Observable upper-model, gradient, cocoercivity, interpolation, and descent inequalities.
The quadratic upper model with estimate M bounds the objective at y.
Equations
Instances For
The two observed gradients satisfy the proposed Lipschitz bound in the dual norm.
Equations
Instances For
Exact current orientation: D_f(x,y) uses the gradient at y.
Equations
- V7.CocoercivityGuard p M oracle x y = (V7.BregmanRemainder oracle x y ≥ V7.lpNorm (V7.conjugateExponent p) (oracle.gradient x - oracle.gradient y) ^ 2 / (2 * M))
Instances For
The observed Euclidean Bregman gap dominates the squared gradient difference.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The terminal gradient step achieves the decrease predicted by the smoothness estimate.
Equations
Instances For
The observable inequalities that a local trial may check.
- upperModel : ObservableGuardKind
- gradient : ObservableGuardKind
- cocoercivity : ObservableGuardKind
- interpolation : ObservableGuardKind
- terminalDescent : ObservableGuardKind
Instances For
The kind and pair of points witnessing a failed observable guard.
- kind : ObservableGuardKind
The inequality that failed.
- x : Point d
The first point of the failed guard.
- y : Point d
The second point of the failed guard.
Instances For
The selected observable inequality fails for the supplied oracle and estimates.
Equations
- One or more equations did not get rendered due to their size.