Deep Learning for Cybersecurity: Assembly Code and Authorship Analytics

点击翻译按钮获取中文摘要

E-commerce sales in Canada are expected to reach $40 billion by 2018.1 Approximately 75% of Canadians use Internet banking.2 Clearly, the Internet and many online information systems have become part of the critical infrastructure of Canada. However, statistics also show that Canadians individuals, industry, and government are not well-prepared for new waves of cyber threats. Deep learning is a major advancement of artificial neural network (ANN), which is a machine learning framework inspired by how the human brain processes information. Due to improvement in high-performance computing, availability of huge volumes of data, and ANN, deep learning has recently shown many promising breakthroughs in multiple areas, from playing chess to self-driving cars. How can we utilize the advancement of deep learning to strengthen the security of our cyberspace and data privacy?The long-term objective of this research program is to enhance the large-scale data analytic capabilities of the cybersecurity and privacy communities so that security professionals can efficiently respond to security incidents and data custodians can effectively protect their clients' data privacy. In this 5-year research program I will focus on the following two short-term objectives.The first objective is to enhance the cybersecurity professionals' deep learning capability on assembly code analytics and provide them a new generation of software reverse engineering methods and tools to efficiently understand the inner workings of benign software and malware binaries. Specifically, the team will develop an assembly code search engine and a description generator using deep learning technology. The applicant closely collaborates with cybersecurity professionals in both public and private sectors; the research results will directly enhance the security of Canadian cyberspace. The projects will be open source; therefore, the results will also benefit the software reverse engineering and machine-learning communities.The second objective is to provide data custodians or individual Internet users with the capability of releasing digital textual documents, e.g., product reviews, opinion articles, blogs, etc., without compromising the identity of the authors due to their digital writing styles. The deliverable will include an open-source privacy-preserving text paraphrasing engine. The research result will enhance the anonymity of users in social media, which in turn promotes web freedom and the fight against social media censorship. The research result will also provide data custodians with an addition layer of privacy protection when they share their administered textual data. This will also indirectly contribute to the open data movement. 1 www.pfsweb.com/blog/2016-canada-ecommerce-market2 www.cba.ca/technology-and-banking