📷Pingwei SUN / 孙平炜
HKUST
👋 Welcome! I’m Pingwei, a postgraduate majoring in AI.
Check out my resumé and portfolio below 😊
Fine-tuning vs Prompting, Can Language Models Understand Human Values?
In this project, the use of PLMs as extractors and the end-to-end fine-tuning approach (prompt tuning, etc.) are compared to validate the comprehension capabilities of PLMs at different scales in understanding human values.
High-performance CPU Design and AI Application Based on MIPS ISA
This is my graduation thesis, using the NECore to build up a SoC with a CNN accelerator.
BERT Based Sentiment Analysis
Based on the BERT model, this project enhances the performance of the model on extremely unbalanced data sets through a variety of finetuning methods and a contrastive learning task.
Medical Image Processing
This project uses deep learning methods to solve the problems of DR diagnosis from fundus photos, the tissue segmentation of cardiac MRI, and the super-resolution of lung CT images.
NECore:A 6-stage Piepline CPU Based on MIPS32
Here is a project for NSCSCC2022 group competition. A CPU called “NECore” is built in Verilog and can be run both on Vivado & FPGA board.
Fundus Image Analysis with Deep Learning Methods
This project includes several submodules of the diabetic retinopathy diagnosis system, including macular positioning, vascular segmentation, and grading. Models like YOLO, UNet, and Efficientnet are applied to analyze fundus images and assist in the diagnosis of DR.