Bilinear Quantum

bilinear-quantum 3.0 is a task-oriented TensorFlow/Keras library built around the Bilinear Quantum Interaction Graph (BQIG). Ordinary users provide data and an objective; the library handles preprocessing, frozen recipe selection, optimization, validation, early stopping, and evidence metadata.

Version 3.0.1 is a maintenance release focused on runtime compatibility and packaging. It does not claim universal model or computational quantum advantage. Such claims require locked, independent confirmatory benchmarks.

Install

Python 3.12 or 3.13:

python -m pip install bilinear-quantum==3.0.1

The public distribution is a CPython-specific, compiled-only wheel. It does not contain implementation source, private recipes, experiment registries, datasets, checkpoints, or unpublished evidence.

Use without model configuration

import bilinear_quantum as bq

classifier = bq.learn.classify(data, target="label")
predictions = classifier.predict(new_data)
probabilities = classifier.predict_proba(new_data)

forecast = bq.sequence.forecast(
    time_series,
    target="power",
    horizon=24,
)
print(forecast.values)

The public task namespaces are:

• bq.learn: classification, regression, embeddings, and clustering;
• bq.sequence: classification, forecasting, events, segmentation, anomalies;
• bq.fusion: paired-source learning, alignment, and matching;
• bq.relation: matching, ranking, retrieval, and recommendation;
• bq.graph: node, edge, and link prediction;
• bq.science: surrogates, dynamics, operator discovery, and data-driven solves;
• bq.quantum: execution brokering, shots, noise, and resources;
• bq.audit: certification, paired comparison, and evidence export.

Researchers can freeze an explicit protocol without changing the ordinary API:

protocol = bq.research.Protocol(
    name="subject-disjoint-confirmatory-v1",
    problem="classification",
    modality="sequence",
    seed=3729,
)
result = bq.research.fit(windows, labels, protocol, groups=subject_ids)

Runtime design

TensorFlow 2.20/Keras 3 is the default training runtime. PennyLane and Cirq are optional quantum execution extras rather than default installation dependencies. TensorFlow Quantum 0.7.6 is kept in an isolated Python 3.12 reference environment because it is not dependency-compatible with the main Python 3.13 runtime. It is never silently emulated.

Scientific safeguards

• preprocessing is fitted on training rows only;
• group/subject and temporal holdouts are first-class split policies;
• recipe selection uses validation data, never confirmatory test data;
• manifests record recipe, split hash, seed, backend, and versions;
• fallback use is explicit and cannot be counted as a BQL win;
• negative and failed benchmark results remain evidence;
• exact BQL/HQBL identities are separated from empirical Deep-BQL claims.

License

Copyright (c) 2026 Nguyen Minh Tuan and Bui Phi Hung. The compiled package is licensed under the accompanying Bilinear Quantum Binary License. It is not an MIT or open-source distribution. Third-party dependencies retain their own licenses.

Project: https://github.com/Thien-y1502/bilinear-quantum

Package: https://pypi.org/project/bilinear-quantum/