Hi, I'm Linyu

I’m Linyu Wu, a Computer Science PhD student at the National University of Singapore (President’s Graduate Fellowship), advised by Jiaheng Zhang. Before Singapore I studied CS in the ACM Honors Class at Shanghai Jiao Tong University.

My current focus is multimodal reinforcement learning — getting unified multimodal models to reason and act through RL. More broadly, I want to make next-generation AI systems efficient, robust, and trustworthy.

Research

• Multimodal reinforcement learning — my current focus. I build UniRL, an open-source RL framework for unified multimodal models, developed at Tencent Hunyuan, where I’m a research intern (Project UP) working with Tianyu Pang and Haonan Wang.
• Diffusion language models. With collaborators: faster self-evaluation for dLLMs (ACL 2026), revisable decoding beyond hard masks (ACL 2026), and DMark, the first watermarking approach built for dLLMs (under review).
• Efficient systems & agent safety. DeServe (affordable, decentralized LLM inference) and Progent (programmable privilege control for LLM agents), from working with Dawn Song at UC Berkeley.

Building

I understand systems by building them from scratch — an OS kernel (Rust, ~10k LoC), a RISC-V CPU (Verilog/FPGA), and a compiler (Kotlin), each ranked first in its course. I also build tools I want to use, like Peninsula, a Dynamic-Island-style window switcher for macOS (400+ stars).

This blog is where I write up what I’m learning — mostly reinforcement learning and systems. Read the posts below, or see my full CV.