Thanks for popping by my github!

• 👋 Hi, I’m Shirley Wang.
• 👀 I’m interested in novel image segmentation models and algorithms, and other interesting computer vision projects.
• 🌱 I have recently completed my Msc at UofT!
• 🔭 I’m currently looking for job opportunities in the Toronto / Vancouver / Seattle area! I'd love to collaborate on some interesting ML problems.
• 😄 In my free time I love learning about new topics, algorithms, and subjects. I also love cats, playing piano, card games, and trying out new foods.

Masters of Computer Science | University of Toronto (2022-2024) | Supervisor: Sanja Fidler

Updated the polygon segmentation model to use a modern transformer backbone and loss function to achieve 85% IoU on five commercial segmentation datasets.

Applied Machine Learning Intern | Vector Institute (2021)

• Researched practical methods for creating domain-specific Conversational AI.
• Finetuned LLMs like GPT2 for multiple NLP tasks, adopting existing techniques to create models capable of extracting reasonable answers to queries from a given context in a natural dialogue manner, which perform well in both zero-shot and finetuned settings.
• Implemented a standardized data pipeline to finetune the model for any domain-specific dataset.

Data Engineer | Xero - Hubdoc (2020-2021)

• Maintained company’s data warehouse and data pipeline by creating and debugging ETL jobs using SQL, Javascript, and Python.
• Mapped out dependencies and created new DAGs to migrate the company data pipeline from a hard-coded schedule to Apache Airflow.
• Created dashboards and tables of KPIs for monthly reporting and other requirements.
• Worked with AWS Lambdas, S3, and Redshift to implement better database monitoring and archiving solutions.

Won 1st Place Best in Show at the UofT@Datafest2020.

Performed time series analysis using a linear mixed model on traffic data in the Greater Toronto Area to show the effects of the Covid lockdown on Toronto life.

Created a video game in C# and Unity, designed to teach basic algebra to children. This is done by representing variables with boxes, and numbers with toys. Done in collaboration with Professor Steve Engels.

Basic rhythm game programmed entirely in Verilog, played on four keys with some extra code to generate songs.

DEMO (Unravel from Tokyo Ghoul)

• Is AI is Coming for the Music Industry?
• DALL-E could be Friends with Flamingo
• Scientific Reasoning Told Me To Write This Article (featuring Meta AI)
• PaLI: Google Once Again Suggests Scaling Up Models
• Can This Model Generate My Head Exploding (.mp4)? Probably Not
• Adding Noise Until The Model Can Generate My Head Exploding (.png): A Brief Overview of Diffusion Models and DALL-E 2
• Generate Stylized MNIST Digits (and Other Cool Stuff) with Differentiable Rasterization
• Rotating The Way We View Position Embeddings