Yu(Jaray) Li 李煜

Senior Student
Hongyi Honor College
Wuhan University
Email:yul79@uci.edu / yuuli2021@whu.edu.cn

[Github] [Linkedin] [CV] [Transcript]

Biography

I am currently a senior in Hongyi Honor College of Wuhan University, majoring in Microelectronics Science and Engineering. I have a certain understanding of digital/analog circuits and computer architecture. My current main research direction is the application of AI in vision, including LLM/VLM research.

Recent News

  • [08/2024] I am looking for a PhD position for Fall 25 entry, and I hope to continue exploring ML as well as VLM and applications in computer vision.

  • [08/2024] Our independent research on methods for fine-tuning LLM parameters in OCR methods was accepted by ICONIP 2024.

  • [05/2024] I will join AICPS at UC Irvine as a research intern, during summer 2024.

  • [05/2024] I was awarded the Innova International Exchange Scholarship, thanks to Dr. Charles Huang and Charles Huang Foundation!

Research

My research focuses on ML&AI application, including improvement of LVLM/VLM, application in medical/healthcare field and integration with EDA.

UCI Fusion Image

Sensor Fusion for Robust and Efficient Autonomous Vehicle Perception
Yu Li, Junyao Wang
2024.6-Present.
[Baseline] [Dataset]

Our main work is to use LVLM to improve the autonomous driving mode in abnormal weather conditions, focusing on improving the multi-modal fusion algorithm of Camera-Lidar branches. Hopefully it will be advanced to hardware deployment in the future.

Ultrasound Image

Ultrasound image segmentation based on SAM and Transformer multi-level feature fusion
Yu Li, Jin Huang
2023.10-2024.5
Under Review, IEEE Journal of Biomedical and Health Informatics.

[Code]

We leveraged the rich semantic segmentation information of SAM and applied its fine-grained attention capability to the feature extraction module of Transformer, achieving SOTA IoU scores in ultrasound image segmentation.

Ultrasound Image 2

Windowed self-attention guided multi-scale feature stream alignment network for ultrasound image segmentation
Jin Huang, Xiaoxiao Li, Yu Li
2023.9-2024.5
Under Review, Biomedical Signal Processing and Control.

[Code]

We explored the self-attention guidance mechanism of the Swin-Transformer structure in breast ultrasound images, used a multi-level encoder to deal with multi-dimensional features, and used spatial pyramid pooling and feature stream alignment to fuse shallow and deep information.

Depth Image

Diffusion model with discriminative priors: Self-supervised monocular depth estimation in endoscopy
Yu Li
2024.2-2024.6
Under Review, International Journal of Computer Assisted Radiology and Surgery.

[Code]

For endoscope medical scenarios, we use the diffusion model for depth estimation. We build a teacher model, set knowledge distillation, optical appearance and ddim losses, and introduce the teacher's discriminative prior, which significantly enhances the accuracy and confidence of the results.

Monte Carlo Simulation Image

SpO2 measurement model based on Monte Carlo simulation
Yu Li, Shing-jiuan Liu, Prof. Weijian Yang
2023.7-2024.5

[Code]

We used the modified Beer-Lambert law for reverse deduction, used Monte Carlo simulation data to improve the SpO2 measurement process model and design algorithms, and finally realized the reverse model for determining physiological parameters from simulation data.

ECG Image

ECG classification based on deep learning
Yu Li, Yunhao Hu
2022.10-2023.3

[Paper] [Code]

For the attempt of deep learning method for ECG classification, the construction of multi-layer neural network and the parameter grid search of effective tool XGBoost.

Projects

Individual or Collaborative Projects

DLoRA module

DLoRA-TrOCR: Mixed Text Mode Optical Character Recognition Based On Transformer.
Yu Li & Da Chang
2024.4, ICONIP 2024.

[Paper] [Code]

We explored the optimization of various full-parameter fine-tuning methods such as LoRA in LLM. For OCR, a hybrid visual-text model, corresponding to the Transformer architecture, DoRA and LoRA have good improvement effects on the visual encoder and text decoder respectively.

UCB eecs151 aisc

A 3-stage CPU that implements the RISC-V ISA. [UCB EECS151 Project]
Yu Li, Timonty Chu
2024.2-2024.4


In the first phase (front-end), I design and implement a 3-stage RISC-V processor in Verilog, and run simulations to test for functionality. Then in the second phase (back-end), I implement front-end design in the SKY130 PDK using the VLSI tools.

UCB ee140 project

LCD Driver Design: 2-Stage Operational Amplifier. [UCB EE140 Project]
Yu Li
2024.2-2024.4


To use a series resistance of 400 Ohms and a capacitance of 60 pF to design the LCD amplifier to the specifications.

Unisoc Competition

FPGA-based video acquisition and performance acceleration
Yu Li
2023.3-2023.6


We developed a multi-channel video acquisition system with hardware acceleration on the Unisoc FPGA. We configured the YoLoV5 model on the host computer, collected video input via HDMI transmission, onboard DDR video frame buffer, and output via PCIE. The FPGA acceleration was achieved using several techniques: 1) optimized precision conversion, 2) pipeline strategy, 3) multi-channel and addition structure, and 4) DDR5 partition data sharing.

Education

Experience

Honors and Awards

Innova International Exchange Scholarship, Wuhan University, 2024
Innova Excellence Scholarship (Top 3%), Wuhan University, 2023
Academic Excellence Scholarship (Top 5%), Hongyi Honor College, 2022 & 2023
First-Class Scholarship (Top 5%), Wuhan University, 2022 & 2023

Misc

National Patent: Energy-saving calculation method, CN116085952.

Skills