With this project, we provide the software for LRNNs (linear recurrent
neural networks). The software has been implemented by the authors in
Python and the scientific programming language Octave (see octave.org,
version 5.2.0). It has been tested under Linux. The main routines can
be found in lrnn implemented in Octave. For further details of LRNNs
and their properties, the interested reader is referred to the
following paper.
The self-contained Python implementation includes the autonomous LRNN library, spectral dimension reduction, tests, a runnable example and reproducible MSO/Python–Octave comparisons. Its accuracy guide explains the fitting settings and measured forecast accuracy.
The research website gives a visual overview of the paper, its learning and reduction method, published experiments and Python reproduction. See website setup for local preview and GitHub Pages hosting.
Frieder Stolzenburg, Sandra Litz, Olivia Michael and Oliver Obst.
Efficient time-series approximation with linear recurrent neural networks:
architecture learning and predictive power.
Neural Computing and Applications 37, 27027–27055 (2025).
DOI: 10.1007/s00521-025-11655-y.
The research reported here has been supported by the German Academic Exchange Service (DAAD) by funds of the German Federal Ministry of Education and Research (BMBF) in the Programmes for Project-Related Personal Exchange (PPP) under grant no. 57319564 and Universities Australia (UA) in the Australia-Germany Joint Research Cooperation Scheme within the project Deep Conceptors for Temporal Data Mining (Decorating).
The Decorating project is released under the BSD 3-Clause licence.