I am a second-year Ph.D. Student in Computer Science at the University of Maryland advised by Prof. Tianyi Zhou. My research focuses on pioneering efficient machine learning methods, encompassing computer vision, natural language processing, and multi-modality. I am particularly passionate about developing adaptive and personalized machine learning models. Before pursuing my Ph.D., I earned my M.S. and B.Eng in Computer Science from ShanghaiTech University, where I delved into research areas such as mixture of experts, domain generalization, and ensemble learning.

Actively looking for 2025 summer research internship opportunities!

🔥 News

• 2025.02:  New on arXiv: “R2-T2: Re-Routing in Test-Time for Multimodal Mixture-of-Experts” is now available on arXiv.
• 2025.01:  🎉🎉 One first-author paper has been accepted to the NAACL 2025 Main Conference.
• 2025.01:  🎉🎉 Two first-author papers have been accepted by ICLR 2025, including one Oral.
• 2024.10:  New on arXiv: “Your Mixture-of-Experts LLM Is Secretly an Embedding Model For Free” is now available on arXiv.
• 2024.03:  New on arXiv: “Many-Objective Multi-Solution Transport” is now available on arXiv.
• 2023.08:  Entering University of Maryland (UMD) as a Ph.D. Candidate in Computer Science.

📝 Publications

• Your Mixture-of-Experts LLM Is Secretly an Embedding Model For Free, Z. Li, T. Zhou, ICLR’25 (Oral, 1.8%)
  • Mixture-of-Experts (MoE) language models can serve as powerful, training-free embedding generators by leveraging the complementarity of routing weights and hidden states, achieving superior performance across diverse embedding tasks.
• Many-Objective Multi-Solution Transport, Z. Li, T. Li, V. Smith, J. Bilmes, T. Zhou, ICLR’25
  • Explores a framework that finds multiple diverse solutions in the Pareto front of many objectives.
• Sparser Mixture-of-Adapters with Cross-Layer Generalization, Z. Li, T. Zhou, NAACL’25
  • Improves efficiency and generalization of large language models by using a unified adapter pool shared across layers.
• SIMPLE: Specialized Model-Sample Matching for Domain Generalization, Z. Li, K. Ren, X. Jiang, Y. Shen, H. Zhang, D. Li, ICLR’23
  • An alternative direction for domain generalization via efficiently leveraging a pool of pretrained models without fine-tuning.
• Towards Inference Efficient Deep Ensemble Learning, Z. Li, K. Ren, Y. Yang, X. Jiang, Y. Yang, D. Li, AAAI’23
  • An inference efficient ensemble learning method to simultaneously optimize for both model effectiveness and efficiency in ensemble learning.
• Protecting the Future: Neonatal Seizure Detection with Spatial-Temporal Modeling, Z. Li $^\dagger$, Y. Fang $^\dagger$, Y. Li, K. Ren, Y. Wang, X. Luo, J. Duan, C. Huang, D. Li, L. Qiu, SMC’23
  • Utilizing spatial-temporal modeling for neonatal seizure detection.
• Why does the president tweet this? Discovering reasons and contexts for politicians’ tweets from news articles, Z. Li, H. Hu, H. Wang, C. Lu, H. Zhang, K. Zhang, Information Processing & Management 2022
  • A tweet interpretation framework which automatically extracts focal tweet clauses and tracks previous news to discover the event-specific reasons that trigger politicians to tweet them.

🎖 Honors and Awards

• 2023,2024: Dean’s Fellowship, University of Maryland.
• 2022: Award of Excellence for Stars of Tomorrow Internship Program, Microsoft Research Asia.

📖 Educations

• 2023 - Present, Ph.D. in Computer Science, University of Maryland.
• 2019 - 2022, M.S. in Computer Science, ShanghaiTech University.
• 2015 - 2019, B.Eng in Computer Science, ShanghaiTech University.
• 2015 - 2019, Minor in Finance, ShanghaiTech University.

💻 Internships

• 2022.06 - 2022.12, Research Assistant, AI/ML group, Microsoft Research Asia, Shanghai, China.
• 2021.06 - 2022.03, Research Intern, AI/ML group, Microsoft Research Asia, Shanghai, China.