About me
I am a PhD candidate in Electrical and Computer Engineering at Purdue University, advised by Prof. Xiaoqian Wang.
Prior to Purdue, I worked with Prof. Andrew Owens and Daniel Geng during my master’s studies. Earlier, I collaborated with Jinge Ma and Yutong Xie under the guidance of Prof. Qiaozhu Mei. I also worked with Ying Yang and Prof. Daniel Klionsky. During my undergraduate studies, I collaborated with Zhiqiang Yuan under the supervision of Prof. Kun Fu and Prof. Xian Sun.
My research focuses on understanding and improving the reliability of machine learning models, particularly their ability to learn from and reason under conflicting or misleading evidence. I develop theoretical foundations and practical methods for building reliable multimodal foundation models. More recently, my work has expanded to address the reliability of large language models (LLMs) and the safety of autonomous AI agents. I am open to new research areas and potential collaborations.
I’m actively looking for internship opportunities for 2027 Summer!
Education
Purdue University 2023–present
PhD in Electrical and Computer Engineering
Advisor: Prof. Xiaoqian Wang
University of Michigan 2021–2023
MS in Electrical and Computer Engineering, specialization in Computer Vision
Advisor: Prof. Andrew Owens and Prof. Qiaozhu Mei
University of Chinese Academy of Sciences 2017–2021
BE in Electronic and Information Engineering
Bachelor’s thesis with honors: Image Caption Generating of High-Resolution Remote Sensing Images
Advisors: Prof. Kun Fu and Prof. Xian Sun
Publications

Beyond Modality Preference: Controlling Multimodal Arbitration under Modality Conflict in MLLMs
A method to identify attention heads that determine whether a multimodal model follows visual or textual evidence when the two conflict, then steer those heads toward a chosen trusted modality at inference time without retraining.

Guiding Data Allocation for Robust Subpopulation Generalization
We show that balanced subgroup data is not the only route to robust generalization and derives gradient-based directions for allocating data to improve adjusted and worst-group accuracy more efficiently.

On the Effects of Adversarial Perturbations on Distribution Robustness.
An analysis on how adversarial perturbations affect robustness under distribution shifts, showing that their impact depends on spurious correlations and feature separability and can sometimes improve distribution robustness.

Think Twice: Test-Time Reasoning for Robust CLIP Zero-Shot Classification.
A inference-time method that uses CLIP’s own semantic cues to localize objects at test time and classify the localized region, reducing reliance on spurious backgrounds without requiring spurious-attribute annotations.
MASSW: A New Dataset and Benchmark Tasks for AI-Assisted Scientific Workflows
A comprehensive text dataset on Multi-Aspect Summarization of Scientific Workflows and it includes more than 152,000 peer-reviewed publications from 17 leading computer science conferences spanning the past 50 years.


Autophagy and Machine Learning: Unanswered Questions

Self-Supervised Motion Magnification by Backpropagating Through Optical Flow
A method for magnifying subtle motion without ground-truth magnified videos, with support for region-specific magnification and single-video overfitting.


Diffusion Model with Detail Complement for Super-resolution of Remote Sensing
A detail-complement diffusion model for remote-sensing image super-resolution.

Projects
- The Evolution of AI and Data Mining. We analyzed computer science papers spanning half a century and studied how the distribution of research domains changed over time. Watch the video.
Teaching
ECE 570: Artificial Intelligence
Graduate teaching assistant, Purdue University, 2026 Fall
Job Interviews in the AI Era: Coding, Systems, and Agents
Student Instructor, Purdue University, 2026 Summer
ECE 570: Artificial Intelligence
Graduate teaching assistant, Purdue University, 2025 Spring
ECE 570: Artificial Intelligence
Graduate teaching assistant, Purdue University, 2024 Fall
ECE 264: Advanced C Programming
Graduate teaching assistant, Purdue University, 2024 Spring
ECE 264: Advanced C Programming
Graduate teaching assistant, Purdue University, 2023 Fall
Service
Reviewing
2026
- Conferences: AAAI, IEEE Big Data, ICLR, AISTATS, ICML, UAI, NeurIPS, WACV, ICASSP
- Journals: Transaction on Graphics
- Workshops: ICBINB@ICLR
2025
- Conferences: CVPR, ICML, NeurIPS
- Journals: Transaction on Graphics
- Workshops: ICBINB@ICLR, MLGenX@ICLR, SRW@ACL
2024
- Conferences: ECCV, NeurIPS, ICLR, AISTATS
- Workshops: MLGenX@ICLR, AIBS@KDD
Recognition
- NeurIPS 2026 Top Reviewer Award
Conference and Workshop Organization
- Website Chair: Learning on Graphs (LoG) 2026
- Workshop Organizer: ICLR 2026 Workshop: I Can’t Believe It’s Not Better: Where Large Language Models Need to Improve
- Workshop Organizer: ICLR 2025 Workshop: I Can’t Believe It’s Not Better: Challenges in Applied Deep Learning
