钱宇阳 Yu-Yang Qian (Y.-Y. Qian) Ph.D. Student, LAMDA Group Department of Computer Science and Technology National Key Laboratory for Novel Software Technology Nanjing University, Nanjing 210023, China Supervisor: Professor Zhi-Hua Zhou Email: qianyy@lamda.nju.edu.cn Laboratory: Computer Science Building, Xianlin Campus of Nanjing University [ Google Scholar, Github ]

Supervisor

      Professor Zhi-Hua Zhou

Biography

• Sep 2016 - Jun 2020 : Receive my B.Sc. degree in School of Information and Communication Engineering, University of Electronic Science and Technology of China.

• Sep 2020 : Admitted to study for a M.Sc. degree in Nanjing University without entrance examination under the guidance of professor Zhi-Hua Zhou.

• Sep 2023 - Now : Curretnly I am a Ph.D. student of School of Artificial Intelligence in Nanjing University and a member of LAMDA Group, where I was very fortunate to be advised by Prof. Zhi-Hua Zhou and Prof. Yuan Jiang.

Research Interests

      My research interests include Machine Learning and Data Mining. Most recently, I am interested in:

• Efficient Machine Learning in Non-stationary and Open World Environments;

• Efficient Post-training for LLMs, including efficient Continual Fine-tuning, RLHF, and Agentic RL;

• Diffusion Large Language Models (dLLMs).

Preprints

• When Drafts Evolve: Speculative Decoding Meets Online Learning. [arXiv] [code] [bibtex]

  Yu-Yang Qian, Hao-Cong Wu, Yichao Fu, Hao Zhang, and Peng Zhao.

  In: Proceedings of the 14th International Conference on Learning Representations (ICLR 2026) Workshop on Lifelong Agents: Learning, Aligning, Evolving, Rio de Janeiro, Brazil. 2026.




• d3LLM: Ultra-Fast Diffusion LLM using Pseudo-Trajectory Distillation. [arXiv] [code] [🤗model] [leaderboard] [bibtex]

  Yu-Yang Qian, Junda Su, Lanxiang Hu, Peiyuan Zhang, Zhijie Deng, Peng Zhao, and Hao Zhang.




• Internalizing Agency from Reflective Experience. [arXiv] [bibtex]

  Rui Ge, Yichao Fu, Yu-Yang Qian, Junda Su, Yiming Zhao, Peng Zhao, and Hao Zhang.

Publications

• Provably Efficient Online RLHF with One-Pass Reward Modeling. [paper] [code] [bibtex]

  Long-Fei Li*, Yu-Yang Qian*, Peng Zhao, and Zhi-Hua Zhou. (* indicates equal contribution)

  In: Advances in Neural Information Processing Systems 38 (NeurIPS 2025), San Diego, California, 2025.




• Handling New Class in Online Label Shift. [paper] [code] [bibtex]

  Yu-Yang Qian, Yong Bai, Zhen-Yu Zhang, Peng Zhao, and Zhi-Hua Zhou.

  IEEE Transactions on Knowledge and Data Engineering (TKDE 2025), 37(9):5257-5270, 2025.




• TreeLoRA: Efficient Continual Learning via Layer-Wise LoRAs Guided by a Hierarchical Gradient-Similarity Tree. [paper] [code] [arXiv] [bibtex]

  Yu-Yang Qian, Yuan-Ze Xu, Zhen-Yu Zhang, Peng Zhao, and Zhi-Hua Zhou.

  In: Proceedings of the 42nd International Conference on Machine Learning (ICML 2025), Vancouver, Canada, 2025.




• Adapting to Generalized Online Label Shift by Invariant Representation Learning. [paper] [code] [bibtex]

  Yu-Yang Qian, Yi-Han Wang, Zhen-Yu Zhang, Yuan Jiang, and Zhi-Hua Zhou.

  In: Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2025), Toronto, Canada, 2025.




• Efficient Non-stationary Online Learning by Wavelets with Applications to Online Distribution Shift Adaptation. [paper] [code] [StreamingWavelet Package] [Pip Install] [bibtex]

  Yu-Yang Qian, Peng Zhao, Yu-Jie Zhang, Masashi Sugiyama, and Zhi-Hua Zhou.

  In: Proceedings of the 41st International Conference on Machine Learning (ICML 2024), Vienna, Austria, 2024.




• Learning with Asynchronous Labels. [paper] [code] [bibtex]

  Yu-Yang Qian, Zhen-Yu Zhang, Peng Zhao, and Zhi-Hua Zhou.

  ACM Transactions on Knowledge Discovery from Data (TKDD 2024), 18(8):1-27, 2024.




• Handling New Class in Online Label Shift. [paper] [code] [bibtex]

  Yu-Yang Qian*, Yong Bai*, Zhen-Yu Zhang, Peng Zhao, and Zhi-Hua Zhou.

  In: Proceedings of the 23rd IEEE International Conference on Data Mining (ICDM 2023), Shanghai, China, 2023.




• Adaptive Learning for Weakly Labeled Streams. [paper] [code] [bibtex]

  Zhen-Yu Zhang, Yu-Yang Qian, Yu-Jie Zhang, Yuan Jiang, and Zhi-Hua Zhou.

  In: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2022), Washington, DC, 2022.

Projects & Open-Source Codes

• 🫧 d3LLM — Ultra-Fast Diffusion LLM using Pseudo-Trajectory Distillation 🚀 [GitHub] [Blog]

  We introduce a novel recipe for building an ultra-fast diffusion language model named d3LLM (pseuDo-Distilled Diffusion LLM).

  Key Features: up to 10× speedup over the vanilla LLaDA / Dream, and 5× speedup over AR models (Qwen-2.5-7B-it) on H100 GPU! With negligible accuracy degradation. PyTorch implementation of the paper "d3LLM: Ultra-Fast Diffusion LLM using Pseudo-Trajectory Distillation".




• 🧑‍🍳 Online RLHF Pipeline — Cook your own RLHF recipe! [GitHub]

  This repository provides a flexible and modular code framework to Reinforcement Learning from Human Feedback (RLHF). A PyTorch implementation of the paper "Provably Efficient Online RLHF with One-Pass Reward Modeling".

  Key Features: You can cook your own Online-RLHF recipe using this repo! Customizable RLHF recipes, modular training components (SFT, RM, PPO, DPO), online learning capabilities, and quickly build up your own RLHF pipeline.




• 🌳 TreeLoRA — Efficient Continual Fine-tuning of LLMs [GitHub]

  Propose a hierarchical tree-structured LoRA adapters for efficient continual fine-tuning of large models. A PyTorch implementation of the paper TreeLoRA (ICML'25).

  Key Features: Efficiently explores task structure using bandits and optimizes parameters through sparse gradient updates, validated on various sizes of models, up to 3.2× speedup for ViTs and 2.4× speedup for LLMs.




• 📈 Streaming Wavelet Operator [Pip Install] [GitHub]

  Sequentially apply wavelet transform to a sequence efficiently in an online manner, instead of recalculation each round. A PyTorch implementation of the paper "Efficient Non-stationary Online Learning by Wavelets with Applications to Online Distribution Shift Adaptation" (ICML'24).

  Can be used for: detect environmental changes, efficiently identify changing points, and analyze variations in sequences.

Awards & Honors

• 2025, National Scholarship for PhD candidates.

• 2024, Ruli Scholarship in Nanjing University

• 2024, achieved an Excellent rating in the Doctoral Qualify Examination (Top 15% of all Ph.D. candidates)

• 2021, Dongliang Scholarship Excellent Award in Nanjing University

• 2021, the First-Class Academic Scholarship in Nanjing University

• 2020, Outstanding Undergraduate Thesis Award

• 2018, Gold Medal, ACM-CCPC (China Collegiate Programming Contest) National Invitational Contest

• 2018, First Prize, the National Mathematical Contest in Modeling

Teaching Assistant

• Advanced Optimization. (With Asst. Prof. Peng Zhao; 2022 Fall)

Academic Service

• Program Committee/Reviewer for Conferences: ICML 2022-2025, NeurIPS 2022-2025, ICLR 2024-2025, KDD 2024-2026, AAAI 2026, AISTATS 2024, UAI 2024.

Correspondence

• Email: qianyy@lamda.nju.edu.cn

• Office: Room 912, Computer Science Building, Xianlin Campus of Nanjing University

• Address: National Key Laboratory for Novel Software Technology, Nanjing University, Xianlin Campus Mailbox 603, 163 Xianlin Avenue, Qixia District, Nanjing 210023, China
  (南京市栖霞区仙林大道163号, 南京大学仙林校区603信箱, 软件新技术国家重点实验室, 210023.)