Runpeng Dai

I am a third-year Ph.D. candidate at the University of North Carolina at Chapel Hill, advised by Prof. Hongtu Zhu. Before that, I obtained my B.S in Statistics from the Shanghai University of Finance and Economics where I was advised by Prof. Fan Zhou. My research sits at the intersection of Reinforcement Learning and LLM Reasoning, bridging theory and practice in AI.

Currently at Apple Cupertino as a Research Intern for the Summer. See you in Bay Area!

Research Highlights

LLM Reasoning & RLVR

Developing advanced reasoning training techniques and applying reasoning to real world applications.
- Efficient Exploration: Curiosity-Driven Exploration for RLVR training.
- Parallel Reasoning: Parallel-R1, the first RL framework to teach LLMs parallel thinking.
- Information Extraction: R1-RE, pioneer LLM reasoning for OOD Relation Extraction.
Efficient Reasoning & Test Time Scaling

Developing intelligent agents and advanced post-training techniques for next-generation AI systems.
- Adaptive Sampling: RL-Guided Sampling, a small RL controller that guides adaptive sampling for efficient test-time scaling.
- Probing: Parallel-Probe, towards efficient parallel thinking via 2D probing.

Research Experience

Tencent AI Lab (Seattle)

Research Scientist Intern

May 2025 - August 2025

Mentor: Dr. Linfeng Song

Develop Curiosity-Driven Exploration leveraging a model's intrinsic sense of curiosity to guide exploration in RLVR
Collaborated with fellow interns and colleagues on Parallel-R1 and VOGUE.

Baidu Qianfan

Research Intern

May 2024 - July 2024

Proposed a transformation-invariant sensitivity measure for LLMs and VLMs.
The measure can be applied to safeguard vulnerable parameters during quantization and model merging.

Selected Publications

	Small RL Controller, Large Language Model: RL-Guided Adaptive Sampling for Test-Time Scaling (arXiv), 2026. [Paper]
	LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling (arXiv), 2026. [Paper]
	CDE: Curiosity-Driven Exploration for Efficient Reinforcement Learning in Large Language Models ICLR 2026, International Conference on Learning Representations, 2026. [Paper]
	Parallel-Probe: Towards Efficient Parallel Thinking via 2D Probing ICML 2026, International Conference on Machine Learning, 2026. [Paper] [Code]
	R1-RE: Cross-Domain Relation Extraction with RLVR ACL 2026 main, The 64th Annual Meeting of the Association for Computational Linguistics, 2026. [Paper] [Code]
	Parallel-R1: Towards Parallel Thinking via Reinforcement Learning ICLR 2026, International Conference on Learning Representations, 2026. [Paper] [Code][Over 200+ Stars]
	StatEval: A Comprehensive Benchmark for Large Language Models in Statistics (arXiv), 2025. [Paper] [Project]
	Breach in the Shield: Unveiling the Vulnerabilities of Large Language Models EACL 2026 main, European Chapter of the Association for Computational Linguistics, 2026. [Paper]
	Causal Deepsets for Off-policy Evaluation under Spatial or Spatio-temporal Interferences (arXiv), 2024. [Paper]
	Deep Distributional Learning with Non-crossing Quantile Network (arXiv), 2025. [Paper]
	Spatio-temporal Prediction of Fine-Grained Origin-Destination Matrices with Applications in Ridesharing Journal of Computational and Graphical Statistics, 1-17, 2026. [Paper]

Teaching & Professional Service

Teaching Assistant of BIOS740 Deep Learning for Biomedical Applications, University of North Carolina at Chapel Hill, Fall 2024
Lecturer of Deep Learning Methods in Advanced Statistical Problems, JSM 2026, ENAR 2026, JSM 2025, ICSA 2024
Graduate Travel Award Recipient, 2026, University of North Carolina at Chapel Hill
Gillings Global Health Expendable Scholarship, 2026, University of North Carolina at Chapel Hill
LAMBDA Research Grant, 2026
ICLR 2026 Travel Award

Life

Outside of research, I enjoy fishing both in freshwater and saltwater. My Fishing Photos in Instagram .