I graduated from University of Wisconsin - Madison in May, 2024 with a B.S degree, Honor in the Major, majoring computer science.

My research interest is in machine learning. Specifically, I am interested in data-efficient models and robust language models.

Ph.D. applicant for Fall 2025.


News

2024.10.15First first-author conference submission to NAACL 2024!
2024.08.12Congratulations on our paper accepted as an oral presentation on ECCV 2024!
2024.07.01Congratulations on our paper acceptance on ECCV 2024!
2024.06.20Congratulations on our paper acceptance on ICML 2024 workshop NextGenAISafety!
2024.05.11Graduated from University of Wisconsin - Madison with B.S. with Honor in the Major, Computer Science!
2024.04.19Presented my senior honor thesis @Senior Honors Thesis Symposium 2024!


Publications

Personalize LLM: Fake it and Then Align.Yijing Zhang, Dyah Adila, Changho Shin, Frederic Sala.
In submission to NAACL 2025
Audio-Guided Visual AnimationLin Zhang, Shentong Mo, Yijing Zhang, Pedro Morgado.
Oral presentation by ECCV 2024. [Code]
Is Free Self-Alignment Possible?Dyah Adila, Changho Shin, Yijing Zhang, Frederic Sala.
Preprint. In submission to ICLR 2024.
Can Language Models Safeguard Themselves, Instantly and For Free?Dyah Adila, Changho Shin, Yijing Zhang, Frederic Sala.
Accepted by ICML 2024 Workshop on NextGenAISafety.
Methods for Domain-specific Fine-tuning for Generative ModelsYijing Zhang, Frederic Sala.
Senior honor thesis, L&S Honors Program, University of Wisconsin - Madison.
OpenASVA: Open-world Audio-Synchronized Visual Animation with Increased ControllabilityLin Zhang, Yufan Zhou, Shentong Mo, Cheng-En Wu, Yibing Wei, Yijing Zhang, Jinhong Lin, Ruiyi Zhang, Tong Sun, Pedro Morgado.
Submission to CVPR 2025.

Education

University of Wisconsin - Madison
B.S with Honor in the Major
Computer Science Major
Sep. 2021 - May. 2024


Research Projects

Project 1: Methods for Domain-specific Fine-tuning for Generative Models

This is an independent research study for my senior honor thesis supervised under Prof. Frederic Sala. It focus on investigating the retrainability of synthetic datasets generated by fine-tuned generative models for domain-specific downstream classification tasks and the fine-tuning efficiency for generating higher-quality synthetic datasets. Presented my work at Senior Honors Thesis Symposium 2024.

Related Topics: NLP, Foundation models, Generative models, Fine-tuning, Data Efficiency

Project 2: Audio-Guided Visual Animation

Supervised by Prof. Pedro Morgado, we worked on highly syncrhonized audio-to-video generation, focusing on generatation of animation with audio guidance. Resulting paper is in submission to ECCV 2024.

Related Topic: Computer vision, Generative models, Audio-to-Video.

Project 3: AlignEZ

Supervised by Prof. Frederic Sala, we worked on self-aligning models without further training. We proposed AlignEZ that uses (1) self-generated preference data; (2) representation editing to provide nearly cost-free alignment. Resulting paper is in submission to ICLR 2025.

Related Topic: NLP, Alignment.

Project 4: P-AlignEZ (Personalized AlignEZ)

Supervised by Prof. Frederic Sala, we worked on utilizing an AlignEZ approach for personalizing LLM. We found that P-ALignEZ is able to perform model larges-scale personalization for multi-users using low resource. Planned submission to ICML 2025.

Project 5: On Generating Better Quality Instruction Tuning Data

Supervised by Prof. Bach and Prof. Frederic Sala, we focused on research for methods for generating better quality instruction tuning dataset from unannotated text for model tuning.


Work Experience

University of Wisconsin - Madison
Research Assistant
Sep.2024 - Present
University of Wisconsin - Madison
CS400 Peer Mentor
Jan.2022 - May.2024
University of Wisconsin - Madison
WISCERS Program - Summer Research Intern
May. 2022 - Aug. 2022