Izabela Savić, a current MSc in the School of Computer Science, won Best Student Paper Award for their paper, Adversarial Sampling Attacks and Defense in DNS Data Exfiltration at the 4th International Conference on Emerging Information Security and Applications [1] (EISA 2023). EISA 2023 was recently hosted on December 6th and 7th in Hangzhou, China.
Cybersecurity continues to be a hot topic in technology as the number and cost of cybercrimes continues to significantly increase worldwide. The average cost of a cyber breach may be significantly lowered if artificial intelligence or machine learning is used as part of a company’s security framework. However, machine learning is vulnerable to a unique kind of attack called adversarial attacks, an attack created by creating slight modifications to a malicious input.
According to Izabela, “this work verifies that this attack can be applied successfully to new areas such as network packets, specifically data exfiltration network packets. The crafted adversarial DNS data exfiltration packets are extremely successful at bypassing traditional machine learning detection methods. To counter this, we introduce the idea of using multiple machine learning models with different architectures together, a tactic called ensemble learning. By using an ensemble learning method the adversarial examples are significantly less effective.
My research provides a foundation to improving machine learning model robustness against adversarial attacks, which in turn increases the security of industries which use machine learning in their security framework (e.g. government, hydro, military defense, and private companies).”
Congratulations on your achievement, Izabela!