Jui-Hsien Wang

PhD Candidate, Stanford University
Institute for Computational and Mathematical Engineering (ICME)

Email: jui-hsien.wang at stanford dot edu
Office: 376 Gates 3B Wing, Stanford CA
Curriculum Vitae
About me

I am a fourth-year PhD student at Stanford University, where I am advised by Doug L. James. My research interest is in developing efficient algorithms for physics-based animation and sound synthesis. In the past, I've worked on research problems in marine geophysics, biomechanics, and fuel cells. I interned at Disney Research with David Levin in 2015, and at Adobe Research with Timothy Langlois in 2017.

Outside of work, I am an enthusiastic badminton player, fifa addict (on PS4), and I enjoy outdoor activities such as camping and swimming. I play ukulele or guitar from time to time, and I once knew how to play the ancient Chinese instrument Sheng (簧笙). I am a madridista. My son, Lucas, was born in March, 2018.

There is no correct pronunciation in English for my name. "Ray" is a close approximation for "Jui", and is what I go by (although I still prefer to write it as Jui).


KleinPAT: Optimal Mode Conflation For Time-Domain Precomputation Of Acoustic Transfer
Jui-Hsien Wang, and Doug L. James
2019, ACM Transactions on Graphics (SIGGRAPH 2019)
PDF Video Demo Project Webpage
We propose a new modal sound synthesis method that rapidly estimates all acoustic transfer fields of a linear modal vibration model, and greatly reduces preprocessing costs. Instead of performing a separate frequencydomain Helmholtz radiation analysis for each mode, our method partitions vibration modes into chords using optimal mode conflation, then performs a single time-domain wave simulation for each chord. We then perform transfer deconflation on each chord’s time-domain radiation field using a specialized QR solver, and thereby extract the frequency-domain transfer functions of each mode. The precomputed transfer functions are represented for fast far-field evaluation, e.g., using multipole expansions. In this paper, we propose to use a single scalar-valued Far-field Acoustic Transfer (FFAT) cube map. We describe a GPU-accelerated vector wavesolver that achieves high-throughput acoustic transfer computation at accuracy sufficient for sound synthesis. Our implementation, KleinPAT, can achieve hundred- to thousand-fold speedups compared to existing Helmholtz-based transfer solvers, thereby enabling large-scale generation of modal sound models for audio-visual applications.

Toward Wave-based Sound Synthesis for Computer Animation
Jui-Hsien Wang, Ante Qu, Timothy R. Langlois, Doug L. James
2018, ACM Transactions on Graphics (SIGGRAPH 2018)
PDF Video Project Webpage
We explore an integrated approach to sound generation that supports a wide variety of physics-based simulation models and computer-animated phenomena. Targeting high-quality offline sound synthesis, we seek to resolve animation-driven sound radiation with near-field scattering and diffraction effects. The core of our approach is a sharp-interface finite-difference time-domain (FDTD) wavesolver, with a series of supporting algorithms to handle rapidly deforming and vibrating embedded interfaces arising in physics-based animation sound. Once the solver rasterizes these interfaces, it must evaluate acceleration boundary conditions (BCs) that involve model and phenomena-specific computations. We introduce acoustic shaders as a mechanism to abstract away these complexities, and describe a variety of implementations for computer animation: near-rigid objects with ringing and acceleration noise, deformable (finite element) models such as thin shells, bubble-based water, and virtual characters. Since time-domain wave synthesis is expensive, we only simulate pressure waves in a small region about each sound source, then estimate a far-field pressure signal. To further improve scalability beyond multi-threading, we propose a fully time-parallel sound synthesis method that is demonstrated on commodity cloud computing resources. In addition to presenting results for multiple animation phenomena (water, rigid, shells, kinematic deformers, etc.) we also propose 3D automatic dialogue replacement (3DADR) for virtual characters so that pre-recorded dialogue can include character movement, and near-field shadowing and scattering sound effects.

Bounce Maps: An Improved Restitution Model for Real-Time Rigid-Body Impact
Jui-Hsien Wang, Rajsekhar Setaluri, Dinesh K. Pai, and Doug L. James
2017, ACM Transactions on Graphics (SIGGRAPH 2017)
PDF Video Project Webpage Bounce Maps Viewer
We present a novel method to enrich standard rigid-body impact models with a spatially varying coefficient of restitution map, or Bounce Map. Even state-of-the art methods in computer graphics assume that for a single rigid body, post- and pre-impact dynamics are related with a single global, constant, namely the coefficient of restitution. We first demonstrate that this assumption is highly inaccurate, even for simple objects. We then present a technique to efficiently and automatically generate a function which maps locations on the object’s surface along with impact normals, to a scalar coefficient of restitution value. Furthermore, we propose a method for twobody restitution analysis, and, based on numerical experiments, estimate a practical model for combining one-body Bounce Map values to approximate the two-body coefficient of restitution. We show that our method not only improves accuracy, but also enables visually richer rigid-body simulations.

Mechano-Chemical Model of Cancer Cell Invasion
Jui-Hsien Wang
2013, Master Thesis
The phenomenon regarding how cancer cell moves against and invades nearby tissue is an interesting yet difficult subject to study due to its multiphysics nature. In this work, we proposed a simplified yet elegant theoretical framework in an attempt to model the system. Physical assumptions were made to amount the modeling of cell locomotion to solving a classical elastostatic problem; during the process, cancer cells can secrete enzymes such as matrix metalloproteases to degrade the material. Therefore, in general the elastostatic problem happens on nonlinear, inhomogeneous substrate, which needs to be mathematically modeled. On the other hand, we hypothesized that multiple species of proteases can be secreted by the cell and they follow diffusion processes and can therefore react with the substrate at different rates. This is a two-way coupled system with cell invasion depth (or substrate deformation) depends on the results of both mechanical and chemical processes, whose outcome can be predicted in a very efficient way under the current framework. In particular, we first applied the regular perturbation method to give an analytical solution for the nonlinear elastostatic equation given a force profile assuming weak nonlinearity. Second, we designed and implemented a numerical, finite-difference based solver to model the diffusion-reaction system that can spatially distribute different species of proteases during the chemical process. This solver is then coupled to the elastostatic problem to close the loop. Several predictions based on this framework were given, such as the parametrized studies for invasion efficiency based on mechanical force distribution and colocalized enzyme distribution.

Effects of sea states on seafloor compliance studies
Jui-Hsien Wang, Wu-Cheng Chi, R. Nigel Edwards, and Eleanor C. Willoughby
2010, Marine Geophysical Research
Gas hydrates affect the bulk physical properties of marine sediments, in particular, elastic parameters. Shear modulus is an important parameter for estimating the distribution of hydrates in the marine sediments. However, S-wave information is difficult to recover without proper datasets. Seafloor compliance, the transfer function between pressure induced by surface gravity waves and the associated seafloor deformation, is one of few techniques to study shear modulus in the marine sediments. The coherence between recorded time series of displacement and pressure provides a measure of the quality of the calculated transfer function, the seafloor compliance. Thus, it is important to understand how to collect high coherence datasets. Here we conducted a 10-month pilot experiment using broadband seismic sensors and differential pressure gauges. We found that data collected in shallow water depth and during rough seas gave high coherence. This study is the first time long-term datasets have been employed to investigate seafloor compliance data quality and its dependence on sea state. These results will help designing future large-scale compliance experiments to study anomalously high shear moduli associated with the presence of gas hydrate or cold vents, or alternatively anomalously low shear moduli, associated with partial melt and magma chamber.

USYMLQ/USYMQR: Two Conjugate-Gradient-Type Methods for Unsymmetric Linear Equations
Jui-Hsien Wang
Final project for CME338: Large-Scale Numerical Optimization
PDF Code
USYMLQ and USYMQR algorithms generalize SYMMLQ and MINRES for solving large, sparse, unsymmetric linear system of equations. It is based on an efficient orthogonal tridiagonalization procedure described in Saunders et al. 1988 paper "Two Conjugate-Gradient-Type Methods For Unsymmetric Linear Equations". USYMLQ can be used for square or under-determined systems, whereas USYMQR can be used for square, under-determined, and over-determined systems. They are iterative solvers and each step has O(N) operations; the overall storage is O(N) as well (unlike GMRES). In the symmetric case, USYMLQ falls back to SYMMLQ and USYMQR falls back to MINRES. The implemented USYMLQ and USYMQR are then tested on real-world data scraped from the UFL sparse matrix collection dataset, and is found to converged relatively quickly for a majority of the problems.
Modal Sound Classifier
Jui-Hsien Wang
Final Project for CS221: General AI
In this project, I trained an AI tool to predict the object and material class from a single input contact sound. With the carefully chosen feature space, I was able to achieve more than 99% test accuracy on a 18-way classification problem.
Real-time Aerodynamic Sound Synthesis for Slender Objects
Jui-Hsien Wang
Final project for CS5643: Physically Based Animation for Computer Graphics
PDF Demo
In this project, I explored the problem of real-time, physics-based aerodynamic sound synthesis for slender objects. It is largely inspired by the paper by Dobashi et al. [Dobashi et al. 2003]. The aerodynamic sound when we swing a slender object, such as a stick, is originated from the complex interaction between the air flow and the stick. Sufficient spatial/temporal resolution was regarded to be essential to capture the physics and thus the characteristics of the sound generated. However, the required fluid simulation is too expensive to run at audio stepping rate. To avoid such computation, I first precomputed a comprehensive database that contains relevant sound textures evaluated from high-quality grid-based fluid simulation, and then at runtime, this database is fetched and textures are blended to effectively resynthesize the aerodynamic swinging sound. Next, to increase the interactivity of the project, I interfaced the sound system with Leap Motion sensor to give real-time motion capture data. The system is proven to be quite reliable and can run at real-time even on a low-end laptop, and create realistic swinging sound.
Bounce Maps Viewer
Jui-Hsien Wang
Supplemental Material for the Bounce Maps paper
Web Viewer
In our SIGGRAPH 2017 paper, we showed that the coefficients of restitution on rigid-bodies have an extremely comlicated distribution, as opposed to what is believed/used today in even the most advanced rigid-body solver -- a single constant number someone typed in: "COF = 0.3". This is a simple web-based viewer that visualizes the beautiful restitution maps on some of the familiar 3D models.