Surgical Hill-Kinetics Field Saturation: Resolving Finite-Time Explosion and NFE Stiffness in Neural Ordinary Differential Equations
Continuous-Time Neural Networks and Neural Ordinary Differential Equations (Neural ODEs) suffer from severe numerical stiffness and finite-time gradient blowup under polynomial weight accumulation, leading to adaptive solver grid collapse (NFE -> infinity), loss oscillations, and machine-precision failures. We propose Surgical Hill-Kinetics Field Saturation...
Authors
- Sjfu
Institutions
- Nexen (Canada) (CA)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-08-03
- DOI
- https://doi.org/10.5281/zenodo.21779035
- Primary Topic
- Model Reduction and Neural Networks
- Type
- preprint