Dissertation Proposal
In Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy
Avisha Das
will defend her proposal
Proactive Defense through Automated Generation of Targeted Attacks
Abstract
Social engineering attacks like phishing, email masquerading, where a perpetrator impersonates a legitimate entity, are a major security threat. However, despite having a higher probability of success, executing such an attack is costly in terms of time and effort. With the advancements in machine learning and natural language processing techniques, attackers can resort to using sophisticated methods to evade detection. Such advanced learning methods include both off-the-shelf generation architectures as well as massively pre-trained language models available publicly today. In a proactive scenario, we presume that attackers would resort to automated methods of attack vector generation. Such a study will not only simulate newer attack scenarios but also equip defenders with tools and resources to better train and update their detection methods. However, the application of automated text generation methods in email generation is fairly challenging, owing to the presence of noise in emails and the diversity in email writing styles. First, we apply state-of-the-art language models to study their effectiveness in content generation and use existing automated metrics to evaluate linguistic quality. We then propose a set of novel techniques to automatically control the coherency and the deceptive intent in the generated output. Additionally, we propose a new set of evaluation metrics and study their efficacy in automatically discerning the quality of generated content in a reference-less setting.Date: Wednesday, December 04, 2019
Time: 11:00 AM - 1:00 PM
Place: PGH 550
Advisor: Dr. Rakesh Verma
Faculty, students, and the general public are invited.