You hear a lot about how great machine learning is, and about how AI will change the world this century, but what you don't tend to hear so much about are the very serious security vulnerabilities inherent to most machine learning methods (both classical and quantum)...
Description
Adversarial machine learning is concerned with the task of tricking machine learning models in some way, e.g. fooling an image classifying convolutional neural network by changing a single pixel in a target image, or implanting undetectable backdoors into a neural network for later exploitation. Quantum adversarial machine learning is a very new (and very interesting) field of research which examines the vulnerabilities of quantum classifiers to these so-called "adversarial attacks", from uniquely quantum mechanical defence strategies to the deep geometric origins of these vulnerabilities in the first place.
We will create a quantum machine learning model which performs really well on a dataset of our choosing (already quite cool) and then show how we can conclusively deceive it using the methods of adversarial machine learning.
Abstract
You hear a lot about how great machine learning is, and about how AI will change the world this century, but what you don't tend to hear so much about are the very serious security vulnerabilities inherent to most machine learning methods (both classical and quantum)...
Description
Adversarial machine learning is concerned with the task of tricking machine learning models in some way, e.g. fooling an image classifying convolutional neural network by changing a single pixel in a target image, or implanting undetectable backdoors into a neural network for later exploitation. Quantum adversarial machine learning is a very new (and very interesting) field of research which examines the vulnerabilities of quantum classifiers to these so-called "adversarial attacks", from uniquely quantum mechanical defence strategies to the deep geometric origins of these vulnerabilities in the first place.
Members
westm2@student.unimelb.edu.au
Deliverable
We will create a quantum machine learning model which performs really well on a dataset of our choosing (already quite cool) and then show how we can conclusively deceive it using the methods of adversarial machine learning.
GitHub repo
This will appear in the fullness of time