# About

I'm a PhD student at Purdue University, advised by Prof. Jing Gao and Prof. Xiaoqian Wang. Most of my recent work is on collaboration between heterogeneous LLM systems, at three layers of the stack: cross-family RL post-training, latent-space communication between heterogeneous VLM agents, and supply-chain risks in cross-family merging. I also work on LLM copyright compliance, spanning detection, agent-based defense, and behavioral unlearning.

# News

• [May 2026] Selected as a Gold Reviewer for ICML 2026. Wrote up some reflections on both sides of LLM-assisted peer review this cycle.
• [May 2026] We introduce Mutual Reinforcement Learning, a framework for concurrent RL post-training in which heterogeneous LLM policies exchange typed experience while keeping separate parameters, objectives, and tokenizers.
• [Mar 2026] I will be joining Microsoft Research - Deep Learning Group this upcoming summer as a Research Intern.
• [Feb 2026] We introduce The Vision Wormhole: Latent-Space Communication in Heterogeneous Multi-Agent Systems, a text-free communication framework that uses a Universal Visual Codec to enable efficient latent-space transfer across heterogeneous VLM agents. Code is available here.
• [Jan 2026] We release tokenforge, exposing tokenizer transplant as a critical supply-chain attack surface in modular AI. We demonstrate how the standard merging pipeline can be exploited to inject "sleeper" backdoors that remain mathematically inert in donor models but activate downstream.
• [Jul 2025] SUV, a scalable copyright compliance framework for large language models (LLMs) with regularized selective unlearning, is accepted by COLM 2025.
• [May 2025] I joined AWS AI Fundamental Research Team as an Applied Scientist Intern for Summer 2025.
• [May 2025] I will present a portion of the tutorial, LLMs and Copyright Risks: Benchmarks and Mitigation Approaches, at NAACL 2025.
• [Jan 2025] FedBiscuit accepted by ICLR 2025.
• [Jan 2025] CausalEval accepted by NAACL 2025.
• [Jan 2025] Survey on Factuality in Large Language Models accepted by ACM Computing Surveys.
• [Dec 2024] GraphEval accepted by IEEE Data Engineering Bulletin.
• [Nov 2024] Listed as one of the Top Reviewers for NeurIPS 2024.
• [Sep 2024] Two papers accepted to EMNLP 2024 Main Conference: SHIELD and SaySelf.

# Recent Publications

Selected first-author and co-first-author work. # denotes equal contribution. See Google Scholar for the complete list.

1. Experience Sharing in Mutual Reinforcement Learning for Heterogeneous Language Models. Xiaoze Liu, Dhananjay Ram, Yuting Zhang, Zhaoyang Zhang, Wei Xia, Stefano Soatto, Preprint, 2026. Blog

2. The Vision Wormhole: Latent-Space Communication in Heterogeneous Multi-Agent Systems. Xiaoze Liu#, Ruowang Zhang#, Weichen Yu, Siheng Xiong, Liu He, Feijie Wu, Hoin Jung, Matt Fredrikson, Xiaoqian Wang, Jing Gao, Preprint, 2026. Blog · Code

3. The Trojan in the Vocabulary: Stealthy Sabotage of LLM Composition. Xiaoze Liu, Weichen Yu, Matt Fredrikson, Xiaoqian Wang, Jing Gao, Preprint, 2026. Blog

4. SUV: Scalable Large Language Model Copyright Compliance with Regularized Selective Unlearning. Tianyang Xu#, Xiaoze Liu#, Feijie Wu, Xiaoqian Wang, Jing Gao, The 2025 Conference on Language Modeling (COLM), 2025. Blog

5. SHIELD: Evaluation and Defense Strategies for Copyright Compliance in LLM Text Generation. Xiaoze Liu#, Ting Sun#, Tianyang Xu, Feijie Wu, Cunxiang Wang, Xiaoqian Wang, Jing Gao, The 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2024. Blog

# Invited Talks & Tutorials

1. LLMs and Copyright Risks: Benchmarks and Mitigation Approaches. NAACL 2025 Tutorials, May 2025. Delivered part of the tutorial session live on Zoom. Tutorial organized by Prof. Zhaozhuo Xu and Prof. Denghui Zhang.

2. SHIELD: Evaluation and Defense Strategies for Copyright Compliance in LLM Text Generation. NICE NLP, November 2024. Delivered talk in Mandarin on mitigating copyright violation via SHIELD.

# Education and Experience

• PhD Student, Purdue University, 2023 - Present
  • Advised by Prof. Jing Gao and Prof. Xiaoqian Wang
  • Applied Scientist Intern, AWS AI Fundamental Research Team, Summer 2025
• Master's Degree, Zhejiang University, 2020 - 2023
  • Awarded Outstanding Master's Thesis of Zhejiang Province (122 out of 35,149, Top 0.3%), and Excellent Postgraduate student's Award of Zhejiang University.
  • Research Intern, AliCloud & Alibaba DAMO Academy, 2021-2023
• Bachelor's Degree, Northeastern University, 2016 - 2020
  • Awarded the National Scholarship of China (Top 0.2%) twice during undergraduate studies, and the Outstanding Graduate of Northeastern University in 2020.
  • Intern, ByteDance, 2019 - 2020
  • Exchange Student, Kanazawa University, 2019, Supported by JASSO Scholarship

# Academic Services

Serve as a reviewer/PC for

• 2026: ICML (Gold Reviewer), NeurIPS, ACL Rolling Review, KDD
• 2025: ICLR, ICML, NeurIPS, ACL Rolling Review, KDD
• 2024: NeurIPS (Top Reviewer), ACM MM, SDM, CIKM, ISWC, ACL Rolling Review, KDD
• 2023: NeurIPS, EMNLP, KDD

Served as a Journal reviewer for Transactions on Machine Learning Research, IEEE Transactions on Pattern Analysis and Machine Intelligence, IEEE Transactions on Knowledge and Data Engineering, IEEE Transactions on Neural Networks and Learning Systems, Pattern Recognition, Information Sciences, IEEE Transactions on Big Data