Yukuan Zhang

I am a Computer Science Ph.D. student at the University of Central Florida, working at the intersection of privacy-preserving computation and large language models. My research asks a simple question: can we make secure machine learning fast enough to be useful at scale?

I build systems that sit across secure multi-party computation (MPC), LLM inference, and high-performance computing — from cryptographic routing inside transformers to deployment on GPU clusters. I believe small, sharp ideas compound.

Publications

• SecureRoute: Input-Adaptive Encrypted Routing for Efficient and Accurate MPC-Based Transformer Inference
  Yukuan Zhang, Mengxin Zheng, Qian Lou
  Proceedings of the 63rd ACM/IEEE Design Automation Conference (DAC), 2026

Experience

Education

Skills

C++, Python, Bash scripting, PyTorch, GPU cluster tuning, and secure multi-party computation frameworks (CrypTen, SPDZ, SecretFlow).

• Focused on frontier AI and privacy-preserving computation — combining secure multi-party computation (MPC) with large language models (LLMs) to drive algorithmic research.
• Comfortable in high-performance computing environments — scripting, deploying, and running large-scale jobs on HPC clusters such as Newton.
• Full-stack troubleshooting — from server deployment (e.g., Aliyun) to algorithm-level implementation, with strong system-level debugging and software engineering skills.

Interests

Outside my core research I keep a close eye on the fast-moving open-source AI agent ecosystem. Two projects I follow and experiment with:

• OpenEvolve — an open-source implementation of DeepMind's AlphaEvolve. It turns LLMs into autonomous code optimizers through an evolutionary loop (prompt sampler, LLM ensemble, evaluator pool, program database), and I'm interested in how this paradigm can push algorithm discovery in secure ML kernels.
• OpenClaw — a self-hosted gateway that bridges everyday messaging apps (Slack, Telegram, Discord, iMessage, Matrix, and more) to AI coding agents. I like the "own your agent" philosophy and the idea of meeting assistants where users already are.

Beyond these, I'm generally drawn to reinforcement learning, LLM routing, and software-engineering agents — and to the academic conversations around them.