Adversarial Robustness of Deep Learning Algorithms on Next-Gen AI Accelerators

Neural Networks play a key role in many modern technologies such as self-driving cars, drones, malware detection, and face recognition. For each of these technologies security and reliability is paramount. Unfortunately, researchers have shown that it is possible to reliably fool the neural networks behind these applications. Which makes identifying the best methods to defend a neural network against an attacker deadset on confusing it very important. This research seeks to compare how various proposed defense methods perform when tested on hardware designed specifically for accelerating neural networks and in doing so develop quick, power efficient defense methods for users of next-gen AMD AI Accelerators.

Faculty Supervisor:

Gennady Pekhimenko

Student:

Partner:

AMD Canada

Discipline:

Computer science

Sector:

Manufacturing; Professional, scientific and technical services

University:

University of Toronto

Program: