Sunny Qin

I am a fourth-year PhD student at Harvard, advised by Sham Kakade and David Alvarez-Melis, and a member of the Machine Learning Foundations Group. I'm honored to have been selected as part of the 2025 Apple Scholars in AI/ML. I'm interested in data-centric AI and the science behind how LLMs learn to reason. My work explores how data, both synthetic and natural, and different training paradigms, such as RL, jointly shape model behavior and capabilities. Ultimately, I aim to uncover general principles for building more capable and interpretable foundation models.

Blog Posts

Backtrack Thumbnail
To Backtrack or Not to Backtrack: When Sequential Search Limits Model Reasoning
Unknowable Thumbnail
Distinguishing the Knowable from the Unknowable with Language Models

Research

COMPASS: A Multi-Turn Benchmark for Tool-Mediated Planning & Preference Optimization. arXiv preprint
Tian Qin, Felix Bai, Ting-Yao Hu, Raviteja Vemulapalli, Hema Swetha Koppula, Zhiyang Xu, Bowen Jin, Mert Cemri, Jiarui Lu, Zirui Wang, Meng Cao.
[ArXiv]
To Backtrack or Not to Backtrack: When Sequential Search Limits Model Reasoning. COLM 2025
Tian Qin, David Alvarez-Melis, Samy Jelassi, Eran Malach.
[ArXiv]
Decomposing Elements of Problem Solving: What "Math" Does RL Teach? arXiv preprint
Tian Qin*, Core Francisco Park*, Mujin Kwun, Aaron Walsman, Eran Malach, Nikhil Anand, Hidenori Tanaka, David Alvarez-Melis.
[ArXiv]
Distributional Scaling Laws for Emergent Capabilities. arXiv preprint
Rosie Zhao*, Tian Qin*, David Alvarez-Melis, Sham Kakade, Naomi Saphra.
[ArXiv]
Sometimes I am a Tree: Data Drives Fragile Hierarchical Generalization. EMNLP 2025
Tian Qin, Naomi Saphra, David Alvarez-Melis.
Distinguishing the Knowable from the Unknowable with Language Models. ICML 2024
Tian Qin*, Gustaf Ahdritz*, Nikhil Vyas, Boaz Barak, Benjamin L. Edelman.
[ArXiv] [GitHub]
A Label is Worth A Thousand Images in Dataset Distillation. NeurIPS 2024
Tian Qin, Zhiwei Deng, David Alvarez-Melis.
[ArXiv] [GitHub]
Distributional Dataset Distillation with Subtask Decomposition. DMLR @ ICLR 2024
Tian Qin, Zhiwei Deng, David Alvarez-Melis.
[ArXiv] [GitHub]
Meta-PDE: Learning to Solve PDEs Quickly Without a Mesh. Preprint
Tian Qin*, Alex Beatson*, Deniz Oktay, Nick McGreivy, Ryan P. Adams.
[ArXiv] [GitHub]

Invited Talks

Teaching & Service

Personal

Outside research, I enjoy spending my free time outdoors — climbing all sorts of things: rocks, boulders, big wall, frozen waterfalls, and big mountains. I am a board member and an education officer of the Harvard Mountaineering Club. I also recently got into surfing!

Contact

Email: tqin[AT]g.harvard.edu

Feel free to reach out - I am always open for collaborations, discussing research ideas.