My name is Zhihan Zhang (张智涵). I am a third-year Ph.D. student in Computer Science and Engineering at the University of Notre Dame, advised by Dr. Meng Jiang. I am currently a member of Dr. Jiang’s Data Mining towards Decision Making (DM²) Lab. Prior to my Ph.D. career, I received my Bachelor’s degree in Computer Science at Peking University, where I had the fortune of working with Dr. Yunfang Wu and Dr. Xu Sun. My research is mainly related to the interaction between natural language instructions and language models, such as instruction tuning and the evaluation of instruction-following language models.

For my past education and research experience, please refer to Experience. For the full list of my publications, please refer to Publications or check my Google Scholar page.

News

• Jun 2024: Several new preprints came out! Check out our latest works:
  • A first-authored paper worked on fine-tuning language models for mathematical reasoning. By augmenting training examples with reflective data, we improved the math abilities of models in both the standard single-round QA and more complex reflective reasoning scenarios.
  • A co-authored paper studied the problem of language models self-correcting their own predictions. We utilized backward verification methods to unleash the model’s capacity to identify and correct their reasoning errors without external feedback.
  • A co-authored paper proposed a new benchmark for comprehensively evaluating the mathematical capabilities of language models. We went beyond single-round QA and assessed the models’ math abilities in multi-turn interactions and open ended generative scenarios.
  • A co-authored paper studied multi-modal mathematical reasoning. We focused on improving MLLMs’ visual understanding capacity on math figures via explicit visual comprehension training, which improved their math reasoning accuracy.
• Jun 2024: I am joining Amazon as a full-time research intern in summer 2024.
• May 2024: One first-authored paper is accepted to ACL 2024 main conference! We studied the challenges in cross-lingual instruction-tuning and proposed a simple yet effective method to improve LLMs’ proficiency in low-resource languages. Code is available on Github.
• Feb 2023: I am joining Tecent America as a full-time research intern in spring 2024.
• Oct 2023: Two papers are accepted to EMNLP 2023! These include one first-authored paper on automatic instruction optimization of LLMs, and one co-authored paper on language model pre-training for comparative reasoning.

Contact

• Email: zzhang23 [at] nd.edu
• Office: 355 Fitzpatrick Hall of Engineering
• Location: University of Notre Dame, Notre Dame, IN 46556