New release of DeckWerk, the open-source, multi-platform, agent-enabled slide editor! Lots of improvements: Better UI, simplified agent support, better headless mode (=local Google Slides), support for code blocks, more shapes, more animations, better keynote import, ...
Download & install binary here:
https://t.co/UO4CZYCqQI
Is it possible to make 3D Gaussian Splatting even faster?
Short answer YES!
Enjoy a 10.5x optimization on forward/backward cuda kernels over the published 3DGS ones!
https://t.co/fVyIexfh6N
@trueelitesheep Beignet is a per-instance function approximator supervised on the PDE residual; it does not do operator learning or use a training set like FNO
Introducing beignet, a new neural field architecture!
beignet uses a multiresolution fourier frequency pyramid to achieve faster convergence and higher precision when solving PDEs and finding singularities.
Teaser: here is beignet at work fitting an image (super fast)!
@cwkx The main difference is that we use fourier (sinc) instead of linear interpolation for space. Spatial derivatives in instant NGP are undefined on grid points which can be bad for pdes
Check out this work led by my student Brandon Zhao (@joiedeposter), with @pratul, @dorverbin, @jon_barron & Y. Wang! We use trainable pyramids to solve PDEs more accurately. I even got to learn a little about PDEs along the way - handy now that they seem to be all over the news!
Recent results on unstable fluid singularities point toward harder open questions, including unforced Navier–Stokes (e.g. https://t.co/zY9Ki2voDi). Burgers' is a much simpler PDE, but we think beignet might be an important building block towards something bigger!