paulslss300,Yuzhi (Paul) Tang,github

Hi there 👋

I'm Yuzhi (Paul) Tang, an MScAC student at the University of Toronto. I am deeply passionate about identifying and mitigating risks in large-scale models. Currently, I'm working at the Machine Learning Group at the University of Toronto with Yangjun Ruan and Honghua Dong, where I developed an automatic pipeline to identify and decompose risks in LLM agent trajectories through prompt-tuning, and extended ToolEmu for large-scale evaluation on state-of-the-art LLM agents. My focus is on enhancing the safety and reliability of AI systems by addressing challenges in large-scale model deployment.

I also received the Undergraduate Student Research Award from NSERC for research in developing a reliability testing framework for computer vision models with Prof. Marsha Chechik. Previously, I applied ML to various healthcare settings, such as adapting a self-supervised pretraining recipe for ultrasound image segmentation at the MiDATA lab with Prof. Pascal Tyrrell and developing an end-to-end ML pipeline for sleep staging at the Sunnybrook Research Institute.

🔭 Current Focus: Risk identification and mitigation in large-scale LLMs.

🌱 Learning: Advanced techniques for improving LLM safety and reliability.

👯 Looking to Collaborate: On projects tackling the complexities and risks of large-scale AI models.

📫 Reach Me: [email protected]

⚡ Fun Fact: I won 1st place at the IJCAI 2023 Intrinsic Error Evaluation Competition!

Yuzhi (Paul) Tang's Projects

brain-mri-segmentation-using-unet

A UNet model trained on the Kaggle Brain MRI segmentation dataset with 84% IoU using PaddleSeg library.

exploratory-statistical-analysis-on-royal-canadian-yacht-club-dataset

Class Project for Introduction to Statistical Reasoning and Data Science: Exploratory Statistical Analysis on Royal Canadian Yacht Club Dataset

grammar-checker

Class Project for Foundations of Computer Science II: Python Grammar Checking Program for English using Constituency Parse Tree

paddleseg

Easy-to-use image segmentation library with awesome pre-trained model zoo, supporting wide-range of practical tasks in Semantic Segmentation, Interactive Segmentation, Panoptic Segmentation, Image Matting, etc.

quick-portfolio

Use this template if you need a quick developer / data science portfolio! Based on a Minimal Jekyll theme for GitHub Pages.

sharkconnect

Sharkconnect is a large iOS project which serves as a centralized platform for high schools, where club presidents and school admins post school-related contents, which are then shared among students. The app also integrates planning tools for students. Winner app of Game of Apps Season 1.

sslmi

Research Poster for ROP299: Self-Supervised Pretraining with Synthetic Data For Improving Ultrasound Medical Image Segmentation. Courtesy of MiDATA lab.

swiftcalendar

SwiftCalendar is a calendar interface written in Swift 4. Main features include the display of months as scrollable Collection View Cells and the display of events associated with a date in a Table View.

Recommend Projects