-
Fast Symbolic Integer-Linear Spectra
Abstract: Here we contribute a fast symbolic eigenvalue solver for matrices whose eigenvalues are $\mathbb{Z}$-linear combinations of their entries, alongside efficient general and stochastic $M^{X}$ generators. Users can interact with a few degrees of freedom to create linear operators, making high-dimensional symbolic analysis feasible for when numerical analyses are insufficient.
Submitted 12 December, 2024; v1 submitted 18 September, 2024; originally announced October 2024.
-
Ensemble Mask Networks
Abstract: Can an $\mathbb{R}^n\rightarrow \mathbb{R}^n$ feedforward network learn matrix-vector multiplication? This study introduces two mechanisms - flexible masking to take matrix inputs, and a unique network pruning to respect the mask's dependency structure. Networks can approximate fixed operations such as matrix-vector multiplication $φ(A,x) \rightarrow Ax$, motivating the mechanisms introduced with… ▽ More
Submitted 9 October, 2023; v1 submitted 12 September, 2023; originally announced September 2023.