FIDE ranks experimental designs by reference-relative law revision
Original title: FIDE: Fiber-Informed Differentiable Experimental Design for Reference-Relative Law Dynamics
It completes aggregate constraints against a frozen baseline and prefers the design needing the least dynamical correction.
FIDE addresses residual freedom in experimental design when aggregate measurements leave an evolving population law underdetermined and an existing dynamical model stays the baseline outside task-valued quantities. Each candidate's aggregate constraints are completed by information projection relative to a common frozen reference. Designs that are scientifically adequate are then ranked by the dynamical revision needed to realize the completed paths. The Full action is the minimum mean-squared velocity correction; its square root is a sharp reference-relative expectation-rate discrepancy. Experiments indicate the preference changes sensor selection, can help or harm withheld predictions depending on baseline reliability, and remains usable for a 32-dimensional configuration law.
Unconfirmed: Whether the preference helps or harms predictions depends on baseline reliability, as reported by the authors.
- The method targets baseline-preserving design rather than generic truth recovery.
- Full action is the minimum mean-squared velocity correction for a completed path.
- Sensor selection changes under this preference, according to the experiments.
- Effects on withheld predictions depend on whether the baseline is reliable.
- The approach stays usable for a 32-dimensional configuration law.
From the original article
1 min read at arxiv.orgAggregate measurements generally leave an evolving population law underdetermined, so different measurement systems can be equally adequate for a scientific task while constraining different aspects of the law. We study how to resolve this residual design freedom in baseline-preserving settings, where an existing dynamical model remains the working baseline outside quantities explicitly valued by the task.
By Pietro Zanotta · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.