About Me
I am a third-year CS PhD candidate at Cornell University, advised by Prof. Claire Cardie. In summer 2026, I was a research intern at Tongyi Lab, Alibaba Group in Bellevue, WA, working on agent harness and post-training. I obtained my MS at Tsinghua University supervised by Prof. Yong Jiang, and my BS at the CS department of Xidian University, where I was supervised by Prof. Yuesheng Xu.
My research focuses on large language models — particularly LLM reasoning (self-play, RL post-training), robustness under noisy or adversarial inputs, and efficient training and serving.
I'm always open to collaboration — feel free to drop me an email!
📍 Bill and Melinda Gates Hall, G23 · Campus Rd, Ithaca, NY 14853
News
[Jul 2026] New preprint: MANTA: Multi-Agent Network Topology Adaptation for Self-Evolving Multi-Agent Systems — letting multi-agent systems rewire their own communication topology at inference time. Read the paper →
[Jun 2026] 🎓 Passed my A exam (admission to candidacy) and am now a PhD candidate at Cornell University!
[Jun 2026] Released HARP — efficient data selection for finetuning large language models via a hierarchical node–leaf structure and empirical Bayes posteriors. Read the paper →
[May 2026] Our Research Arena paper — How Far Are We From True Auto Research? — is now on arXiv.
[Apr 2026] Released Research Arena — an evaluation of CLI-based AI agents (Claude Code, Codex, Kimi Code) on end-to-end scientific research across 13 CS domains. Read the analysis →
[Apr 2026] Two papers accepted to ACL 2026 Findings: Better LLM Reasoning via Dual-play and GSM-Noise: Exploring and Enhancing Large Language Models' Reasoning under Noisy Inputs. See you in San Diego!
[Nov 2025] We released Better LLM Reasoning via Dual-play, a framework that trains two nearly identical models adversarially, enabling them to co-evolve through iterative competition without external supervision. Read the blog post →
Publications
Preprints
-
MANTA: Multi-Agent Network Topology Adaptation for Self-Evolving Multi-Agent Systems
-
HARP: Efficient Data Selection for Finetuning Large Language Models
-
Bootstrapping Post-training Signals for Open-ended Tasks via Rubric-based Self-play on Pre-training Text
-
AppellateGen: A Benchmark for Appellate Legal Judgment Generation
-
Towards More Robust Retrieval-Augmented Generation: Evaluating RAG Under Adversarial Poisoning Attacks
-
Graph-based Multi-Task Learning Framework for Model Generalization across Network Topologies
2026
-
GSM-Noise: Exploring and Enhancing Large Language Models' Reasoning under Noisy InputsFindings of ACL, 2026
-
HumanGuideNet: Adapter-Based Alignment of Deep Neural Networks with Human Similarity JudgmentsWACV, Mar 2026
-
HAPO: Training Language Models to Reason Concisely via History-Aware Policy OptimizationAAAI, Jan 2026
2025
-
Learning-Enhanced High-Throughput Pattern Matching Based on Programmable Data PlaneUSENIX ATC, July 2025
2024
-
Quantized Side Tuning: Fast and Efficient Tuning of Quantized Large Language Models ★ Outstanding PaperACL, Aug 2024 · Oral presentation
-
SpecInfer: Accelerating Large Language Model Serving with Tree-Based Speculative Inference and VerificationASPLOS, July 2024
-
-
2023
* denotes equal contribution
Education
Cornell University
Ph.D. in Computer Science · Advisor: Prof. Claire Cardie
2024 — Present
Tsinghua University
M.S. in Computer Science and Engineering · GPA 3.88/4.0
2021 — 2024
Xidian University
B.S. in Software Engineering · GPA 3.86/4.0 · Rank 2/110
2017 — 2021
Experience
Tongyi Lab, Alibaba Group · Bellevue, WA
Research Intern
Agent harness and post-training
May 2026 — Aug 2026
Tencent, Shenzhen
Research Scientist Intern
XDP and eBPF algorithm design
Sep 2022 — Aug 2023
Awards
Talks
• Cornell NLP Seminar: Quantized Side Tuning: Fast and Efficient Tuning of Quantized Large Language Models. Sep 2024.
• Tencent: Regular Expressions in the Data Plane. Dec 2023.
Academic Service
• ICLR Reviewer 2025
• AAAI Programme Committee 2025
• NeurIPS Reviewer 2024, 2025
• COLM Reviewer 2025