Fitting a 4D Mandelbulb with KS3 vs MLP [R]
Summary
A video comparing KS3, the latest hermite spline regression block in the K-Splanifolds series, against MLPs on fitting a 4D Mandelbulb over 9,000 gradient descent steps, with KS3 achieving the lowest loss at nearly identical forward/backward latency.
View Cached Full Text
Cached at: 10/28/26, 02:42 AM
Similar Articles
you can now fine tune Prism-ML's ternary Bonsai models
Prism-ML announces that their ternary Bonsai models can now be fine-tuned, with example code and a recommendation to use a high learning rate.
$R^2$-dLLM: Accelerating Diffusion Large Language Models via Spatio-Temporal Redundancy Reduction
R²-dLLM introduces spatio-temporal redundancy reduction techniques that cut diffusion LLM decoding steps by up to 75% while preserving generation quality, addressing a key deployment bottleneck.
I Compressed Bad Apple into a 3MB Neural Network [P]
The author compressed the Bad Apple animation into a small MLP with sine activations (SIREN), using only 790k parameters (~3MB) to represent 2.7 billion pixels, with improvements like time-stretching and motion-focused sampling reducing validation MSE by ~9x.
Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering
Imagine3D-LLM proposes teaching multimodal LLMs to imagine a compact 3D representation of a scene (decoded from learnable summary tokens as 3D Gaussian Splatting via photometric reconstruction loss) before answering multi-view spatial reasoning questions, outperforming prior 3D-aware MLLM approaches on multiple benchmarks.
Kimi K3 Architecture Overview and Notes
Sebastian Raschka provides an architectural overview of the open-weight Kimi K3 model, highlighting its scaling from 48B to 2.8T parameters, new LatentMoE and attention residual components, removal of RoPE in favor of NoPE, and native multimodal support. The model emphasizes inference efficiency and matches frontier performance.