Abstract: Applied Natural Language Processing for cybersecurity: Taming the bleeding edge language models for practical security use cases
We have new and powerful natural language models cropping up almost every month, each with more than a billion parameters, capable of numerous open-ended human-like language tasks – like conjuring up crazy concoctions of realistic images from unrealistic human descriptions. But how can they be useful for practical purposes in the cybersecurity domain? Have we solved all the low hanging fruits related to existing security bottlenecks and automation of all kinds of security events analysis?
Following questions will be proposed to the panel for discussion:
1. Large Language Models – A new Moore’s Law? Can multibillion parameter models be finally used for practical infosec use cases?
2. Have we explored tried and tested NLP techniques being successfully used in other domains – for e.g. Topic Modeling in advertisement and SEO (Search Engine Optimization) industry – are these being successfully adapted for infosec use cases? What are other examples?
3. Infosec benchmark datasets for language modeling – is enough work being done here? How can we move the needle here?
4. Dangers and pitfalls of open-ended language models in infosec
Bio: Nick is currently a software engineer at Google working on macOS endpoint security systems. He was previously a senior threat researcher at Capsule8 (acquired by Sophos), focusing on Linux server defense. His background is primarily in low-level systems and kernel exploitation research. Nick is also a Hacker in Residence and former student of NYU Tandon School of Engineering's OSIRIS Lab.