VoodooNet: Achieving Analytic Ground States via High-Dimensional Random Projections

TL;DR

VoodooNet is a neural architecture that uses high-dimensional projections and an analytic solution to achieve high accuracy on image classification tasks without iterative training.

Contribution

It introduces a non-iterative, closed-form neural network using Galactic Expansion and Moore-Penrose pseudoinverse for rapid, high-dimensional feature untangling.

Findings

Achieves 98.10% accuracy on MNIST

Achieves 86.63% accuracy on Fashion-MNIST

Surpasses SGD baseline on Fashion-MNIST with reduced training time

Abstract

We present VoodooNet, a non-iterative neural architecture that replaces the stochastic gradient descent (SGD) paradigm with a closed-form analytic solution via Galactic Expansion. By projecting input manifolds into a high-dimensional, high-entropy "Galactic" space ($d \gg 784$), we demonstrate that complex features can be untangled without the thermodynamic cost of backpropagation. Utilizing the Moore-Penrose pseudoinverse to solve for the output layer in a single step, VoodooNet achieves a classification accuracy of \textbf{98.10\% on MNIST} and \textbf{86.63\% on Fashion-MNIST}. Notably, our results on Fashion-MNIST surpass a 10-epoch SGD baseline (84.41\%) while reducing the training time by orders of magnitude. We observe a near-logarithmic scaling law between dimensionality and accuracy, suggesting that performance is a function of "Galactic" volume rather than iterative refinement. This "Magic Hat" approach offers a new frontier for real-time Edge AI, where the traditional training phase is bypassed in favor of instantaneous manifold discovery.