This project focuses on the analysis of invariant manifolds within the context of piecewise linear RNN like the ALRNN and shPLRNN models. This is the codebase for the paper Eisenmann et al. (2025) Detecting Invariant Manifolds in ReLU-Based RNNs.
The codebase provides a comprehensive set of tools for exploration and understanding of manifold structures in high-dimensional spaces.
Figures and results presented in the paper can be reproduced by running the scripts located in the evaluation folder. These scripts are designed to demonstrate the capabilities of the provided functions and to validate the theoretical concepts discussed.
A general pipeline for executing the entire workflow will be added soon, which will streamline the process of setting up, running, and analyzing manifold-related experiments.