combines historical data with real-time before. Thankfully,
behavior to understand or predict future behavior. With this, we can answer the question, we also have more
“What happens next?” compute power,
better algorithms, and
But our vision for security analytics doesn’t stop here. Predictive analytics is just one piece of broader investment
a much larger puzzle that can give us much more useful insight for security teams. The ideal in research and
analytics paradigm combines intelligent sensor and ubiquitous data sources—desktops and technologies to help
servers, mobile, cloud, social networks, open data, etc.—with multiple advanced analytical us make sense of
approaches to behavioral and threat analysis, including forensic analysis, risk modeling, this data through
anomaly detection, behavioral and response optimization, and more. mathematics. By all
accounts, we believe
This means that we can do far more than predict or identify a threat. It allows us to go even security analytics is
further to offer not just advanced detection but insight into how to respond most effectively. just getting started.
Security analytics gives us the power to answer other key questions, like “How many threats
are there?” and “What is the best possible reaction?”
We haven’t seen other classes of analytics like optimization methods applied to cybersecurity
yet, but they have immense potential. These techniques look at all the possible reactions to a
security risk and determining the best response. Yes, there are ways to do this with math.
For example, optimization methods are used when you place a call to your cell phone service
provider with an issue. They are not randomly making a recommendation on whether or
not to upgrade your service plan at a discount; they rely on a set of mathematics in the
background that looks at your call logs, the number of dropped calls, how your history
compares with that of other users, etc. It even calculates the probability that you might
switch to another service provider. Then, out of all the possible next steps, it computes
the best next step to maximize customer retention.
The same math can be applied to a security team to identify a risk, provide a number of ways
in which to react, and determine mathematically the best response to maximize containment
of this particular risk.
The rapid rise and evolution of security threats make this type of response efficiency critical.
We have more data today than ever before. Thankfully, we also have more compute power,
better algorithms, and broader investment in research and technologies to help us make
sense of this data through mathematics. By all accounts, we believe security analytics is
just getting started.
9
Contact us at CyberRes.com
Like what you read? Share it.
762-000016-001 | M | 04/21 | © 2021 Micro Focus or one of its affiliates. Micro Focus and the Micro Focus logo, among others,
are trademarks or registered trademarks of Micro Focus or its subsidiaries or affiliated companies in the United Kingdom, United States
and other countries. All other marks are the property of their respective owners.