Scientific publications
Stéphane Gael Raymond Ekodeck
Prosper Rosaire MAMA ASSANDJE
St´ephane Willy Mossebo Tcheunteu
Ewane Ulrich Arthur
Damaris Belle M. Fotso
Figshare
This supplementary dataset contains the reproducibility code and numerical outputs supporting the manuscript “Fisher--Rao Geometric Optimization and Detection for Generative Steganographic Transport.”<br>The package provides Python scripts for reproducing the numerical simulations reported in the paper. These simulations cover: (i) Fisher--Rao geodesics on Gaussian statistical manifolds, (ii) natural-gradient versus Euclidean-gradient convergence in diagonal and non-diagonal Gaussian families, (iii) comparisons with adaptive optimizers such as Adam and RMSprop, (iv) curvature estimation on Gaussian-mixture statistical manifolds, (v) geodesic versus Euclidean path comparisons, and (vi) local Fisher-ball volume estimation for representative Gaussian-mixture points.<br>The main script `run_all.py` executes the principal simulations S1--S4 in sequence. Individual scripts are also provided for each experiment: `sim_s1_geodesics.py`, `sim_s2_gradient.py`, `sim_s2_nondiagonal.py`, `sim_s2_adaptive_comparison.py`, `sim_s3_curvature.py`, `sim_s4_path_comparison.py`, and `compute_fisher_ball_volumes.py`.<br>The scripts generate CSV data files, PDF figures, and LaTeX tables used to support the manuscript’s numerical claims. In particular, the Fisher-ball volume script estimates local Fisher-ball volumes for a one-dimensional two-component Gaussian mixture model and outputs `fisher_ball_volumes.csv` and `fisher_ball_volumes.tex`.<br>The supplementary material is intended for academic reproducibility of the theoretical and numerical results. The simulations are performed on low-dimensional Gaussian and Gaussian-mixture statistical manifolds and should not be interpreted as an operational benchmark of a complete image-steganography system.
