Anka He Chen
@AnkaHeChen
Followers
1,554
Following
198
Media
25
Statuses
391
PhD from the University of Utah | Research Scientist @ NVIDIA| Physics Solver | Motion Capture | warp | GAIA | VBD
Joined December 2021
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Sinwar
• 1173497 Tweets
Hamas
• 732935 Tweets
#يحيي_السنوار
• 433966 Tweets
Bear
• 262831 Tweets
Bluesky
• 259827 Tweets
Monsalve
• 105631 Tweets
Tyler
• 102857 Tweets
#TUDUMNaLata
• 100507 Tweets
غير مدبر
• 97042 Tweets
Rafah
• 86323 Tweets
UNRWA
• 84884 Tweets
Niall and Harry
• 83925 Tweets
O Louis
• 51628 Tweets
Louis Tomlinson
• 49844 Tweets
Louis y Zayn
• 49200 Tweets
ابو ابراهيم
• 48881 Tweets
Mark Cuban
• 42956 Tweets
#GHGala7
• 40287 Tweets
Corinthians
• 38258 Tweets
Niall y Harry
• 35907 Tweets
Payno
• 33456 Tweets
Isabel
• 32447 Tweets
Silvia
• 31406 Tweets
Macaya
• 28681 Tweets
対象作品
• 22235 Tweets
Tapia
• 22081 Tweets
Sofía Delgado
• 20414 Tweets
Vural Çelik
• 19872 Tweets
Broncos
• 18574 Tweets
BAUTI X ALGO DE MÚSICA
• 16379 Tweets
Kawhi
• 14705 Tweets
ENIGMA DO MEDO
• 13200 Tweets
Pinned Tweet
#SIGGRAPH2024
We present Vertex Block Descent (VBD), the NEXT simulator.:
nearly zero-compromised, fast🚀🚀, massively parallel💻💻, convergent📉📉, and unconditionally stable💎💎.
Project Page:
Code (coming soon):
18
166
1K
🎉🎉I finally released my physics simulator, Gaia:
The Gaia engine is a C++ codebase primarily designed for physics-based simulations. It can be compiled as a standalone simulator or integrated into other applications as a third-party module.
13
121
809
I just successfully defended my PhD!!! I am a Dr. now! Thanks to all my nice committee members,
@cem_yuksel
@giljoo_nam
@Mike
Kirby
@SheldonAndrews
@YinYang24414350
. Special thanks to my advisor Prof. Cem Yuksel, I couldnt do it without you. Next stop:
@nvidia
30
8
374
I will give a talk on my Vertex Block Descent paper at SIGGRAPH on Wednesday, July 31st, starting at 3:55 PM. The session will be held in Miles High 4. Remember to come🤩!
2
22
336
🥳🥳🥳I have open-sourced the code of our Siggraph 2023 paper:
shortest path to boundary for self-intersecting meshes
at:
The code includes a test example that allows you to quickly get your hands on the algorithm. Please see the repo for more info.
7
63
314
#SIGGRAPH2024
🥳🥳I've open-sourced the code for our Vertex Block Descent (VBD) paper as a part of Gaia:
This is a collaborative work with
@lzh2199
. Thanks for your interest in VBD and Gaia! I released it ahead of schedule for you!
4
55
311
Check out the trailer for Gaia's next major update [], where we introduce a new solver that enables simulations like this to run in just a few seconds per frame.
The details will be released soon.
8
20
264
I will give the presentation of my paper "Shortest Path to Boundary for Self-intersecting Meshes" at Siggraph this afternoon (Aug 10), at 2:00PM (PDT)
Location: Petree Hall C at the Los Angeles Convention Center.
I make the presentation very listener-friendly!
3
29
249
I am glad to see that someone has implemented VBD using compute shader. This extends VBD to GPUs without CUDA support. This motivates me to write a compute shader based VBD and open-source it.
@AnkaChen1
@YouTube
I've implemented a very basic VBD setup in TouchDesigner using compute shaders. Looking forward to further exploring the technique. Thank you so much for sharing your research!
3
0
36
2
6
128
10 mins presentation for a Siggraph paper is ridiculous. I use to have a good presentation that every one can learn a lot from. Now it's crippled. 🥲
3
4
82
What an amazing cloth simulation work using VBD and Metal! It really looks close to my implementation now!
Here we go, I finally implemented VBD with the StVK hyperelastic model (λ = 1e5, μ = 5e4).
Currently, I do:
- 8 (max) force elements per vertex in parallel
- 4 vertices in an SIMD group (8*4)
- 0 loops for reduction
- A lot of loops for Hessian 😅
3
1
90
1
8
74
This is the full video of our
#SIGGRAPH2024
paper, Vertex Block Descent:
Project page:
----------------------------------
via
@YouTube
1
8
71
Very proud to announce our new paper:
[Siggraph 2023] Shortest Path to Boundary for Self-Intersecting Meshes via
@YouTube
where we proposed an efficient way to build collision constraints for self-intersecting meshes, enabling fast simulation.
2
6
60
Great work from the community using VBD!
0
1
43
This looks amazing! And super fast!
Optimizing for robustness over accuracy can get you pretty good results!
Now I am doing soft vertex-vertex collision after initialization and hard vertex-triangle collisions during the solve step.
Here is a stress test with pretty high velocity and elasticity —
5
6
75
1
0
34
Disappeared from twitter for a while. I just made an submission for Siggraph. It's quite a crunch for me, but we eventually submitted a well polished paper. That is something truly great I am really proud of, looking forward to it.
2
0
33
Big thanks to Károly for making such an amazing video to introduce our work. Keep bringing us those breakthroughs in science & tech, and, what a time to be alive!
New video - 50,000,000 Point Bouncy Jelly Simulation!
0
5
34
2
0
29
Almost done with cleaning the Gaia Repo for the release of VBD's code.
0
0
23
So happy that my paper got selected into the technical paper trailer. Not bad for an algorithm that only takes a week to come up with. But of course, implementing it is another story. 😁
1
1
21
@boiledwater
我不明白，为什么六年级就写作业写到1点。我高中也就10点后就不学习了。我六年级没上跳了一级，整个六年级的课程我不到一周就自学完了。不明白是增加了什么健身内容要让孩子天天学到1点？
17
0
20
We have a picture of Bùi Tường Phong on our wall of alumini, and he looks like none of the picture you can find online
1
0
21
VBD is parallelized by coloring the mesh to achieve Gauss-Seidel iterations, similar to (X)PBD. However, VBD colors the vertices directly, whereas (X)PBD colors the constraints, resulting in 10 times better parallelism than (X)PBD.
2
1
16
Unlike alternative techniques requiring a global linear solve like Newton's method or projective dynamics, VBD utilizes local, vertex-level Gauss-Seidel iterations, making it faster, more parallelizable, and less prone to numerical issues.
1
0
7
The code will be released as the next update of my Gaia library. But we do have a tiny demo ready:
Code (tiny demo):
Full video:
Paper:
You can join my discord for discussion:
2
0
12
Note that this is the code for the shortest path to surface algorithm. It does not include the simulation part, which will be released separately, as Gaia physics engine:
0
0
11
This remind me of my undergrand time. I was in a library studying math. After a long time I stand up for dinner because it's dark. Then I tried to remember what I ate for lunch. Then I realized I forgot to have lunch.
extremely relatable, this happens to me like every other night (though I kind of like it)
23
30
593
0
0
12
With assistance from enthusiastic community members, I've resolved several bugs in Gaia that were causing compilation issues. If you previously encountered difficulties with compiling, please try again now. Special thanks to
@colemapt
.
0
0
12
I think this raises questions about the essence of creativity. The development of AI suggests that art and writing (at least some forms of them) might not be as inherently creative as we once believed.
3
0
11
😅So true. Don't get me started, otherwise I am gonna start by comparing Riemannian integration and Lebesgue integration and the measurement of a set.
1
0
11
Fellow researchers, please give my friend some advice and encourage
I am considering pursuing a part-time PhD in fields such as discrete differential geometry or differentiable physics. Is it worth it?
Are there any benefits beyond career and collaboration opportunities in research?
6
0
25
1
0
12
@MathMatize
It couldn't be homeomorphic to a sphere. The legs and heads and arms must come out through holes, which make it homeomorphic to a at least genus 5 surface
0
0
9
@thejcedeno
@cem_yuksel
Just my own 3080Ti😇. Our method only adds a tiny overhead to the simulation (around 5% of the total run time). Also, this method is embarrassingly parallel.
0
1
8
@sellan_s
Implicit functions are those you dont have explicit map from one to the other. It require a solving of a equation to calculate the independent variable, such as sin(x+log(y))=exp(xy)sinh(y)
0
0
10
@cem_yuksel
Thank you Cem! We are planning on releasing the code soon, please follow to receive notifications!
1
0
8
Gaia provides a set of useful tools, including a powerful triangular/tet mesh data structure, a convenient parameter IO module, a set of efficient collision detectors, and a virtual physics framework that can be extended to support all sorts of solvers.
1
0
8
Gaia is engineered to enhance the efficiency of both developers and the hardware. Compromises may be made since optimizing for one can adversely affect the other. It may not always have the smartest design. However, I can assure you that Gaia is free from any stupid designs.
0
0
7
@Luckyballa
Strongly suggest you do a PhD, but not a part-time one. I don't believe a part-time PhD is possible in the area you want to pursue, especially if you want to excel. The top conference in that area is SIGGRAPH, and the work for a SIGGRAPH submission requires fulltime dedication
1
0
8
This method is ideal for those quick but not penetration-free frameworks, like XPBD. Our method is even stable for severely self-intersecting shapes and capable of covering them to a penetration-free state.
0
1
7
@thejcedeno
@cem_yuksel
It can be real-time. With our GPU implementation, it can simulate models with a total of 100K vertices in real time (30 FPS).
0
1
6
Great opportunity to learn about Warp!
We're running a course on NVIDIA Warp at SIGGRAPH bright and early on Sunday morning 9am-12:15pm, including presentations from
@peterchencyc
,
@pika7ma
,
@zeshunz
,
@shinjin_cr
, and
@c0d1f1ed
- hope to see you there!
2
13
65
0
0
7
@CFD_UniBw_M
This is the kind work I want to see, instead of: hey we let neuron networks learn to estimate 90% of the physics and it's only 100x slower
0
0
7
@DrJimFan
At least in the computation & physics community, "simulation" means you know exactly how things runs, which are usually described by a few physics laws, and you try to imate the process of natue using computation. The computational method must be proved, to provide some level of
1
0
4
We believe we have discovered a nearly zero-compromised simulator. VBD has all the desired properties: stability, parallelism, speed, convergence, and support for arbitrary energy formulations. To our knowledge, no previous has all those properties. VBD delivers it all.
1
1
6
@miniapeur
Genghis is not his real name, his name is Temüjin. Genghis Khan is a respectful title he gets after he became Khan and united the Mongolian clans
0
0
7
On the other hand, doing chores demands the delicate control of thousands of muscles throughout the body, while also coordinating multimodal inputs such as vision, tactile sense, hearing, balance, etc. This level of complexity remains highly challenging for robots to achieve.
1
1
6
@Luckyballa
Have you considered using the 2 level parallelism I introduced in the paper? You can use a work group (I am not sure how it's called in Metal) to evaluate and aggregated the hessian in parallel! In that case it doesn't need to loops over the adjacent hessians and is much faster!
0
0
6
@XorDev
Basically all the answers are about rendering... Let me provide my perspective from the simulation side. Simulate approximately the same number of primitives in real time as we can render them. With unconditional stability and support for collisions & fractions
0
0
4
@ogaelinasai
@suohawking
Idk, 我本科是电气工程专业的，没有学过数学专业课。我大四喜欢上了数学，就花了半年时间把大学本科课程学完了。然后我参加了数学系的研究生考试，我成绩排名第一，其他人都是科班出身的数学系……
1
0
5
VBD is not only fast but also achieves highly accurate simulation results through its ability to achieve numerical convergence and outperform alternative techniques. Additionally, it fits within any computation budget by adjusting the iteration count.
1
0
4
😂😂
1
1
5
Currently, the XPBD-based simulator, along with some running parameters that replicate the experiments included in my paper "Shortest Path to Boundary for Self-Intersecting Meshes" are provided.
1
0
4
A paper from 2000 with no updates, and still rules (some) lifes is ridiculous.
1
0
5
@kiaran_ritchie
Actually I just discovered a bug in the implementation of acceleration. Let me fix it.
1
0
4
Join the HPG student competition and win a RTX 4080!
0
0
3
@Luckyballa
why not also do the solf collision during the solve stage as and apply the hard constraint at the end. My observation is using hard constraint alone could inject energy
1
0
3
@ThePhDPlace
It also depends on what kind of advisor you get. I am lucky to eventually having a great advisor. But there are people who likes to exploit their students, who like to lie, who refuse to fund their students, who like to take credit for their students' woek, who like to torture
1
0
4
This work is completed in collaboration with my amazing colleagues: Ziheng Liu
@lzh2199
Yin Yang
@YinYang24414350
and Cem Yuksel
@cem_yuksel
1
0
4