Bilinear Quantum Learning / Hirota Quantum Bilinear Learning

This repository is the canonical implementation and evidence package for a gate-based Bilinear Quantum Learning (BQL) foundation and its Hirota Quantum Bilinear Learning (HQBL) hierarchy.

The project is governed by falsification-first gates:

1. no backend skip is promoted to a pass;
2. no learning experiment opens before exact mathematical invariants pass;
3. classical, quantum, ordinary-bilinear, and Hirota comparisons are matched;
4. negative results remain part of the evidence package;
5. quantum advantage is not an assumed contribution.

Canonical runtime

The reproducible binary runtime is Linux x86-64, CPython 3.12, TensorFlow 2.18.1, TF-Keras 2.18, TensorFlow Quantum 0.7.6, Cirq 1.5, NumPy 2.0, SciPy 1.15, and SymPy 1.14. The exact environment is verified by scripts/verify_environment.py; version drift is a hard failure.

Repository map

• src/bilinear_quantum: installable TensorFlow/TFQ library;
• tests: mathematical, backend, layer, serialization, and gate tests;
• configs: immutable experiment specifications;
• scripts: environment, gate, experiment, analysis, and release entrypoints;
• docs: master specification, traceability, and library documentation;
• manuscript: submission manuscript and supplementary material;
• references: immutable source packages supplied for the project;
• artifacts: generated evidence, reports, figures, and tables.
• manuscript: editable publication sources and bibliography;
• output: generated DOCX/PDF manuscripts, packages, and release bundle.

Current status

Work is accepted only through machine-readable gate reports. See artifacts/gates/ after executing the canonical gate runner.

Install and verify

Install the immutable public release on Linux x86-64 with CPython 3.12:

python -m pip install bilinear-quantum==1.0.0

Run a packaged smoke experiment without a repository checkout:

bq-run-experiment E1 --profile smoke --output-root reproduced/smoke

For development and full-profile evidence reproduction from a clean source release:

python -m pip install -r requirements-lock.txt
python -m pip install --no-deps -e .
export TF_USE_LEGACY_KERAS=1
bq-run-gates --project-root .

To reproduce into a fresh immutable destination without touching the evidence shipped with the release:

python scripts/run_reproduction.py \
  --profile full \
  --artifact-root reproduced/full-20260825

The exact API contract is documented in docs/API.md; the end-to-end data and evidence procedure is in docs/REPRODUCIBILITY.md. Full-profile experiments refuse an uncommitted or dirty source tree, and completed output directories cannot be overwritten.

Authors and citation

The software release is authored by Nguyen Minh Tuan and Bui Phi Hung, Faculty of Information Technology, Posts and Telecommunications Institute of Technology, Ho Chi Minh City, Vietnam. Citation metadata is provided in CITATION.cff.

License and third-party software

Original project code is released under the Apache License 2.0. Runtime dependencies remain governed by their respective licenses and are installed separately; no Elsevier template or third-party research document is bundled in the Python distribution.