Newest papers

Adaptive Normalized Gradient Method for $(L_0,L_1)$-smooth Problems: Estimating Problem Parameters as a Region

R. Okazaki, Shotaro Yagishita, Hideitsu Hino · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.27681 · cited by 0 works (OpenAlex) · open-access PDF

German

Wir schlagen ein parameterfreies und effizientes Gradient-Descent-Algorithmus für $(L_0,L_1)$-smooth-Optimierung vor. Dieses Konzept der Glattheit, eingeführt von Zhang et al. (2020), ermöglicht die Glattheit gebunden mit der Gradient-Norm zu wachsen. Ähnlich wie die Cutting-Plane-Lokalisierung verwendet unsere Methode eine Sequenz von Beobachtungen, um eine Region von plausiblen Werten für $(L_0,L_1)$ zu halten, und verwendet diese Region, um einzelne Schrittgrößen zu wählen. In der konvexen Einstellung erreicht unsere Methode die Standard-Iteration-Komplexität gebunden $\de\mathcal_O_left auf $$0_tiliestation

Machine translation (Meta M2M100); the English original follows.

Abstract

We propose a parameter-free and efficient gradient descent algorithm for $(L_0,L_1)$-smooth optimization. This notion of smoothness, introduced by Zhang et al. (2020), allows the smoothness bound to grow with the gradient norm. Similar to cutting-plane localization, our method uses a sequence of observations to maintain a region of plausible values for $(L_0,L_1)$ and uses this region to select step sizes. In the convex setting, our method attains the standard iteration complexity bound $\tilde{\mathcal{O}}\left(\left(L_0/ε+ L_1^2\right)\|x^0-x^*\|^2\right)$ for finding an $ε$-optimal solution, while in the nonconvex setting it finds an $ε$-stationary point within $\tilde{\mathcal{O}}\left(\left(L_0/ε^2 + L_1/ε\right)(f(x^0)-f^*)\right)$ iterations. To the best of our knowledge, this is the first work to establish, for a single parameter-free algorithm, the best-known iteration-complexity guarantee in the nonconvex setting while also providing the above iteration-complexity guarantee in the convex setting. Experiments on logistic regression and real phase retrieval demonstrate that our method can exploit $(L_0,L_1)$-smoothness and perform efficiently on nonconvex problems.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Smoothness, Gradient descent, Mathematics, Regular polygon, Mathematical optimization, Gradient method

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange