Explore the vulnerabilities of machine learning methods used for cyber attack detection in this 45-minute conference talk. Delve into the increasing complexity of communication networks and the rising interest of attackers in gaining access to information. Examine why machine learning methods are proposed for effective cyber attack detection, including their scalability, speed, and ability to protect against unknown threats. Discover various inherent properties of machine learning methods that allow attackers to bypass detection systems. Learn about specific attacks on these methods and their impact on detection performance. Gain insights into current open problems in machine learning security and cyber attack detection. Understand topics such as model stealing, model inversion, evasion attacks, and defensive strategies like distillation and effective defenses. Conclude with a discussion on evolution scenarios, cross-transfer ability, and limitations in the field.
Attacking Machine Learning Methods Used for Detection of Cyber Attack
Overview
Syllabus
Introduction
Who am I
About my work
Application domains
Why Machine Learning for Security
Machine Learning in Practice
Model Stealing
Model Inversion
Evasion Attack
Other Attacks
How does it work
How can we automate
Publicly available libraries
Intra transferability
Why
Can we defend
First defense
Defensive distillation
Defensive mentality
Effective defenses
Evolution Scenario
Cross Transfer Ability
Limitations
Conclusion
Questions
Taught by
WeAreDevelopers
Related Courses
-
Bot vs. Bot for Evading Machine Learning Malware Detection
-
Trusted and Responsible AI - Explainability, Adversarial Attacks, Bias, and Fairness
-
Fighting New Era Bandits Who Are Plundering Your AI - ML Application
-
Adversarial Examples in Machine Learning - Crafting and Defending Against Attacks
-
Duplicating Black Box Machine Learning Models
-
Defeating Machine Learning - Systemic Deficiencies for Detecting Malware
