A warm-start plus early-stopping estimator gives variable importance close to full retraining for wide neural networks and gradient-boosted trees, with lower compute.
A time-dependent energy-momentum method
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
abstract
We devise a generalisation of the energy momentum-method for studying the stability of non-autonomous Hamiltonian systems with a Lie group of Hamiltonian symmetries. A generalisation of the relative equilibrium point notion to a non-autonomous realm is provided and studied. Relative equilibrium points of non-autonomous Hamiltonian systems are described via foliated Lie systems, which opens a new field of application of such differential equations. We reduce non-autonomous Hamiltonian systems via the Marsden-Weinstein theorem and we provide conditions ensuring the stability of the projection of relative equilibrium points to the reduced space. As an application, we study the stability of relative equilibrium points for a class of mechanical systems, which covers rigid bodies as a particular instance.
fields
stat.ML 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Reliable and scalable variable importance estimation via warm-start and early stopping
A warm-start plus early-stopping estimator gives variable importance close to full retraining for wide neural networks and gradient-boosted trees, with lower compute.