Source and reproducibility repository for:
ROSA: Metric Amplification on Noisy Graphs with Theoretical Guarantees for Amplified Spectral Distances Ben Cardoen and Fabian Spill Submitted to SIAM Journal on Mathematics of Data Science (SIMODS), 2026
paper-siam.qmd Main manuscript (Quarto → LaTeX → PDF)
references.bib Bibliography
figures/ Signed-off figure PDFs used by the manuscript
reproduction/ Julia scripts, inputs, and paper-facing results
Project.toml Direct dependencies pinned to exact commits
Manifest.toml Complete reproducibility environment
- Quarto (≥ 1.4)
- TeX Live 2025 (or equivalent LaTeX distribution with
pdflatex,biber) - SIAM document class (
siamart251216.cls, included in the repository)
./build_paper.shOutput: output/paper-siam.pdf
The checked-in Julia environment pins the paper's project dependencies to exact commits in the following repositories:
| Package | Repository |
|---|---|
| ROSA.jl | Amplification framework |
| DynamicGeometricGraphs.jl | Geometric graph types |
| CalibratedTemperedPowerLaw.jl | Noise models |
| GraphSpectra.jl | Graph-level spectral workflows |
| Spectral.jl | Sparse partial-spectrum solver |
| SkeletonExtractor.jl | Image-to-graph extraction |
cd reproduction
julia --project=. -e 'using Pkg; Pkg.instantiate()'
julia --project=. -e 'using Pkg; Pkg.test()'
julia --project=. figure_2/generate.jlSee reproduction/README.md for the mapping from manuscript figures and tables to their reproduction entry points.
Continuous integration resolves the complete pinned public environment without
precompiling it, then runs fast static integrity checks over the reproduction
source and committed results. These checks cover script syntax, path hygiene,
figure provenance, tabular schemas, and paper-facing data invariants; they do
not rerun the numerical experiments. The Pkg.test() command above runs the
full Julia helper tests locally.
Running all experiments is computationally expensive. The full run was tested on an Apple M4 Max with 64 GB RAM and takes approximately 2–4 hours; a machine with 16 GB RAM and four cores may require 6–12 hours.
The original software and scripts in this repository are available under the MIT License. Copyright in the manuscript and original figures remains with their authors, and those materials are not covered by the MIT software license. Third-party materials and data retain the terms specified by their respective sources.