Publications

You can also find my articles on my Google Scholar profile

Research Outlines

My main research interests lie in the following three areas:

• Evolving AI Systems. Lifelong model editing, parameter generation and understanding, and memory management for foundation models, including model editing, unlearning, merging, and generation. Recursive self-improving embodied agentic systems.

• Agentic Intelligence. Multimodal and embodied agent systems, agentic reinforcement learning, and self-evolving agents.

• Collaborative Deep Learning. Human–AI collaborative learning, proactive user-centric agent systems, personalized deep learning, and distributed edge–cloud learning under heterogeneous data. I am also interested in their generalization, personalization, and training dynamics, as well as related deep learning phenomena and socio-technical issues.

Main Publications

* indicates equal contributions, # indicates corresponding author, † indicates mentorship.

2026

• WorldLines: Benchmarking and Modeling Long-Horizon Stateful Embodied Agents
  Yehang Zhang, Jianchong Su, Haojian Huang, Yifan Chang, Tianhao Zhou, Xinli Xu, Yingjie Xu, Yinchuan Li, Zexi Li†, Ying-Cong Chen#
  EMNLP 2026 Main.
• IntentRL: Training Proactive User-intent Agents for Open-ended Deep Research via Reinforcement Learning
  Haohao Luo*, Zexi Li*, Yuexiang Xie, Wenhao Zhang, Yaliang Li, Ying Shen†
  International Conference on Machine Learning (ICML) 2026.
• AnyEdit++: Adaptive Long-Form Knowledge Editing via Bayesian Surprise
  Bowen Tian, Caixue He, Jiemin Wu, Jingying Wang, Wenshuo Chen, Zexi Li†, Yutao Yue#
  International Conference on Machine Learning (ICML) 2026.
• Sparsity Curse: Understanding RLVR Model Parameter Space from Model Merging
  Chenrui Wu*, Zexi Li*, Jiajun Bu, Jiangchuan Liu, Haishuai Wang#
  Proceedings of the 32nd ACM SIGKDD Conference, KDD 2026.
• Tina: A Diffusion Neural Network for Generating Personalized AI Models from Text Prompts
  Zexi Li*#, Lingzhi Gao*, Dongqi Cai, Nicholas D. Lane#, Chao Wu#
  Patterns, Cell Press (Cover Invitation).
• Editing as Unlearning: Are Knowledge Editing Methods Strong Baselines for Large Language Model Unlearning?
  Zexi Li*#, Xiangzhu Wang*, William F. Shen, Meghdad Kurmanji, Xinchi Qiu, Dongqi Cai, Chao Wu#, Nicholas D. Lane#
  AAAI 2026 AI Alignment Track; NeurIPS 2025 Workshops on LLM Evaluation & Lock-LLM.
• R³L: Reflect-then-Retry Reinforcement Learning with Language-Guided Exploration, Pivotal Credit, and Positive Amplification
  Weijie Shi, Yanxi Chen, Zexi Li, Xuchen Pan, Yuchang Sun, Jiajie Xu, Xiaofang Zhou, Yaliang Li
  EMNLP 2026 Main.

2025

• Improving Model Fusion by Training-time Neuron Alignment with Fixed Neuron Anchors
  Zexi Li, Zhiqi Li, Jie Lin, Tao Shen, Jun Xiao, Yike Guo#, Tao Lin#, Chao Wu#
  IEEE Transactions on Pattern Analysis and Machine Intelligence. [arxiv]
• Editing as Unlearning: Are Knowledge Editing Methods Strong Baselines for Large Language Model Unlearning?
  Zexi Li*†, Xiangzhu Wang*, William F. Shen, Meghdad Kurmanji, Xinchi Qiu, Dongqi Cai, Chao Wu#, Nicholas D. Lane#
  AAAI 2026; NeurIPS 2025 Workshops Lock-LLM & LLM Evals. [arxiv]
• FedGuCci: Making Local Models More Connected in Landscape for Federated Learning
  Zexi Li*, Jie Lin*, Zhiqi Li*, Didi Zhu, Tao Shen, Tao Lin#, Chao Wu#, Nicholas D. Lane
  ACM KDD 2025. [arxiv]
• Towards Universal Personalization in Federated Learning via Collaborative Foundation Generative Models
  Chenrui Wu*, Zexi Li*, Fangxin Wang, Hongyang Chen, Jiajun Bu, and Haishuai Wang#
  IEEE Transactions on Mobile Computing. [paper]
• FediOS: Decoupling Orthogonal Subspaces for Personalization in Feature-skew Federated Learning
  Lingzhi Gao*, Zexi Li*#†, Yang Lu, and Chao Wu#
  Machine Learning. [arxiv]
• You Are Your Own Best Teacher: Achieving Centralized-level Performance in Federated Learning under Heterogeneous and Long-tailed Data
  Shanshan Yan, Zexi Li†, Chao Wu, Meng Pang, Yang Lu#, Yan Yan, Hanzi Wang
  ICCV 2025. [arxiv]
• Text2Weight: Bridging Natural Language and Neural Network Weight Spaces
  Bowen Tian*, Wenshuo Chen*, Zexi Li†, Songning Lai, Jiemin Wu, Yutao Yue
  ACM MM 2025.
• Merging LoRAs like Playing LEGO: Pushing the Modularity of LoRA to Extremes Through Rank-Wise Clustering
  Ziyu Zhao, Tao Shen, Didi Zhu, Zexi Li, Jing Su, Xuwu Wang, Kun Kuang, Fei Wu
  ICLR 2025. [arxiv]

2024

• Text-to-Model: Text-Conditioned Neural Network Diffusion for Train-Once-for-All Personalization
  Zexi Li*, Lingzhi Gao*, Chao Wu#
  preprint. [arxiv]
• WISE: Rethinking the Knowledge Memory for Lifelong Model Editing of Large Language Models
  Peng Wang*, Zexi Li*, Ningyu Zhang#, Ziwen Xu, Yunzhi Yao, Yong Jiang, Pengjun Xie, Fei Huang, Huajun Chen#
  NeurIPS 2024. [arxiv]
• Model Tailor: Mitigating Catastrophic Forgetting in Multi-modal Large Language Models
  Didi Zhu, Zhongyi Sun, Zexi Li, Tao Shen, Ke Yan, Shouhong Ding, Kun Kuang#, Chao Wu#
  ICML 2024. [arxiv]
• Neural Collapse Anchored Prompt Tuning for Generalizable Vision-Language Models
  Didi Zhu, Zexi Li, Min Zhang, Junkun Yuan, Yunfeng Shao, Yinchuan Li, Jiashuo Liu, Kun Kuang, Chao Wu#
  ACM KDD 2024. [arxiv]
• OpenFedLLM: Training Large Language Models on Decentralized Private Data via Federated Learning
  Rui Ye, Wenhao Wang, Jingyi Chai, Dihan Li, Zexi Li, Yinda Xu, Yaxin Du, Yanfeng Wang, Siheng Chen
  ACM KDD 2024. [arxiv]

2023

• Revisiting Weighted Aggregation in Federated Learning with Neural Networks
  Zexi Li, Tao Lin#, Xinyi Shang, and Chao Wu#
  ICML 2023. [paper][github][arxiv]
• No Fear of Classifier Biases: Neural Collapse Inspired Federated Learning with Synthetic and Fixed Classifier
  Zexi Li, Xinyi Shang, Rui He, Tao Lin#, and Chao Wu#
  ICCV 2023. [paper][arxiv]
• Universal Domain Adaptation via Compressive Attention Matching
  Didi Zhu*, Yinchuan Li*, Junkun Yuan, Zexi Li, Kun Kuang#, Chao Wu#
  ICCV 2023. [paper]
• Edge-cloud Collaborative Learning with Federated and Centralized Features
  Zexi Li*, Qunwei Li*, Yi Zhou, Wenliang Zhong#, Guannan Zhang, and Chao Wu#
  SIGIR 2023. [paper][arxiv]
• Learning Cautiously in Federated Learning with Noisy and Heterogeneous Clients
  Chenrui Wu*, Zexi Li*†, Fangxin Wang#, and Chao Wu#
  (Oral) ICME 2023. [paper][arxiv] –

2022

• Towards Effective Clustered Federated Learning: A Peer-to-peer Framework with Adaptive Neighbor Matching
  Zexi Li, Jiaxun Lu, Shuang Luo, Didi Zhu, Yunfeng Shao, Yinchuan Li, Zhimeng Zhang, Yongheng Wang#, and Chao Wu#
  • Long version in IEEE Transactions on Big Data. (JCR Q1, IF: 7.2) [paper][arxiv]
  • Short version in MLSys2022 Workshop. [poster]
• Can We Share Models If Sharing Data Is Not an Option?
  Zexi Li, Feng Mao#, and Chao Wu#
  Patterns, Cell Press. (JCR Q1, IF: 6.5) [paper]

2021

• Boosting the generalization ability of Vis-NIR-spectroscopy-based regression models through dimension reduction and transfer learning
  Xiaoli Li*, Zexi Li*, Xufeng Yang, and Yong He#
  Computers and Electronics in Agriculture. (JCR Q1, CAS Top, IF: 8.3) [paper]
• An early assessment of the County Medical Community reform in China: a case study of Zhejiang province
  Chao Wu, Yixin Tu, Zexi Li, Jianxing Yu#
  Journal of Chinese Governance. (SSCI, JCR Q1, IF: 3.0) [paper]