Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SpectralUnmix: A Torch-Based Regularized Non-negative Matrix Factorization

Rafael S. de Souza, Paula Coelho, Niranjana P, Ana L. Chies-Santos, Rogério Riffel

Abstract

We present SpectralUnmix, an R package for regularized non-negative matrix factorization (NMF), implemented in torch with optional GPU acceleration. The package estimates low-rank non-negative representations through proximal-gradient updates and allows smoothness regularization along the spectral axis. As a compact demonstration, we apply the method to a subset of stellar spectra and compare the recovered NMF components with principal-component directions and representative stellar spectra. The package is released under the MIT license at \href{https://rafaelsdesouza.github.io/SpectralUnmix/}{this repository}.

SpectralUnmix: A Torch-Based Regularized Non-negative Matrix Factorization

Abstract

We present SpectralUnmix, an R package for regularized non-negative matrix factorization (NMF), implemented in torch with optional GPU acceleration. The package estimates low-rank non-negative representations through proximal-gradient updates and allows smoothness regularization along the spectral axis. As a compact demonstration, we apply the method to a subset of stellar spectra and compare the recovered NMF components with principal-component directions and representative stellar spectra. The package is released under the MIT license at \href{https://rafaelsdesouza.github.io/SpectralUnmix/}{this repository}.
Paper Structure (4 sections, 3 equations, 1 figure)

This paper contains 4 sections, 3 equations, 1 figure.

Table of Contents

  1. Introduction
  2. Method
  3. Analysis
  4. Conclusions

Figures (1)

  • Figure 1: Panel A shows four representative normalized spectra out a sample of 80. Panels B and C compare PCA and NMF eigenspectra derived from the same data; colors indicate the stellar class assigned to each component through a one-to-one matching based on a simple $\chi^2$-like distance between normalized spectral shapes. Panel D summarizes the class--component distances for PCA and NMF.