Projected Newton-type methods in machine learning

Schmidt, M.; Kim, D.; Sra, S.

Lokale TagsFreigabegeschichteDetailsÜbersicht

Projected Newton-type methods in machine learning

Schmidt, M., Kim, D., & Sra, S. (2011). Projected Newton-type methods in machine learning. In S. Sra, S. Nowozin, & S. J. Wright (Eds.), Optimization for Machine Learning (pp. 305-330). Cambridge, MA, USA: MIT Press.

Item is Freigegeben

einblenden: alle ausblenden: alle

Basisdaten

einblenden: ausblenden:

Datensatz-Permalink: https://hdl.handle.net/11858/00-001M-0000-0010-4BF4-B Versions-Permalink: https://hdl.handle.net/11858/00-001M-0000-0010-4BF5-9

Genre: Buchkapitel

ausblenden:

Urheber:
Schmidt, M.^{1, 2}, Autor
Kim, D., Autor
Sra, S.³, Autor

Affiliations:
1Dept. Metastable and Low-Dimensional Materials, Max Planck Institute for Intelligent Systems, Max Planck Society, ou_1497645
2Dept. Modern Magnetic Systems, Max Planck Institute for Intelligent Systems, Max Planck Society, ou_1497648
3Dept. Empirical Inference, Max Planck Institute for Intelligent Systems, Max Planck Society, ou_1497647

Inhalt

einblenden:

ausblenden:

Schlagwörter: MPI für Intelligente Systeme; Abt. Schölkopf;

Zusammenfassung: We consider projected Newton-type methods for solving large-scale optimization problems arising in machine learning and related fields. We first introduce an algorithmic framework for projected Newton-type methods by reviewing a canonical projected (quasi-)Newton method. This method, while conceptually pleasing, has a high computation cost per iteration. Thus, we discuss two variants that are more scalable, namely, two-metric projection and inexact projection methods. Finally, we show how to apply the Newton-type framework to handle non-smooth objectives. Examples are provided throughout the chapter to illustrate machine learning applications of our framework.