arXiv:2606.21199v2 Announce Type: replace-cross Abstract: We introduce a semi-parametric framework for nonlinear system identification, which decouples discrepancy functions from physics-based components. Orthogonal Gaussian process regression balances sparse parameter selection (the white box) with discrepancy learning (the black box) to produce interpretable models from incomplete physics.
The increasing complexity of AI models and the demand for explainability in scientific and engineering applications are driving the need for frameworks that integrate physics with machine learning.
This development offers a pathway to more interpretable and reliable AI models, particularly in domains where physical laws govern system behavior, bridging the gap between 'black-box' and 'white-box' AI.
The ability to decouple discrepancy functions from physics-based components will allow for more robust and transparent system identification, potentially accelerating discovery and optimization in complex physical systems.
- · AI researchers (interpretable AI)
- · Engineers (system identification)
- · Scientific computing sector
- · Manufacturing (predictive maintenance)
- · Pure 'black-box' AI solutions in scientific domains
- · Traditional modeling approaches without AI integration
- · Sectors reliant on opaque, uninterpretable models
Improved simulation accuracy and predictive power in fields like materials science, fluid dynamics, and climate modeling.
Accelerated design cycles for new technologies by enabling AI to learn from incomplete physics while accounting for unknown discrepancies.
Potential for autonomous scientific discovery systems that can formulate hypotheses, conduct experiments, and refine models based on real-world observations and physical principles.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG