A Deep Learning Approach to Fast, Format-Agnostic Detection of Malicious Web Content

Explore a cutting-edge deep learning approach for detecting malicious web content in this 15-minute IEEE conference talk. Discover how this innovative method operates directly on language-agnostic token streams extracted from static HTML files, offering speed and efficiency suitable for high-frequency data contexts like firewalls and web proxies. Learn about the hierarchical spatial scale examination technique that captures locality and provides superior accuracy compared to traditional bag-of-words models. Gain insights into the impressive 97.5% detection rate at a 0.1% false positive rate, and the ability to classify over 100 small-batched web pages per second on commodity hardware. Understand the potential applications of this fast and accurate approach for deployment to endpoints, firewalls, and web proxies in the fight against malevolent web pages.

A Deep Learning Approach to Fast, Format-agnostic Detection of Malicious Web Content