Zujing Liu

Zujing Liu

ByteDance

Hi there πŸ‘‹. This is Liu Zujing (刘η₯–靖). Currently, I am working as a Research Engineer at ByteDance. I received my Master’s degree from Wuhan University, advised by Prof.Gui-song Xia and Prof.Yuan Gao at Captain. I completed my undergraduate study in Computer Science and Technology at Wuhan University and received an honor degree from Hongyi Honor College.

Currently, I’m working on Multimodal Large Language Models (MLLMs) and AI Agents with Reinforcement Learning. Before that, I conducted research on MLLM reasoning as well as Efficient LLM/MLLM.

πŸ”₯ News

  • 2026.07: Β πŸ€“πŸ€“ Great honor to join ByteDance.
  • 2026.06 Β πŸŽ“πŸŽ“ Farewell WHU! Cherishing my great time here.
  • 2025.05: Β πŸŽ‰πŸŽ‰ One papers got accepted by ACL 2025.
  • 2025.04: Β πŸ₯ΊπŸ₯Ί Soooo lucky to join ByteDance as a Research Intern.

πŸ“ Publications

Arxiv
sym

Bypass Back-propagation: Optimization-based Structural Pruning for Large Language Models via Policy Gradient (ACL 2025 Main Conference)

Yuan Gao*, Zujing Liu*, Weizhong Zhang*, Bo Du, Gui-Song Xia

  • A novel optimization-based structural pruning method for LLM that optimizes directly for model loss.
  • Learning pruning mask using Bernoulli distribution via policy gradient estimator to avoid back-propagation.
  • Superior performance compared to existing techniques, with efficient computation on a single GPU.
Arxiv
sym

On the Faithfulness of Visual Thinking: Measurement and Enhancement (Arxiv Preprint)

Zujing Liu*, Junwen Pan*, Qi She, Yuan Gao, Gui-Song Xia

  • Investigate the faithfulness of multimodal chain-of-thought (MCoT) in LVLMs, and propose an MCoT faithfulness evaluation framework.
  • Propose a reinforcement learning method based on Sufficient-Component Cause Model (SCCM) modeling to improve MCoT reasoning faithfulness.

* Equal contribution

πŸŽ– Honors and Awards

  • 2023-2026(M.Eng.) Second-Class Academic Scholarship of Wuhan University, DiDi Scholarship
  • 2019-2023(B.Eng.) Second-Class Academic Scholarship of Wuhan University (Top 10%), Outstanding Students of Wuhan University (Top 10%)

πŸ“– Educations

  • 2023.09 - 2026.06, M.Eng. in Computer Science, School of Computer, Wuhan University
  • 2019.09 - 2023.06, B.Eng. in Computer Science, School of Computer, Wuhan University

πŸ’» Internships

  • 2025.05 - 2025.10, Business Integrity at ByteDance, Research Intern.