Understanding Security Threats Against Machine - Deep Learning Applications

Explore the world of security threats against machine and deep learning applications in this comprehensive conference talk. Gain a quick understanding of neural networks and delve into associated hacking methods, including trojaning, adversarial examples, adversarial patches, data poisoning, model extraction, and training data leakage. Learn about the complexities of machine learning security, explained in a way that benefits developers of all levels. Discover the nuances of neural networks, whitebox vs blackbox approaches, differentials, derivatives, fast gradient sign method, Jacobian map approach, Carlini Wagner attack, genetic algorithms, and synthesizing data. Understand the importance of these concepts in the context of AI and application security, presented by Abraham Kang, an experienced security researcher with a background in AI, security, and development from companies like Fortify and Cornell University.

Introduction
Review of Machine Learning
Example
Adversary Inputs
Neural Network
Why is this important
How does this work
Neural Networks
Whitebox vs Blackbox
Differentials
Derivative
math
fast gradient sign
Jacobian map approp approach
Carlini Wagner
Genetic Algorithms
Synthesising Data
Adversary Patches