Alec Jacobson

📢 libigl v2.5 released 🐍 pip install libigl Eigen → 3.4 CGAL → 5.6 OpenGL → 4.1 CGAL and its dependencies (GMP+MPFR) compile everywhere from source via cmake Github actions finish faster and rarely fail due to heap size Shadow mapping in viewer …

✌️ not going to be on this site much anymore. See you somewhere else.

@rms80 Hold em hostage. Only fix it if they hire you back for $10M an hour. Surely no one else can find let along fix this bug. You have all the leverage. Don't settle for less. #geometryguru #unreal #unrealstoreloadingscreenisryansfault

tfw when you discover you shipped a bug in a ray-plane intersection function 🪦☠️⚰️

🖖 #include <autodiff/reverse/var.hpp> template <typename T> T f(T & x) { // "clever" branch to avoid 0/0 at x=0 if(x == 0.0) return 1.0; return sin(x)/x + x*(x-1.0); }

You can do this so your long titles don't look as terrible: \newcommand{\abbrtitle}{Clever Short Title} \title[\abbrtitle]{\abbrtitle:\\ Long Part Causing Line Breaks in Title and Spills in Headers}

@mnewberg @rms80 I knew it.

Today is my last day at Epic Games, please let me know if there is anything you’d like me to get done before I go.

Trying out a Role-Playing Paper-Reading Seminar in the style of @colinraffel's blog colinraffel.com/blog/role-play… in my History of Language and Computing graduate course this semester. Eager to see how it plays out, but I wanted to show off the class materials that just arrived :)

@_jessethomason_ @colinraffel amazing. I love this.

@Arxiv_Daily nah

@_AlecJacobson is this actually hard?

Q: We're almost out of time. Do you have any questions for me? A: Yes, how do you represent a mesh with a slit along a single edge?

you may also like libigl 🤝 TinyAD github.com/alecjacobson/l…

Typical. More time wrangling cmake than writing C++. Here's an example project using both libigl and autodiff.github.io (a nice C++ library for forward and reverse automatic differentiation). github.com/alecjacobson/l…

@tzumaoli Heh. I was just reading this this morning.

@_AlecJacobson Yes, this is a long-standing problem in autodiff literature. For example, Chapter 14 of Griewank's book talked about it. The most recent discussion I saw is in Hückelheim et al.'s very nice paper "Understanding Automatic Differentiation Pitfalls" arxiv.org/abs/2305.07546

Mathematically is there any reason to think that: f(x) = x is different from g(x) = / 0 if x=0 \ x if x≠ 0 ?

This kind of example pops up as a pitfall in autodiff discussions. And in a recent discussion regarding symbolic differentiation, somebody appeared to argue that f and g are different in some essentially way which excuses the symdiff library from finding equivalent derivatives.

I would think of the piecewise definition as an _implementation_ of the same function as f. Thus they're the same and consequently have the same derivative ∂f/∂x = ∂g/∂x, even at x=0.

I no longer have to send Wayback Machine links to my own blog articles! Successfully transferred my wordpress blog from bluehost to my static academic page. Old post urls should redirect correctly. alecjacobson.com/weblog/

@benhouston3d I didn't try out any higher-order ones yet, but I'm bothered by the fact that the simplest case already has non-trivial error 0---1 | | 1----? ? should be √2 but FMM predicts 1+√½ Fitting a quadratic to this case would get it right. Not sure how that generalizes

@_AlecJacobson Then you'd be left with errors at 45/2 degrees. I wonder if a higher order Eikonal equation is possible?

Playing with the fast marching method for the first time.

been doing this for a while and I'm still not sure I know whether "strongly" or "enthusiastically" is better.