About me

Welcome! I’m Eric Liu. I am broadly interested in efficient/trustworthy AI and applications of ML.
You can find me on LinkedIn and GitHub, or reach me at eliu4913 at usc dot edu.

---

Education

University of Southern California, Viterbi School of Engineering

• B.S. Computer Science & Business Administration, Aug 2020 - Dec 2025
• M.S. Computer Science (Progressive Degree), Aug 2024 - Dec 2025
• Ph.D. Computer Science, Aug 2026 - Present

---

Publications

• Trajectory Graph Learning: Aligning with Long Trajectories in Reinforcement Learning Without Reward Design
  Yunfan Li, Eric Liu, Lin Yang
  NeurIPS 2025 Spotlight

• SimWorld: An Open-ended Simulator for Agents in Physical and Social Worlds
  Xiaokang Ye, Jiawei Ren, Yan Zhuang, Xuhong He, Yiming Liang, Yiqing Yang, Mrinaal Dogra, Xianrui Zhong, Eric Liu, Kevin Benavente, Rajiv Mandya Nagaraju, Dhruv Vivek Sharma, Ziqiao Ma, Tianmin Shu, Zhiting Hu, Lianhui Qin
  NeurIPS 2025 Spotlight

• Graph Neural Networks for Bridge Swap Link Prediction in Uniswap v3
  Qingran Zhou, Eric Liu, Alessio Brini
  ACM ICAIF 2025 Best Student Paper Honorable Mention

---

Preprints and Manuscripts

• FlashCP: Load-Balanced Communication-Efficient Context Parallelism for LLM Training
  Zheng Wang, Eric Liu, Linan Jiang, Zhongkai Yu, Zaifeng Pan, Yue Guan, Yuke Wang, Yufei Ding
  arXiv preprint (2026)

---

Research

• USC, May 2025 - Present
  With Prof. Sai Praneeth Karimireddy. Leading research on LLM unlearning and privacy auditing.

• UC San Diego, PICASSO Lab, March 2025 - Jan 2026
  With Prof. Yufei Ding. Co-developed FlashCP, a whole-document context-parallelism framework that resolves workload imbalance and achieves up to 1.63× end-to-end speed-ups over SOTA long-context training baselines.

• UCLA, May 2024 - Jan 2026
  With Prof. Lin Yang. Co-developed Trajectory Graph Learning, a framework that formulates policy alignment as an MWIS problem and enables coherent long-horizon policy learning.

• UC San Diego, SimWorld, March 2025 - Sept. 2025
  With Prof. Lianhui Qin. Worked on the experimental pipeline for an LLM-driven delivery-agent benchmark in a UE5 urban simulator, including task variants, scripts, and baselines.