A constraint-based approach to function interpolation, with application to performance estimation for weakly convex optimisation
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , |
|---|---|
| Format: | Preprint |
| Publié: |
2024
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866908539008581632 |
|---|---|
| author | Rubbens, Anne Hendrickx, Julien M. |
| author_facet | Rubbens, Anne Hendrickx, Julien M. |
| contents | We consider the problem of obtaining interpolation constraints for function classes, i.e., necessary and sufficient constraints that a set of points, function values and (sub)gradients must satisfy to ensure the existence of a global function of the class considered, consistent with this set. The derivation of such constraints is crucial, e.g., in the performance analysis of optimization methods, since obtaining a priori tight performance guarantees requires using a tight description of function classes of interest. We propose an approach that allows setting aside all analytic properties of the function class to work only at an algebraic level, and to obtain counterexamples when a condition characterizing a function class cannot serve as an interpolation constraint. As an illustration, we provide interpolation constraints for the class of weakly convex functions with bounded subgradients, and rely on these constraints to outperform state-of-the-art bounds on the performance of the subgradient method on this class. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2405_08405 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | A constraint-based approach to function interpolation, with application to performance estimation for weakly convex optimisation Rubbens, Anne Hendrickx, Julien M. Optimization and Control 90C25, 90C20, 68Q25, 90C22, 49M29 We consider the problem of obtaining interpolation constraints for function classes, i.e., necessary and sufficient constraints that a set of points, function values and (sub)gradients must satisfy to ensure the existence of a global function of the class considered, consistent with this set. The derivation of such constraints is crucial, e.g., in the performance analysis of optimization methods, since obtaining a priori tight performance guarantees requires using a tight description of function classes of interest. We propose an approach that allows setting aside all analytic properties of the function class to work only at an algebraic level, and to obtain counterexamples when a condition characterizing a function class cannot serve as an interpolation constraint. As an illustration, we provide interpolation constraints for the class of weakly convex functions with bounded subgradients, and rely on these constraints to outperform state-of-the-art bounds on the performance of the subgradient method on this class. |
| title | A constraint-based approach to function interpolation, with application to performance estimation for weakly convex optimisation |
| topic | Optimization and Control 90C25, 90C20, 68Q25, 90C22, 49M29 |
| url | https://arxiv.org/abs/2405.08405 |