CV

Linyu Wu · linyuwu@u.nus.edu · linyuwu.com · github.com/Celve · Google Scholar

Research Interests

I aim to advance the robustness, safety, and efficiency of next-generation AI systems. My current focus is multimodal reinforcement learning and diffusion large language models (dLLMs).

Education

• National University of Singapore — PhD in Computer Science, Oct 2025 – present. President’s Graduate Fellowship. Advised by Prof. Jiaheng Zhang.
• Shanghai Jiao Tong University — Honors B.Eng. in Computer Science (ACM Honors Class), Sep 2021 – Jun 2025.

Research Experience

• Tencent (Project UP) — Research Intern, Feb 2026 – present. Multimodal RL; advised by Dr. Tianyu Pang.
• Sunblaze Lab, UC Berkeley — Research Intern, Feb 2024 – Dec 2024. Decentralized ML systems and AI safety; advised by Prof. Dawn Song.
• ADB Lab, University at Buffalo — Research Intern (remote), Dec 2023 – May 2024. Federated query processing and DBMS; advised by Prof. Zhuoyue Zhao and Prof. Dong Xie.

Publications

• UniRL: A Reinforcement Learning Framework for Unified Multimodal Models. Haonan Wang, Linyu Wu, et al. Open-source framework, Tencent Hunyuan.
• Efficient Self-Evaluation for Diffusion Language Models via Sequence Regeneration. Linhao Zhong*, Linyu Wu*, et al. ACL 2026. ~32× faster than Monte Carlo evaluation.
• Beyond Hard Masks: Progressive Token Evolution for Diffusion Language Models. Linhao Zhong*, Linyu Wu*, et al. ACL 2026.
• DMark: Order-Agnostic Watermarking for Diffusion Large Language Models. Linyu Wu*, Linhao Zhong*, et al. Under review. First watermarking approach for dLLMs.
• DeServe: Towards Affordable Offline LLM Inference via Decentralization. Linyu Wu*, Xiaoyuan Liu*, et al. Decentralized pipeline parallelism; 12.6× throughput in high-latency settings.
• Progent: Programmable Privilege Control for LLM Agents. Tianneng Shi, Jingxuan He, …, Linyu Wu, et al.

* equal contribution.

Projects

• OS kernel — Rust, ~10k LoC, from scratch (ranked 1st in course).
• Compiler — Kotlin/ANTLR, ~10k LoC, from scratch (ranked 1st in course).
• RISC-V CPU — Verilog/FPGA, ~3k LoC, from scratch (ranked 1st in course).
• Peninsula — a Dynamic-Island-style window switcher, notification center, and file shelf for macOS (400+ stars).

Teaching

• Teaching Assistant, Operating Systems — Spring 2024. Main contributor to ACore, a guidebook for building a microkernel from scratch.

Skills

• Languages: Rust, C/C++, Python, Kotlin, Java, Swift, Verilog.
• Systems & ML: OS kernels, compilers, RISC-V/FPGA, distributed systems, PyTorch, CUDA.
• Tools: Git, Linux, Typst.