Qian Wang

Qian Wang

CS Ph.D. candidate @ NUS

I am a Computer Science Ph.D. candidate at the National University of Singapore, advised by Prof. Bingsheng He. I work on LLM for Finance: teaching LLMs to read financial text and turn it into investment decisions, along three threads —

  • Financial benchmarks & data. Building the datasets this field is missing, e.g., CrossAlpha (EMNLP 2026 Findings), an annual-report benchmark testing whether disclosures in one market predict stock returns in another, and EX-Graph (ICLR 2024), a pioneering dataset bridging Ethereum transactions and X accounts.
  • LLM trading agents. Agents that reason over market information to make trading decisions, e.g., CryptoTrade (EMNLP 2024) and FS-ReasoningAgent (ICLR 2025 Financial AI Workshop).
  • Trustworthy LLM judgment. Financial decisions need unbiased judges — I characterize LLM judging bias empirically (COLM 2025) and mitigate it via reinforcement learning (Treat Bias as Noise).

Before this, I worked broadly on LLM multi-agent systems (MegaAgent, ACL 2025 Findings — a 590-agent fully autonomous system) and graph learning. My work has appeared at ICLR, NeurIPS, ICML, ACL, EMNLP, COLM, and LOG, with several Oral presentations and an Outstanding Reviewer Award (EACL 2026).

I am always open to collaborations — and happy to chat about research, PhD applications, or life choices. Email me at persdre@gmail.com!

📰 News

  • 2026.08 — 🎉 CrossAlpha: An Annual-Report Benchmark for Cross-Market Factor Research was accepted to EMNLP 2026 Findings. Hope to see you all in Budapest!
  • 2026.05 — 📈 New preprint CrossAlpha: an annual-report benchmark for cross-market factor research, covering ~3,600 firms across 5 markets with ~19M firm-pair scores. arXiv
  • 2026.05 — 🎉 4 papers accepted to the ICML 2026 Agents-in-the-Wild Workshop. Looking forward to meeting you in Seoul!
  • 2026.05 — 🏆 Received the Outstanding Reviewer Award from EACL 2026.
  • 2026.05 — 🎯 Treat Bias as Noise (bias-robust LLM reasoning via reinforcement learning, with collaborators from UC Berkeley) was accepted to the ICML 2026 AI4GOOD Workshop. arXiv
  • 2026.04 — 🎉 3 papers accepted to ACL 2026 (incl. Findings).
  • 2026.01 — 🎉 LLM DNA: Tracing Model Evolution via Functional Representations was accepted to ICLR 2026 as an Oral presentation.
  • 2025.10 — 1 paper accepted to LOG 2025 as an Oral presentation.
  • 2025.09 — 🛡️ Towards Evaluating Fake Reasoning Bias in Language Models was accepted to the NeurIPS 2025 Lock-LLM Workshop.
  • 2025.07 — Our paper Assessing Judging Bias in Large Reasoning Models: An Empirical Study was accepted to COLM 2025.
  • 2025.06 — 🤝 Multiple papers accepted to ICML 2025 workshops (incl. the Multi-Agent Systems Workshop), covering multi-agent systems and LLMs for finance.
  • 2025.05 — Our paper MegaAgent: A Large-Scale Autonomous LLM-based Multi-Agent System Without Predefined SOPs was accepted to ACL 2025 Findings.

📝 Selected Publications

Click a topic to filter; my name is shown in bold. The complete list is on my Google Scholar

ACL 2025 Findings ICLR 2025 FM-Wild Workshop Oral MegaAgent: A Large-Scale Autonomous LLM-based Multi-Agent System Without Predefined SOPs An autonomous multi-agent framework that decomposes tasks, spawns agents on the fly, and coordinates up to 590 of them — no predefined SOPs. Paper Code Qian Wang, T. Wang, Z. Tang, Q. Li, N. Chen, J. Liang, B. He
ICML 2025 Multi-Agent Systems Workshop AgentTaxo: Dissecting and Benchmarking Token Distribution of LLM Multi-Agent Systems Dissects where tokens actually go inside LLM multi-agent systems and benchmarks their token efficiency. Paper Qian Wang, Z. Tang, N. Chen, T. Wang, B. He
ACL 2026 Findings Diversity Collapse in Multi-Agent LLM Systems: Structural Coupling and Collective Failure in Open-Ended Idea Generation Shows how tightly-coupled multi-agent LLM discussion collapses open-ended idea generation into the same few ideas. N. Chen, Y. Tong, Y. Yang, X. Zhang, Qian Wang, B. He
COLM 2025 Assessing Judging Bias in Large Reasoning Models: An Empirical Study An empirical audit of large reasoning models as judges, uncovering systematic bandwagon, authority, and position biases. Paper Qian Wang, Z. Lou, Z. Tang, N. Chen, X. Zhao, W. Zhang, D. Song, B. He
NeurIPS 2025 Lock-LLM Workshop Towards Evaluating Fake Reasoning Bias in Language Models Shows LLM judges reward text that merely looks like reasoning, and builds a benchmark to measure this bias. Paper Qian Wang, Z. Tang, Z. Lou, N. Chen, W. Wang, B. He
Preprint 2025 JudgeLRM: Large Reasoning Models as a Judge Trains large reasoning models into better judges with judgment-oriented reinforcement learning. Paper N. Chen, Z. Hu, Q. Zou, J. Wu, Qian Wang, B. Hooi, B. He
ICML 2026 AI4GOOD Workshop Treat Bias as Noise: Training Bias-Robust LLM Reasoning via Reinforcement Learning RL training that treats biased cues as noise, yielding LLM reasoning that stays robust on biased prompts. Paper Qian Wang, X. Zhao, Z. Zhang, Z. Lou, N. Chen, D. Song, B. He
Preprint 2026 Learning to Learn-at-Test-Time: Language Agents with Learnable Adaptation Policies Language agents with learnable policies that decide how to adapt themselves at test time. Paper Z. Lou, H. Chen, Y. Li, Qian Wang, B. Hooi
Preprint 2026 RL-RIG: A Generative Spatial Reasoner via Intrinsic Reflection A generative spatial reasoner that learns to reflect on its own intermediate steps via reinforcement learning. Paper T. Wang, Z. Ma, Qian Wang, X. Zhang, X. Long, B. Zhou
ICLR 2026 Oral LLM DNA: Tracing Model Evolution via Functional Representations Gives every LLM a functional 'DNA' embedding, making model lineage and evolution traceable across families. Paper Z. Wu, H. Zhao, Z. Wang, J. Guo, Qian Wang, B. He
Preprint 2026 LLM Agent Memory: A Survey from a Unified Representation–Management Perspective A survey unifying LLM agent memory research under one representation–management framework. Paper Z. Tang, X. He, T. Zhao, F. Wei, X. Liu, P. Dong, Qian Wang, et al.
ICML 2026 Agents-in-the-Wild Workshop Parameters as Agentic Memory: Internalizing Long-Horizon Memories for Efficient LLM Agents Internalizes an agent's long-horizon memory into model parameters instead of ever-growing context. Z. Tang, F. Wei, P. Dong, X. Liu, Qian Wang, X. Chu, B. Li
EMNLP 2026 Findings CrossAlpha: An Annual-Report Benchmark for Cross-Market Factor Research A public annual-report benchmark testing whether firm disclosures in one market predict stock returns in another. Paper Qian Wang, Z. Tong, N. Chen, Z. Wu, B. He
ICLR 2025 Financial AI Workshop Exploring LLM Cryptocurrency Trading Through Fact-Subjectivity Aware Reasoning Splits crypto-trading reasoning into factual and subjective paths, improving LLM trading decisions. Paper Code Qian Wang, Y. Gao, Z. Tang, B. Luo, N. Chen, B. He
EMNLP 2024 CryptoTrade: A Reflective LLM-based Agent to Guide Zero-shot Cryptocurrency Trading A reflective LLM agent that fuses on-chain and off-chain signals for zero-shot cryptocurrency trading. Paper Code Y. Li, B. Luo, Qian Wang, N. Chen, X. Liu, B. He
ICLR 2024 EX-Graph: A Pioneering Dataset Bridging Ethereum and X The first public dataset linking Ethereum transaction wallets with X (Twitter) accounts. Paper Code Qian Wang, Z. Zhang, Z. Liu, S. Lu, B. Luo, B. He
LOG 2025 Oral Less is More: Using Buffer Nodes to Reduce Excessive Majority Node Influence in Class Imbalance Graphs Inserts buffer nodes into graphs to damp excessive majority-class influence in imbalanced node classification. Paper Qian Wang, Z. Liu, Z. Zhang, B. Luo, B. He
NeurIPS 2024 Datasets & Benchmarks Multi-Chain Graphs of Graphs: A New Approach to Analyzing Blockchain Datasets A graphs-of-graphs dataset spanning multiple blockchains, enabling cross-chain analysis. Paper B. Luo, Z. Zhang, Qian Wang, B. He
IEEE TKDE 2025 A Survey of Imbalanced Learning on Graphs: Problems, Techniques, and Future Directions A systematic survey of imbalanced learning on graphs: problems, techniques, and future directions. Paper Z. Liu, Y. Li, N. Chen, Qian Wang, B. Hooi, B. He
Preprint 2025 LLM-based Human Simulations Have Not Yet Been Reliable Argues LLM-based human simulations are not yet reliable, and maps out where and why they fail. Paper Qian Wang, J. Wu, Z. Tang, B. Luo, N. Chen, W. Chen, B. He
ICLR 2025 Blogposts Can LLM Simulations Truly Reflect Humanity? A Deep Dive A deep dive into whether LLM simulations truly reflect human behavior. Paper Qian Wang, Z. Tang, B. He

🏆 Awards

🎤 Invited Talks & Interviews

  • 2025, Invited talk at Qube Research & Technologies (Singapore office) on leveraging LLMs to make fair and unbiased judgments about factors.
  • 2025, Invited talk at AI4X 2025 on utilizing LLMs to make trading decisions in the cryptocurrency market.
  • 2025, Invited talk on LLM-based human simulations at Renmin University of China, hosted by Yunhai Wang. Slides
  • 2025, Invited talk by AI Time on my ICLR 2025 BlogPost Can LLMs Truly Simulate Humanity? A Deep Dive. Video
  • 2024, Interviewed by the Open Source Promotion Plan (OSPP), a summer program organized by the Institute of Software, Chinese Academy of Sciences. Interview
  • 2023, Invited talk on EX-Graph at The Hong Kong University of Science and Technology (Guangzhou). Slides

📖 Education

  • 2023 - Now, Ph.D. Student, Computer Science, National University of Singapore
  • 2019 - 2022, Bachelor, Computer Science with a Minor in Economics, National University of Singapore
  • 2017 - 2019, Undergraduate, Chemistry, Shanghai Jiao Tong University

💼 Industry Experience

  • 2026.01 - Now, Quant Researcher, MS Capital, Singapore
  • 2022, Algorithm Tutor, OI Wiki, GitHub
  • 2021, Backend Engineer Intern, Ant Group, Shanghai
  • 2020, Python Tutor, InterMine, University of Cambridge

💬 Projects

  • SurviveSJTU Manual: I authored a chapter for the latest edition of the SurviveSJTU Manual — an online survival guide for students at Shanghai Jiao Tong University. You can browse the manual and my contributions on GitHub.

🔥 Service

  • 2025, ICAIF 2025 Secure FinAI Contest Local Organizer
  • 2023 - Now, Seminar Organizer, Xtra Lab, NUS
  • 2023 - Now, Fire Warden, School of Computing, NUS