Overview
Explore the vulnerabilities of deep learning systems in this comprehensive conference talk from GrrCon 2016. Delve into the world of hacking machine learning, focusing on deep learning techniques and their widespread applications. Understand neural networks, convolutional networks, and recurrent neural networks, along with their roles in text generation, speech recognition, and long-term memory. Discover the potential attack vectors in the text economy and learn about blind spots in machine learning models. Examine three key steps and methods for exploiting deep learning systems, including transferability and substitute models. Gain insights into false assumptions and privacy concerns surrounding deep learning technologies. Conclude with a Q&A session to address specific inquiries about machine duping and pwning deep learning systems.
Syllabus
Introduction
Hacking Machine Learning
Deep Learning
Deep Learning is Everywhere
Why Use Deep Learning
Neural Networks
Convolutional Networks
Layered Learning
Recurrent Neural Networks
Text Generation
Long Term Memory
Speech Recognition
Machine Learning
Attack Text Economy
Blind Spots
Three Steps
First Way
Transferability
Substitute Models
False Assumptions
Three Methods
Deep Boning
Deep Learning Privacy
Questions