Sirui Chen
Ph.D. Student in Computer Science
University of Illinois Urbana-Champaign
I'm a Ph.D. student in Computer Science at the University of Illinois Urbana-Champaign, advised by Prof. Jingrui He. Before that, I received my M.Eng. from Renmin University of China, advised by Prof. Jun Xu, and my B.A. in Japanese from Tsinghua University.
My long-term goal is to build reliable, controllable AI systems that improve themselves, closing the loop from data curation to model training to evaluation. My research currently focuses on two directions:
- Reinforcement learning and reasoning. How can a model improve by learning from its own experience? I study how LLMs and agents can explore, learn from feedback on their own trajectories, and turn extra computation into better reasoning, making this self-improvement loop efficient and reliable in training and at inference time.
- Data-centric learning. What a model becomes is shaped by what it learns from. I study how to measure the value of data for a given model and use that signal to select and build training data automatically, so that models can curate their own learning material as they evolve.
Before LLMs, I worked on reinforcement learning for recommender systems and multi-agent RL, including research at Alibaba. I'm always happy to chat about research or collaborations, so feel free to reach out.
News
NewTwo papers accepted to NeurIPS 2026, on better rollouts for on-policy distillation and on training agents to scale inference-time reasoning.
Two papers accepted to EMNLP 2026, on adaptive-length infilling for diffusion language models and calibration data selection for LLM pruning.
Two papers accepted to COLM 2026, on structured pruning for LLM compression and data-efficient LLM adaptation.
Two papers accepted to ICML 2026, on in-context demonstration selection and dynamic tool selection for agentic reasoning.
T-RAG won the Outstanding Award at the Logical Reasoning of LLM Workshop @ ICLR 2026.
Two papers accepted to ACL 2026, on online rollout pruning for RLVR (Oral) and retrieval-augmented generation over tables (Findings).
Excited to join Amazon Rufus Foundation Models as an Applied Scientist Intern.
One paper accepted to ICLR 2026, on influence-preserving proxies for data selection in LLM fine-tuning.
AutoTool won the Best Paper Award at the MMRAgI Workshop @ ICCV 2025.
Started my Ph.D. at UIUC.
ReCODE was selected as a Best Short Paper Nominee at SIGIR 2024.
One paper accepted to RecSys 2024, on learning re-ranking models at serving time without user feedback.
Three papers accepted to SIGIR 2024, on long-term exploration in recommendation, repeat consumption modeling and legal document retrieval.
One paper accepted to AAAI 2024, on credit assignment for multi-agent RL.
One paper accepted to KDD 2023, on controllable multi-objective re-ranking.
Publications
* Equal contribution · Also on Google Scholar
Training Agent to Scale Inference-Time Reasoning
Reinforcement Learning & Reasoning
5 papersImproving how LLMs and agents reason, through reinforcement learning, distillation, tool use and retrieval.
-
NeurIPS 2026
Proposing Better Rollouts for On-Policy Distillation
-
NeurIPS 2026
Training Agent to Scale Inference-Time Reasoning
- ICML 2026
- ACL 2026 Findings
- arXiv Preprint
Data-Centric Learning
4 papersMeasuring and selecting the data that LLMs learn from, for fine-tuning, in-context learning and adaptation.
- ICML 2026
- ICLR 2026
-
EMNLP 2026 Findings
Learning What Matters: End-to-End Adaptive Calibration Data Selection for LLM Pruning
-
COLM 2026
EvoSelect: Data-Efficient LLM Evolution for Targeted Task Adaptation
Efficient LLMs
3 papersMaking LLMs cheaper to train and run, from rollout pruning in RL to structured pruning and faster decoding.
-
EMNLP 2026
Predict, Don't Iterate: Efficient Adaptive-Length Infilling for Diffusion Language Models
- COLM 2026
- ACL 2026
Recommendation, Retrieval & Multi-Agent RL
8 papersReinforcement learning for recommender systems, re-ranking, information retrieval and multi-agent credit assignment.
- SIGIR 2024
- AAAI 2024
- SIGIR-AP 2023
- KDD 2023
- RecSys 2024
- SIGIR 2024
- SIGIR 2024
- WWW 2023
Experience & Education
Industry
-
May-Dec 2026AmazonApplied Scientist Intern · Rufus Foundation ModelsEvolving long-term agentic reasoning · Mentor: Dr. Zheng Li -
2021-2022AlibabaResearch Intern · Taobao GroupRL for private-domain recommendation · Mentor: Dr. Yuan Wang
Education
-
Since 2024University of Illinois Urbana-ChampaignPh.D. in Computer ScienceAdvisor: Prof. Jingrui He -
2020-2023Renmin University of ChinaM.Eng. in Big Data Science & EngineeringAdvisor: Prof. Jun Xu -
2016-2020Tsinghua UniversityB.A. in Japanese
Service
- Conference reviewer
- NeurIPS · ICML · ICLR · ACL · EMNLP · CVPR · AAAI · WWW · DAI · IEEE DSAA
- Journal reviewer
- TNNLS · TKDD · TIST