FIDE
A differentiable experimental design method for benchmark-preserving settings. It uses information projection to complete aggregate constraints into full dynamics paths, ranks candidate designs by a minimum-correction principle, and supports gradient-based optimization.
Timeline (last 90 days)
The paper "FIDE: Fiber-Informed Differentiable Experimental Design for Reference-Relative Law Dynamics" proposes the Fiber-Informed Differentiable Experimental Design (FIDE) method for experimental design in benchmark-preserving settings, providing a conditional minimal-correction principle.
FIDE uses information projection to complete the aggregate constraints produced by each candidate design into a full dynamical path, that is, taking the frozen benchmark model as reference and selecting, among all possible paths satisfying the observation constraints, the path with the minimal correction relative to the benchmark.
FIDE defines a metric called "Full action," namely the minimum mean-square velocity correction required to realize the full path; its square root is the sharp reference-relative expected rate difference on the gradient-normalized Sobolev class.
FIDE's computation process is differentiable, allowing gradient-based methods for optimization; experiments show that the method remains computationally usable for 32-dimensional configuration laws.
Experiments found that FIDE's preferences changed sensor selection, favoring designs that make the minimal correction to the benchmark model.
Depending on the reliability of the benchmark, FIDE may help or harm held-out downstream predictions: when the benchmark model is reliable, FIDE can improve prediction performance; when the benchmark model is unreliable, it may have a negative impact.
FIDE can be used prospectively from aggregate predictions, that is, making design choices without real data.
FIDE offers a new perspective on experimental design: in benchmark-preserving settings, one should not seek to recover the "true" law, but rather to make minimal corrections on top of the existing benchmark.
All Facts (8)
The paper "FIDE: Fiber-Informed Differentiable Experimental Design for Reference-Relative Law Dynamics" proposes the Fiber-Informed Differentiable Experimental Design (FIDE) method for experimental design in benchmark-preserving settings, providing a conditional minimal-correction principle.
50%UnverifiedFIDE uses information projection to complete the aggregate constraints produced by each candidate design into a full dynamical path, that is, taking the frozen benchmark model as reference and selecting, among all possible paths satisfying the observation constraints, the path with the minimal correction relative to the benchmark.
50%UnverifiedFIDE defines a metric called "Full action," namely the minimum mean-square velocity correction required to realize the full path; its square root is the sharp reference-relative expected rate difference on the gradient-normalized Sobolev class.
50%UnverifiedFIDE's computation process is differentiable, allowing gradient-based methods for optimization; experiments show that the method remains computationally usable for 32-dimensional configuration laws.
50%UnverifiedExperiments found that FIDE's preferences changed sensor selection, favoring designs that make the minimal correction to the benchmark model.
50%UnverifiedDepending on the reliability of the benchmark, FIDE may help or harm held-out downstream predictions: when the benchmark model is reliable, FIDE can improve prediction performance; when the benchmark model is unreliable, it may have a negative impact.
50%UnverifiedFIDE can be used prospectively from aggregate predictions, that is, making design choices without real data.
50%UnverifiedFIDE offers a new perspective on experimental design: in benchmark-preserving settings, one should not seek to recover the "true" law, but rather to make minimal corrections on top of the existing benchmark.
50%