Hyperparameter Trajectory Inference with Conditional Lagrangian Optimal Transport
Harry Amad, Mihaela van der Schaar · 2026
Read Hyperparameter Trajectory Inference with Conditional Lagrangian Optimal Transport by Harry Amad, Mihaela van der Schaar (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-01771.txt.
Description
Harry Amad and Mihaela van der Schaar predict how a neural network output shifts when its design settings change. Their 2026 work names that task Hyperparameter Trajectory Inference. Settings like reward weights are fixed at design time, yet user tastes can drift after deployment and force costly retraining.
Their approach learns from observed outputs how the output distribution moves as settings vary. A surrogate model then mimics the network at settings never actually trained. The method rests on conditional Lagrangian optimal transport, which learns both the motion law and the best maps between output states. Biases from the manifold hypothesis and least-action ideas keep inferred paths feasible. Tests show the surrogate rebuilds network outputs across setting ranges better than rival methods, which could spare many retrains.
Catalog record
| CREATOR | Harry Amad, Mihaela van der Schaar |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 82963 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.01771