Defeating Machine Learning - Systemic Deficiencies for Detecting Malware

Explore the systemic deficiencies in machine learning for malware detection in this 45-minute BSidesLV conference talk. Delve into the evolution of the security landscape over the past 30 years, examining sandbox evasion techniques and current security postures. Gain insights into supervised machine learning, its potential vulnerabilities, and common model problems. Discover machine learning security solutions and learn about obfuscation techniques, balancing replacement and addition, and efficacy results from the speakers' lab experiments. Understand the benefits and challenges of vulnerability testing, continuous learning, and crowd-sourcing in improving malware detection systems. Examine feature vectors and their role in enhancing machine learning models for cybersecurity applications.

Intro
Who are we
Agenda
Security landscape 30 years ago
Sandbox evasion techniques
Security posture today
Machine learning
Supervised machine learning
Potential vulnerabilities of machine learning
Common model problem
Machine Learning Security Solutions
Our Lab
Obfuscating
Balancing Replacement and Addition
Efficacy Results
Benefits
Summary
Vulnerability Testing
Continuous Learning
Crowd Sourcing
Feature Vectors