L21_sgdmet - Lecture 21 Subgradient Methods April 9 2007...

Info iconThis preview shows pages 1–8. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Lecture 21: Subgradient Methods April 9, 2007 Lecture 21 Outline • Subgradients and Level Sets • Subgradient Method • Convergence and Convergence Rate Convex Optimization 1 Lecture 21 Subgradients and Level Sets A vector s is a subgradient of a convex function f : R n 7→ R at ˆ x ∈ dom f when subgradient inequality holds f ( z ) ≥ f (ˆ x ) + s T ( z- ˆ x ) for all z ∈ dom f • We have interpreted the subgradient inequality in terms of a hyperplane in R n +1 supporting the epigraph epi f at (ˆ x,f (ˆ x )) Convex Optimization 2 Lecture 21 • Projecting the hyperplane and the epigraph on the set { ( x,w ) | x ∈ R n , w = f (ˆ x ) } • The subgradient inequality can be interpreted in terms of a hyperplane in R n supporting the level set L γ ( f ) at ˆ x for γ = f (ˆ x ) • For z in the level set L f (ˆ x ) ( f ) = { z | f ( z ) ≤ f (ˆ x ) } , the subgradient inequality implies s T ( z- ˆ x ) ≤ for all z ∈ L f (ˆ x ) ( f ) and the inequality is tight at z = ˆ x . Convex Optimization 3 Lecture 21 Another Interpretation of Subgradients • A subgradient of f at any ˆ x defines a hyperplane that cuts R n in two regions and one of them contains the level sets L γ ( f ) for all γ ≤ f (ˆ x ) • The optimal set X * lies entirely on one side of each such hyperplane • Cutting-plane method can be interpreted as successively “reducing” the size of a region where the “optimal solutions” are located, by increasing the number of hyperplanes that “cut” the space R n Convex Optimization 4 Lecture 21 Subgradient Method Constrained Convex Minimization minimize f ( x ) subject to x ∈ X General Assumption • The function f ( x ) is convex and dom f = R n • The set X is nonempty closed and convex Subgradient Method • Having the current iterate x k , a subgradient s k of f at x k , and a stepsize value α k > , the next iterate is given by x k +1 = P X [ x k- α k s k ] • Polyak 1969, Ermoliev 1969, Shor 1985 Convex Optimization 5 Lecture 21 Property “ensuring” that subgradient methods work Along the negative subgradient direction...
View Full Document

This note was uploaded on 08/22/2008 for the course GE 498 AN taught by Professor Angelianedich during the Spring '07 term at University of Illinois at Urbana–Champaign.

Page1 / 19

L21_sgdmet - Lecture 21 Subgradient Methods April 9 2007...

This preview shows document pages 1 - 8. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online