For non-convex objectives satisfying the Łojasiewicz inequality of order 2, the DIN continuous-time dynamics converge in function value at a rate arbitrarily close to e^{-2√µ t}, which is worst-case optimal within this dynamics family.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.OC 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Heavy-ball dynamics with Hessian-driven damping for non-convex optimization under the {\L}ojasiewicz condition
For non-convex objectives satisfying the Łojasiewicz inequality of order 2, the DIN continuous-time dynamics converge in function value at a rate arbitrarily close to e^{-2√µ t}, which is worst-case optimal within this dynamics family.