Reverse Engineering a Neural Network's Clever Solution to Binary Addition

I recently finished up the process of fixing a bug in the Chromium/Google Chrome web browser. It was my first time contributing to the Chromium project or any other open source project of that scale, and it was a very unique experience compared to other open source work I've done in the past. I figured that I'd write up an overview of the whole...

For a couple of years now, I've been working on and off on some 3D scenes and levels that run in the browser. It started off as a scattering of self-contained demos for some custom shaders or similar that I wanted to try out, but over time the project has grown into a pretty substantial interconnected game-like thing. One of the unifying themes...

There's been a fair bit of buzz about Kolmogorov-Arnold networks online lately. Some research papers were posted around claiming that they offer better accuracy or faster training compared to traditional neural networks/MLPs for the same parameter count. I was compelled by these claims and decided to test them out myself. Here are my main...

Overview When I started off, my goal was simply to subdivide 3D meshes. I wanted to add more geometry to simple meshes so that I could apply programmatic deformation like greebling and noise to give them increased detail and realism in procedurally and semi-procedurally-generated scenes. It turns out that basic mesh subdivision is indeed very...

Background As part of some other work I was doing that involved programmatically manipulating 3D meshes, I encountered the need to compute normals for them from scratch. Usually this isn't necessary since applications like Blender handle doing this automatically. But in my case, I was doing these transformations dynamically and in the web...