Fitting a 4D Mandelbulb with KS3 vs MLP [R]

For the past year I have been studying using a fast K-dim hermite spline manifold algorithm, called K-Splanifolds or KS, for general purpose fast regression. In this video I show the third version, KS3, compared to MLP and older KS versions fitting a 4D mandelbulb via gradient descent on random samples. KS3 converges faster and has nearly identical latency on a GPU as MLP. KS blocks are suitable for a wide range of regression problems and I have found KS3 ranges from considerably faster convergence as MLP (noise removal, geometric fit, financial prediction) to neck and neck (LLMs). You can view code for all three versions of the algorithm and a paper for KS1 here: github.com/curvedinf/k-splanifolds I am an independent and if anyone would like to explore this with me, let me know. Also if anyone would like to donate access to 8x H100, I would like to attempt a world record nanogpt speedrun with KS3 using configurations I have found beat the WR on proxy hardware.