All-Optical Deep Learning with Quantum Nonlinearity
* Equal contribution.
All-optical deep learning leveraging quantum nonlinearity; collaborative work on optical computing.
Yutian Tao
Hi, I’m Yutian Tao, a final-year Ph.D. student in Computer Science at the University of Wisconsin–Madison, advised by Prof. Eftychios Sifakis. Before my Ph.D., I earned my B.S. in Computer Science from Nanjing University. My research centers on physics-based simulation and numerical methods, with a focus on building fast, robust solvers for discretized PDE systems that arise in graphics and physical modeling. In particular, I work on Stokes and other saddle-point problems, multigrid and Krylov methods, and efficient simulation pipelines.
I enjoy working at the boundary of theory and systems by developing algorithms that are mathematically grounded and engineering them into implementations that scale and behave reliably in practice. I’m also interested in how physics-based simulation can be applied to robotics, and in how large language models (LLMs) and their extensions may reshape the way we learn, create, and work.
Research
Interpolated Adaptive Linear Reduced Order Modeling for Deformation Dynamics
Adaptive reduced-order models for deformation dynamics using interpolated linear bases.
Retrieval-augmented generation as noisy in-context learning: A unified theory and risk bounds
Theoretical framework connecting RAG to noisy in-context learning with risk bounds.
A symmetric multigrid-preconditioned Krylov subspace solver for Stokes equations
Symmetric multigrid preconditioning for Krylov solvers applied to Stokes flow in graphics and simulation.
A computer based facial flaps simulator using projective dynamics
Facial flap surgery simulator built with projective dynamics for real-time, stable biomechanical simulation.
Long-term results of the Murawski unilateral cleft lip repair
Long-term clinical outcomes of unilateral cleft lip repair; supporting simulation and validation pipeline.
Optimized processing of localized collisions in projective dynamics
Efficient handling of localized collisions in projective dynamics for surgical and graphics applications.
Experience
Work / Research Experience
Meta
Jun 2025 – Oct 2025
Research Scientist Intern · Reality Labs
Built an AI-driven simulation pipeline with reduced-order modeling (300k+ soft-body samples) and a physics-informed autoencoder for 1k× compression and 70% lower simulation error.
Foundation LLM
Jan 2024 – Jan 2025
Applied Research Intern
Built Marching Cubes voxelization and mesh QC for 3D asset validation, focusing on watertight mesh generation and removing 87% of malformed meshes for simulation-conditioned vision models.
Microsoft Research Asia
Jul 2020 – Dec 2020
Research Intern · Internet Graphics
Ported and profiled the Incremental Potential Contact (IPC) algorithm in modern C++, uncovering bottlenecks and outlining optimizations for graphics research.