Skip to content

Ethics & Scientific Deontology

This document outlines the ethical guidelines, scientific deontology principles, and standards of academic integrity governing research software developed with this deep learning boilerplate template.

📜 Scientific Deontology Principles

Scientific software engineering requires the same rigor, transparency, and reproducibility as empirical experiment design.

1. Transparency & Reproducibility

  • All hyperparameter configurations, random seeds, data preprocessing steps, and model weights must be explicitly logged and archived.
  • Deterministic flags (torch.use_deterministic_algorithms(True)) should be enabled for published baseline benchmarks.

2. Peer-Reviewed Citations & Provenance

  • Every neural layer, loss function, and architectural primitive implemented in this repository includes peer-reviewed paper citations (Author, Venue, Year, DOI/arXiv URL) directly in class docstrings.
  • Developers extending this codebase must maintain citation provenance for external algorithms and baseline implementations.

3. Ethical AI & Automated Assistance

  • AI pair programming assistants (e.g. Google's Gemini, Antigravity) are recognized as productivity tools.
  • Authors remain solely responsible for the scientific validity, correctness, and accuracy of published models and experimental claims.

⚖️ Research Integrity Checklist

  • No Ghost Writing: All automated code generation is audited by domain experts.
  • Citation Transparency: Original paper authors are credited in docstrings and documentation.
  • Data Integrity: Datasets are accessed legally according to publisher licenses.
  • Open Access: Code and benchmarks are published openly under MIT License.