Hello! I am Murdock.

murdock-aubry

murdock@cs.toronto.edu

@murdock.aubry

Who am I?

If you haven't already guessed, my name is Murdock. I am originally from Collingwood, Ontario, however I now live in downtown Toronto. I have always pushed myself to be the best at everything I can, and in the former half of my life, this largely took the form of freestyle skiing; some videos can be found here. However, for the last several years, most of my efforts have been focussed on cultivating my technical skills. More generally, I love solving difficult problems.

I recently graduated with a BSc in Mathematics and Physics from UofT, and am currently pursing a Master of Science in Applied Computing (MScAC) at UofT in the applied mathematics stream. I research deep learning architectures, particularly transformers, at the Vector Institute under the supervision of Professor Vardan Papyan. I have previously conducted research on topics in numerical analysis under the supervision of Professor James Bremer, and also on particle physics and deep learning at DUNE under Professor Nikolina Ilic. More information on my current and past projects can be found here. A closer look at my academic background can be found here.

What is next?

My personal research interests are two-fold:

1. Interpretability and explainability of deep learning architectures, particularly transformer networks, from a rigorous mathematical and physical standpoint. I perform extensive empirical experiments and theoretical explorations to quantify relationships between distinct neural connections.
2. Designing and training generative models to adhere to first principles by modulating their outputs by the differential equations which govern the laws of physics. This can potentially be achieved by deploying the concept of physics-informed neural networks, however, I am openly exploring several other methods.
I aim to utilize my remaining time in graduate school to study and implement these machine learning architectures, and moreover, analyze and replicate the behavior of these systems in CAD and CFD software. Following this, I hope to obtain a research position at such a software development company where I can continue to research the application of physics-informed machine learning to front-end applications, and back my exploration based on my extensive interpretability research.