Stochastic relaxed inertial forward-backward-forward splitting for monotone inclusions in Hilbert spaces (Q2082546)

The problem of monotone inclusions of an operator, given by the sum of a maximal monotone operator and a single-valued monotone, Lipschitz continuous, and expectation-valued operator, defined on a Hilbert space. A stochastic extension of the relaxed inertial forward-backward-forward method is considered. Employing an online variance reduction strategy via a mini-batch approach, it is shown that the proposed method produces a sequence that weakly converges to the solution set. The rate at which the discrete velocity of the stochastic process vanishes, is estimated. Under strong monotonicity, strong convergence, a comprehensive assessment of the iteration and oracle complexity of the scheme, are demonstrated. It is shown, for instance, that when the mini-batch is raised at a geometric rate, the rate statement can be strengthened to a linear rate while the oracle complexity of computing an \(\epsilon\)-solution improves to \(\mathcal{O}(1/\epsilon)\). This allows for possibly biased oracles, which in turn, permits a far broader applicability. Defining a restricted gap function based on the Fitzpatrick function, it is shown that the expected gap of an averaged sequence diminishes at a sublinear rate of \(\mathcal{O}(1/k)\) while the oracle complexity of computing a suitably defined \(\epsilon\)-solution is \(\mathcal{O}(1/{\epsilon}^{1+a})\) for \(a > 1.\) Numerical results on two-stage games and an overlapping group Lasso problem are presented as illustrations.