IDC — Iterative Domain Contraction for Neural-Network Response Optimization

This repository is the public companion to the paper "Neural network response optimization via iterative domain contraction" (Scala, Lopez, Assumpção, Dewil at Advances in Computational Science and Engineering / AIMS).

https://doi.org/10.3934/acse.2026010

It contains the worked examples reported in §7 of the paper, the benchmark suite, the held-out validation protocol, and a reproduction recipe — all built on top of the open-source library OpenNN.

Status. The C++ worked examples (§7.3–§7.5), the BBOB / Olympus validation runners, the photo_pce10 21-seed sweep + aggregation, the surrogate-quality audit, the MO figure-regeneration, the pymoo/pycma baselines for the three §7 example problems (benchmarks/baselines/), and the in-tree datasets are present and runnable. OpenNN is pinned to the immutable tag v1.2-IDC-paper for byte-reproducibility. Only the broader ~30-problem benchmark catalog the paper's §7.1 points to and its baseline sweep are run from the authors' workspace.

---

About OpenNN

OpenNN (Artelnics/opennn) is an open-source neural-networks C++ library released under LGPL v3. IDC is implemented as the ResponseOptimization class inside OpenNN; this companion repository ships worked examples and reproduction tooling around that core, version-pinned to a specific OpenNN release so the paper's numbers are byte-reproducible.

If you want to use IDC inside your own project, depend on OpenNN directly. This repo is the paper-reproduction layer.

---

Quick start

git clone https://github.com/Siscanalysis/IDC.git
cd IDC
mkdir build && cd build
cmake ..        # downloads + builds the pinned OpenNN release
cmake --build . --config Release
./bin/photo_pce10   # runs the §7.4 real-application SO case study

Expected output: a result.csv containing the IDC-recommended input configuration and corresponding surrogate output, matching the headline number in the §7.4 table of the paper.

---

Reproducing the paper

./scripts/reproduce_paper.sh   # build + the three C++ case studies + BBOB + Olympus

This configures and builds (fetching the pinned OpenNN), runs the three C++ case studies (§7.3 MOEED13, §7.4 photo_pce10, §7.5 concrete_uci_mo), then runs the §7.2 BBOB validation, the §6.3 f15–f24 stress test, and the §7.4 Olympus real-data SO sweep. The three C++ examples finish in seconds; the BBOB / Olympus runners take longer, dominated by the pymoo baselines (IDC itself is sub-second per seed). It then runs the photo_pce10 21-seed sweep + aggregation, renders the surrogate-quality audit, and regenerates the §7 MO figures from the committed result CSVs. The pymoo/pycma baselines for the three §7 example problems ship in benchmarks/baselines/ and run on the same surrogate + YAML as the C++ binaries. Only the broader ~30-problem benchmark catalog (the §7.1 catalog) and its baseline sweep, plus the SO holdout cross-table, are run from the authors' workspace and are not bundled in this companion.

---

Worked examples

The paper's §7 reports four headline case studies, split into a validation block and a real-applications block. The two C++ neural-network case studies are under examples/; the analytical BBOB validation is driven from benchmarks/bbob/:

§	Example	Block	Type	Location
8.2	BBOB bi-objective mixed-integer	Validation	Analytical MO	benchmarks/bbob/
8.3	MOEED13 economic-emission dispatch	Validation	Simulator MO	examples/moeed13/
8.4	photo_pce10 (Olympus OPV)	Real application	Real SO	examples/photo_pce10/
8.5	concrete_uci_mo (UCI Concrete)	Real application	Real MO	examples/concrete_uci_mo/

The broader catalog (~30 additional benchmark problems §7.1 refers to — other BBOB suites, other Olympus tasks, classical engineering, chemistry HTE) is reproduced from benchmarks/, plus two extra real-data examples under examples/additional/ that are not shown in the manuscript.

---

Benchmark suite

benchmarks/ contains the validation and catalog runners used in the paper:

• run_olympus.py — Olympus real-data runner; --task selects the task (default photo_pce10; other tasks are not shown in the paper)
• bbob/run_bbob_suites.py — COCO suite driver; --suite selects the suite (default bbob-biobj-mixint, the only one shown explicitly)
• bbob/run_bbob_stress.py — the f15–f24 hard-multimodal stress test (§6.3 limitations)
• baselines/run_baselines.py — pymoo/pycma baselines (CMA-ES/DE/GA/PSO, NSGA-II/III/MOEA-D) for the three §7 example problems, on the same surrogate + YAML as the C++ IDC binaries; emits feasibility and the mean constraint-violation magnitude (see benchmarks/baselines/)
• make_convergence_figure.py — regenerates the §7.4 convergence figure from the committed extra_results/ CSVs
• requirements.txt — Python dependencies for the runners

See benchmarks/README.md for the switch reference and details.

---

Repository structure

IDC/
├── README.md                    ← this file
├── LICENSE                      ← LGPL v3, matching OpenNN
├── CITATION.cff                 ← paper citation metadata
├── CMakeLists.txt               ← C++ build entry point
├── cmake/
│   └── FindOrFetchOpenNN.cmake  ← three-tier OpenNN resolution
├── examples/                    ← C++ NN case studies (§7.3–§7.5) + additional/
├── benchmarks/                  ← BBOB (§7.2) + catalog sweep + figures + switches
├── scripts/                     ← reproduction orchestrators
├── data/                        ← curated dataset subset
└── docs/
    ├── architecture.md          ← IDC overview + how OpenNN is used
    ├── reproducing.md           ← exact reproduction recipe
    └── holdout_procedure.md     ← held-out validation protocol

---

Citation

@article{Scala2026,
  title   = {Neural network response optimization via iterative domain contraction},
  author  = {Scala, Simone and Matias, Jose and Dewil, Raf and Lopez, Roberto},
  journal = {Advances in Computational Science and Engineering},
  volume  = {9},
  pages   = {1--46},
  year    = {2026},
  issn    = {2837-1739},
  publisher = {American Institute of Mathematical Sciences (AIMS)},
  doi     = {10.3934/acse.2026010},
  url     = {https://doi.org/10.3934/acse.2026010}
}

https://doi.org/10.3934/acse.2026010

---

License

Released under the GNU Lesser General Public License v3.0 (LGPL-3.0-or-later) — matching OpenNN.

See LICENSE for the full text.

---

Acknowledgments

Funded by the European Union (Marie Skłodowska-Curie Grant Agreement no. 101169541 — NEUTEN). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Executive Agency (REA). Neither the European Union nor the granting authority can be held responsible for them.