Matthew Choi

Applied ML specialist - Vector Institute

Hi I’m Matt! I’m currently working as an applied ML specialist at the Vector Institute in Toronto, Canada. I graduated with a BASc in Biomedical Mechanical Engineering and a BSc in Computing Technology from the University of Ottawa in 2020.

Although my university experience was rooted in applied mechanics, I’ve developed a thorough passion for machine learning. My current interests are a combination of optimization, distributed computing, and large language models.

If you have any questions, feel free to reach out on either Linkedin or via email. If you just want to chat, that’s fine too!

Work Experience

Mar. 2023 - Now: Applied ML specialist - Vector Institute, Toronto, Canada

Dec. 2021 - Mar. 2023: Associate applied ML specialist - Vector Institute, Toronto, Canada

Sept. - Dec. 2021: Applied ML intern - Vector Institute, Toronto, Canada

May - Aug. 2019: Mechanical Engineering Intern - TWD EPCM, Burlington, Canada

Publications

FlexModel (arxiv coming soon): A Framework for Interpretability of Distributed Large Language Models - github

Projects

For a full list of personal projects, check out my github.

JAX LLM Scaling - Spring GTC 2023

Distributed Language Modelling - Language modelling with Megatron-LM distributed techniques using both PyTorch Distributed RPC and JAX pjit.

Conditional PATE-GAN - Tabular data generative model with differential privacy. An implementation of the PATE-GAN paper.

Graph Neural Networks in JAX - Exploratory project into various GNNs using JAX / Flax / j-raph.

Annealed Importance Sampling - Reproduction of results from the AIS paper by R. Neal (1998). Really interesting method for Monte-Carlo expectation approximation.

MetNet-2 - Framework reproduction of the MetNet-2 paper for spatiotemporal forecasting. This MetNet version includes the usage of ConvLSTMs and dilated convolutions, rather than the original axial attention modules.

Passive Lower-Body Exoskeleton - University capstone project, done using SolidWorks and Matlab.

More to come!