🔬 Scientific Background & Acknowledgments

⬅️ README | 📚 THEORY | 👥 AUTHORS

This project came to life by implementing cutting-edge research in white-box machine learning. We would like to deeply acknowledge the following colleagues for the works, research teams, and foundational authors whose theoretical breakthroughs made the TAM framework possible:

📐 Foundational Theory & Inspiration (GAMs)

  • Simon N. Wood: We acknowledge his definitive foundational work on Generalized Additive Models (GAMs). His research mathematically underpins continuous structural regularization, Penalized Splines, and scale-invariant tensor products.

🏗️ Core WeaKL Framework (StaticTAM, AdaptiveTAM, HierarchicalTAM)

  • Nathan Doumèche, Francis Bach, Éloi Bedek, Gérard Biau, Claire Boyer, and Yannig Goude: Their 2025 research on forecasting time series with constraints serves as the theoretical foundation for the WeaKL algorithm.

  • Nathan Doumèche: His 2025 PhD thesis provides the central reference for the “Online WeaKL” approach, utilized as a tensorial alternative for managing structural breaks.

  • Nathan Doumèche, Francis Bach, Gérard Biau, and Claire Boyer: Their work on Physics-Informed Kernel Learning (PIKL) demonstrates that linear differential constraints can be reduced to a convex quadratic penalty, forming the basis of our exact physics module.

⚖️ Expert Aggregation (OperaTAM)

  • Pierre Gaillard and Yannig Goude: We acknowledge their foundational 2016 OPERA package. Their theoretical frameworks for the MLpol and EWA algorithms are implemented natively for the OperaTAM module.

🌐 Extended Mathematical & Algorithmic Influences

This framework is also built upon the mathematical foundations laid by the researchers cited in the references.bib file, encompassing foundational breakthroughs in generalized linear models, robust statistics, extreme value theory, and conformal prediction.