<会議発表論文>
Adaptive Proximal Gradient Methods Are Universal Without Approximation

作成者
本文言語
出版者
発行日
収録物名
開始ページ
終了ページ
会議情報
出版タイプ
アクセス権
権利関係
権利関係
関連DOI
関連URI
関連HDL
概要 We show that adaptive proximal gradient methods for convex problems are not restricted to traditional Lipschitzian assumptions. Our analysis reveals that a class of linesearch-free methods is still co...nvergent under mere local Hölder gradient continuity, covering in particular continuously differentiable semi-algebraic functions. To mitigate the lack of local Lipschitz continuity, popular approaches revolve around ε-oracles and/or linesearch procedures. In contrast, we exploit plain Hölder inequalities not entailing any approximation, all while retaining the linesearch-free nature of adaptive schemes. Furthermore, we prove full sequence convergence without prior knowledge of local Hölder constants nor of the order of Hölder continuity. Numerical experiments make comparisons with baseline methods on diverse tasks from machine learning covering both the locally and the globally Hölder setting.続きを見る

本文ファイル

pdf 7234388 pdf 1.75 MB 11  

詳細

EISSN
レコードID
関連URI
助成情報
登録日 2024.09.13
更新日 2024.09.18