Laura Graves

Machine Learning & Security

Laura Graves
Laura Graves

About Me

Graduate researcher at the University of Waterloo in Waterloo, Ontario.

I like breaking neural networks. My research focuses on networks that do things they aren't supposed to do and how we can exploit that, fix it, or learn from it. I also work with explainable AI systems and how we can use them to make the systems we use better, and I'm interested in what we can find at the intersection of machine learning and formal logic. When I'm not doing research or tinkering with side projects I like to spend my time playing music, reading, taking pictures of birds (gotta catch em all!), and tinkering in the kitchen. You can find me on Github or LinkedIn


Machine Learning Security

Machine learning models learn all sorts of things they weren't intended to learn. A consequence of that is that they also leak all sorts of information that wasn't meant to be leaked, from class information to private records to unintentional decision boundaries. I'm intrigued by how we can use these unintentional properties to learn more about how these models learn, how we can exploit these properties, and most importantly how we can protect against people trying to exploit them.

Algorithmic Fairness

Deep networks often learn to discriminate in unintentional or opaque ways, and I believe that making those predictions easier to understand and prevent biased behavior is extremely important for production models. To this end, I'm working on ways to detect unfair or biased behavior and mitigate it for future predictions.

Computer Science Theory

I helped tutor at the MUN help centre for courses such as data structures and algorithms, logic for computer scientists, and object oriented programming. I've studied higher level theory such as heuristics and algorithms for hard problems, including satisfiability solving and using evolutionary algorithms to solve NP-hard problems.

St. Johns, NL



Amnesiac ML

As neural networks are trained on more and more personal data, regulations such as the right to be forgotten highlight the need for methods of removing learned data from trained networks. This research focuses on these methods that cause networks to forget what they've learned about sensitive data without harming the efficacy of the model.

In this project I was the primary researcher and writer, and I developed the codebase in PyTorch. Accepted for publication at AAAI-2021.

Check it out
xAI-Gan Architecture


xAI-GAN leverages explainable AI systems to focus the gradient descent process for generator training on the most important features, leading to improved data efficiency and a new way of looking at GAN training. Accepted for publication at the Explainable AI workshop at AAAI-2021.

In this project I was a primary researcher and developer along with a colleague, and I assisted with writing

Check it out
GANs n Reels Process

Generating traditional Irish music with a Generative Adversarial Network

Under the supervision of Dr Kolokolova, a group of undergrads used a GAN to create original music that matched the formed of traditional Irish music. This project was recently featured on CBC Radio!

Check it out

University of Waterloo

In progress: Masters of Applied Science in Electrical & Computer Engineering (expected August 2021)

Under the supervision of Dr. Vijay Ganesh

Awarded an Engineering Excellence Fellowship - MASc

GPA 4.0

Memorial University of Newfoundland

BSc. in CS, Minor in Mathematics. GPA 3.90, major GPA 4.0

Faculty of Science Dean's List 2017-2018 and 2018-2019

University Medal for Academic Excellence in Computer Science, 2019.

Courses of note

CS858 - Advanced Topics in Security and Privacy for AI and Machine Learning (Dr. Florian Kershbaum)

Advanced graduate course covering state-of-the-art research on AI security and privacy research.

Grade: 93

CS885 - Reinforcement Learning (Dr. Pascal Poupart)

Covers reinforcement learning topics such as Markov decision processes, model based and model free RL, deep RL, heirarchical RL, inverse RL, and meta learning. Literature survey on attacking deep RL

Grade: 89

COMP6902 - Theory of Computation (Dr. Antonina Kolokolova)

Graduate course dealing with the problems of computational complexity including decidability of languages, polynomial-time hierarchy, and P vs NP. Final project on the performance of SAT solvers on Sudoku puzzles.

Grade: 99

COMP6915 - Machine Learning Computation (Dr. Lourdes Peña-Castillo)

Graduate course covering machine learning methods, focusing on applications in bioinformatics. Literature survey on adversarial attacks on ML models.

Grade: 94


Download my resume