Short Bio
Welcome to my personal webpage! My name is Yang Xiao (Chinese: θι³) and Iβm currently a Masterβs candidate of Computer Science at Northeastern University (Vancouver, Canada) and also a intern at TikTok. Before that, I earned my B.E. degree(2019) from Beihang University (Beijing, China), supervised by Prof. Junchen Wang. I also have 4 years experience(including internships and full-time jobs) in AI industry.
My research interests mainly focus on 2D computer vision, including but not limited to image classification/detection/segmentation, semi-supervised learning, contrastive learning, human image generation, diffusion model, etc.
π Educations
- 2023.01 - now, Master, Khoury College, Northeastern University
- 2015.09 - 2019.07, Undergraduate, ShenYuan Honors College, Beihang University
π» Internships
- 2024.04 - now, TikTok, Vancouver, Canada
- 2020.07 - 2021.05, DiDi, Beijing, China
- 2019.12 - 2020.07, 4paradigm, Beijing, China
- 2019.06 - 2019.09, Haier Uhome, Beijing, China
- 2018.09 - 2018.12, ZhenRobotics, Beijing, China
π Research
- 2024.04 - now: Research Assistant
- Northeastern University, Vancouver, Canada
- Directions: Fire Prediction, Diffusion Model, Remote Sensing, etc.
- Supervisor: Professor Ryan Rad
π Teaching
- Spring 2024: Teaching Assistant
- Northeastern University, Vancouver, Canada
- Course: Foundation of AI
- Supervisor: Professor Richard Hoshino
- Fall 2023: Teaching Assistant
- Northeastern University, Vancouver, Canada
- Course: Programing Desgin Paradigm
- Supervisor: Professor Jack Thomas
π§ Projects
Pet Avatar Customized and Animated Generation via Diffusion Model
Apr. 2024
Master's capstone
code page
Master's capstone
code page
Built a pipeline aiming at generating pet images and videos in a personalized and customized way. To better preserve the original pet features, we followed DreamBooth + LoRA paradigm; furthermore, we inserted the LoRA weight into AnimateDiff framework to animate the original pet.
A Comparative Study between SLMs and LLMs in Customer Review Analysis
Apr. 2024
Master's capstone
code
Master's capstone
code
Investigated and compared the performance and computational costs of SLMs as well as LLMs in two tasks, sentiment polarity classification and correlation analysis; Explored the potential of combining LLMs with SLMs in customer review analysis to achieve better results and lower costs than either technique alone.
Real-time Interactive Online-Classroom
Nov. 2023
Course project
code
Course project
code
Built a online classroom where students and teachers can chat and draw some sketches in real-time. The tech stack is based on Typescript, Node.js, React and Redis. To ensure real-time and low-lantency, communication between front-end and back-end was implemented via websocket.
Lane Segementation Challenge on ApolloScape Benchmark
Dec. 2020
6/94 place on lane segementation track
6/94 place on lane segementation track
Combined multiple augmentation strategies, class-balanced loss functions and a sementic segmention Network(HRNet+OCR) to parse real-world image.
Chinese Artificial Intelligence Competition
Aug. 2019
3rd place on same source image retrieval category
3rd place on same source image retrieval category
Used ImageNet-pretrained backbone to extract image feature and trained a MLP to obtained the image embedding with triple loss.
3D Reconstruction with an Two-stage Stereo Matching Approach
