{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

3sgmethod

# 3sgmethod - EE236C(Spring 2008-09 3 Subgradient method...

This preview shows pages 1–4. Sign up to view the full content.

EE236C (Spring 2008-09) 3. Subgradient method subgradient method convergence analysis optimal step size when f is known alternating projections optimality 3–1 Subgradient method to minimize a nondifferentiable convex function f : choose x (0) and repeat x ( k ) = x ( k 1) t k g ( k 1) , k = 1 , 2 , . . . g ( k 1) is any subgradient of f at x ( k 1) step size rules fixed step: t k constant fixed length: t k bardbl g ( k 1) bardbl 2 constant ( i.e. , bardbl x ( k ) x ( k 1) bardbl 2 constant) diminishing: t k 0 , k =1 t k = Subgradient method 3–2

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

View Full Document
Assumptions f has finite optimal value f , minimizer x f is convex, dom f = R n f is Lipschitz continuous with constant G > 0 : | f ( x ) f ( y ) | ≤ G bardbl x y bardbl 2 x, y this is equivalent to bardbl g bardbl 2 G for all g ∂f ( x ) , all x Subgradient method 3–3 Analysis the subgradient method is not a descent method the key quantity in the analysis is the distance to the optimal set bardbl x ( i ) x bardbl 2 2 = vextenddouble vextenddouble vextenddouble x ( i 1) t i g ( i 1) x vextenddouble vextenddouble vextenddouble 2 2 = bardbl x ( i 1) x bardbl 2 2 2 t i g ( i 1) T ( x ( i 1) x ) + t 2 i bardbl g ( i 1) bardbl 2 2 bardbl x ( i 1) x bardbl 2 2 2 t i parenleftBig f ( x ( i 1) ) f parenrightBig + t 2 i bardbl g ( i 1) bardbl 2 2 define f ( k ) best = min 0 i<k f ( x ( i ) ) , and combine inequalities for i = 1 , . . . , k : 2( k summationdisplay i =1 t i ) parenleftBig f ( k ) best f parenrightBig bardbl x (0) x bardbl 2 2 − bardbl x ( k ) x bardbl 2 2 + k summationdisplay i =1 t 2 i bardbl g ( i 1) bardbl 2 2 bardbl x (0) x bardbl 2 2 + k summationdisplay i =1 t 2 i bardbl g ( i 1) bardbl 2 2 Subgradient method 3–4
fixed step size t i = t f ( k ) best f bardbl x (0) x bardbl 2 2 + kt 2 G 2 2 kt does not guarantee convergence of f ( k ) best for large k , f ( k ) best

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

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}