2 problems

0 votes0 replies1 view
The unimodal minimal filling architecture conjecture for polynomial neural networks
Let $r$ be the activation power, let $L$ be the depth, and fix input and output widths $d_0$ and $d_L$ . For an architecture $\mathbf{d}=(d_0,d_1,\dots,d_L)$ , call it minimal fillin…
Conjectureopenneuroalgebraic-geometry polynomial-neural-networks tensor-decompositionsarXiv posted 2 days ago
0 votes0 replies0 views
The generic-fiber symmetry conjecture for algebraic multilayer perceptrons
Let an algebraic multilayer perceptron (MLP) be a neural network whose parameterization is algebraic, and consider the fibers of its parameterization map. Its parameter symmetries…
Conjectureopenalgebraic-geometry neural-networks neuroalgebraic-geometryarXiv posted 2 days ago

The unimodal minimal filling architecture conjecture for polynomial neural networks