A general method of finding the direction of descent in \(\varepsilon\)-subgradient methods (Q1912454)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | A general method of finding the direction of descent in \(\varepsilon\)-subgradient methods |
scientific article |
Statements
A general method of finding the direction of descent in \(\varepsilon\)-subgradient methods (English)
0 references
30 May 1996
0 references
Let \(f(x)\) be a quasiconvex function defined on \(E^n\). For \(x\in E^n\), \(\varepsilon>0\) define \[ W_\varepsilon(f,x)= \{g\in E^n;\;\langle g,y-x\rangle\leq 0,\;\forall y\in E_\varepsilon(f,x)\}, \] where \(E_\varepsilon(f,x)= \{y;\;y\in E^n,\;f(y)<f(x)-\varepsilon\}\). The elements of the set \(W_\varepsilon(f,x)\) are called \(\varepsilon\)-generalized support vectors for the function \(f(x)\) at the point \(x\). In this paper, the problem of finding a direction \(S\in E^n\) at the point \(x\) along with the function \(f(x)\) can be decreased by at least \(\varepsilon\). If no such direction exists, then \(x\) is an \(\varepsilon\)-optimal point. This problem lies at the basis of the problem of choosing the direction of descent in the \(\varepsilon\)-subgradient method.
0 references
minimization
0 references
descent direction
0 references
\(\varepsilon\)-subgradient method
0 references