I am a PhD candidate at Princeton University and a Princeton AI Lab Fellow, working with Prof. Mengdi Wang, Prof. Andy Yao, and Prof. Quanquan Gu, where my research focuses on building scalable and capable large language models (LLMs) and multimodal foundation models. My work explores methods to improve LLM reasoning via reinforcement learning, data curation and algorithms for foundation models, as well as the development of new attention mechanisms, positional encodings, and model architectures. Previously, I was a visiting PhD student at the University of California, Los Angeles.
About Me
Research Interests
- Language Modeling and Pretraining
- LLM Reasoning and Reinforcement Learning
- Physics of Deep Learning
You can find my publications on Google Scholar.
You can find my blog posts at Yifan's Blog.
Selected Works
Deep Delta Learning
arXiv:2601.00417
[GRAPE] Group Representational Position Encoding
International Conference on Learning Representations (ICLR 2026)
[RPG] On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning
International Conference on Learning Representations (ICLR 2026); See also Thinking Machines Tinker
[TPA] Tensor Product Attention Is All You Need
Conference on Neural Information Processing Systems (NeurIPS 2025 Spotlight)
[GPM] Beyond Bradley-Terry Models: A General Preference Model for Language Model Alignment
International Conference on Machine Learning (ICML 2025)
[AutoMathText] Autonomous Data Selection with Zero-shot Generative Classifiers for Mathematical Texts
Findings of the Association for Computational Linguistics (ACL 2025 Findings)
[MMIQC] Augmenting Math Word Problems via Iterative Question Composing
AAAI Conference on Artificial Intelligence (AAAI 2025)
[Proposer-Verifier] Cumulative Reasoning with Large Language Models
Transactions on Machine Learning Research (TMLR)
(* denotes equal contribution)
Recent Publications
Group Representational Position Encoding
International Conference on Learning Representations (ICLR 2026)
On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning
International Conference on Learning Representations (ICLR 2026); See also Thinking Machines Tinker and DeepSeek V3.2
Tensor Product Attention Is All You Need
Conference on Neural Information Processing Systems (NeurIPS 2025 Spotlight)
Autonomous Data Selection with Zero-shot Generative Classifiers for Mathematical Texts
Findings of the Association for Computational Linguistics (ACL 2025 Findings)
Beyond Bradley-Terry Models: A General Preference Model for Language Model Alignment
International Conference on Machine Learning (ICML 2025)
Augmenting Math Word Problems via Iterative Question Composing
AAAI Conference on Artificial Intelligence (AAAI 2025)
Beyond Squared Error: Exploring Loss Design for Enhanced Training of Generative Flow Networks
International Conference on Learning Representations (ICLR 2025 Spotlight)
(* denotes equal contribution)
Blog Highlights
ShortSWA Is the Next-Generation N-gram Embedding
Yifan's Blog, January 12, 2026
Revisiting Variance Reduction in Policy Gradients for LLM Reinforcement Learning
Yifan's Blog, December 27, 2025
Rethinking SWA: Why Short Sliding Window Attention Will Replace ShortConv
Yifan's Blog, December 16, 2025
Matrix Exponential Attention
Yifan's Blog, December 15, 2025
Professional Activities
Teaching
- Teaching Assistant, Machine Learning for Yao class, IIIS, Tsinghua University
Academic Services
- Conference Reviewer: NeurIPS, ICLR, ICML, COLM, AAAI, AISTATS
- Journal Reviewer: ACM TKDD, Neural Computing, Neural Networks