Using a machine learning approach based on a technique known as reinforcement learning an adaptive cyber defense against these attacks is now in development.
A moving target defense (MTD) is an adaptive machine learning-driven method to address current limitations to detect and respond to cyber-attacks, said Minghui Zhu, Penn State University associate professor of electrical engineering and computer science and Institute for Computational and Data Sciences co-hire.
Typical responses to an attack can take up to 15 days, which can use up significant funds and resources for an organization, according to the researchers,
Zhu said Zero Day attacks are among the most dangerous threats to computer systems and can cause serious and lasting damage. As an example, the WannaCry ransomware attack, which occurred in May 2017, targeted more than 200,000 Windows computers across 150 countries, and caused an estimated $4 billion to $8 billion worth of damage.
The team’s approach relies on reinforcement learning, which, along with supervised and unsupervised learning, is one of the three main machine learning paradigms.
Reinforcement learning is a way that a decision maker can learn to make the right choices by selecting actions that can maximize rewards by balancing exploitation — leveraging past experiences — and exploration — trying new actions, said Peng Liu, the Raymond G. Tronzo, MD Professor of Cybersecurity in the Penn State College of Information Sciences and Technology.

A machine learning algorithm may give organizations a powerful and cost-effective tool for defending against Zero Day attacks on vulnerable computer networks and cyber-infrastructure.
Source: PIXAHIVE
“The decision maker learns optimal policies or actions through continuous interactions with an underlying environment, which is partially unknown,” Liu said. “So, reinforcement learning is particularly well-suited to defend against Zero Day attacks when critical information — the targets of the attacks and the locations of the vulnerabilities — is not available.”
The researchers tested their reinforcement learning algorithm in a 10-machine network. They added although a 10-computer network may not seem very large, it is actually more than robust enough for the test. The setup also included web and mail servers, a Gateway server, SQL server, DNS server and admin server. A firewall ended up installed to prevent access to the internal hosts. The researchers also selected vulnerabilities that could produce multiple attack scenarios for the test.
The researchers added there is room for further improvement for their approach. For example, their algorithm relies on model-free reinforcement learning, which requires a large amount of data or a large number of iterations to learn a relatively good defense policy. In the future they would like to incorporate model-based approaches to accelerate the learning process.
Click here to view more information on the researchers’ findings in ACM Transactions on Privacy and Security.

