A method for searching for a stochastic saddle point with constraints satisfied almost surely.

The problem of searching for \(x^{*}\) and \(y^{*}\) satisfying the minimax \[ F^{*} = \min_{x\in X} \max_{y\in Y} F(x,y), \qquad F(x,y) = E_z f(x, y, z), \quad \tag{1} \] where \[ X =\{x\in X \mid g(x,y)\leq 0 \text{ a.s.r. } \forall z \in Z \}, \quad Y =\{y\in Y \mid h(x,y)\leq 0\text{ a.s.r. } \forall z \in Z \}, \tag{2} \] and \(E_{z}\) is a mean on \(z \in Z\), is considered. The term ``almost surely restrictions and the mean \(E_{z}\) are understood in the sense of a measure \(P\) of a probability space \((\Omega, A, P)\), where \(\Omega = Z\) and \(A\) is a \(\sigma\)-algebra on \(\Omega\). The problem (1), (2) can be interpreted as an antagonistic 2-person game with the winning (loosing) function \(f(x,y,z)\) depending both on the strategies of the players \(x\) and \(y\) and on the subsequent realization of the accidental factor \(z\) distributed on \(Z\) in accordance with the probability measure \(P\). An algorithm for searching a value and satisfying the minimax convex-concave regression function with restrictions which must be almost surely realized (a.s.r.), is proposed. Weaker restrictions providing the convergence of the algorithm are found.

Recommendations

Cited in

(5)

This page was built for publication: A method for searching for a stochastic saddle point with constraints satisfied almost surely.

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1348537)