объясни как работает variational inequality для за...
Создано: 12 апреля 2025 г.
Создано: 12 апреля 2025 г.
объясни что здесь происходит:
Variational inequalities for convex cost functions. Let X ⊂ R
d
and F : X → R
d be a continuous mapping. The variational inequality problem (Harker and Pang, 1990) is:
find x
∗ ∈ X such that F(x
∗
)
(x − x
∗
) ≥ 0 , ∀x ∈ X . (VIP)
We call optimal set the set X
∗ of x ∈ X verifying (VIP). A standard assumption on F is monotonicity:
(F(x) − F(y))>(x − y) ≥ 0 ∀ x, y ∈ X . (8)
If F(x) = ∇f(x), it is equivalent to f being convex. If F can be written as (6), it implies that the cost functions are convex.6 When the operator F is monotone, we have that
F(x
∗
)
(x − x
∗
) ≤ F(x)
(x − x
∗
), ∀x, x
∗
. Hence, in this case (VIP) implies a stronger
formulation sometimes called Minty variational inequality (Crespi et al., 2005):
find x
∗ ∈ X such that F(x)
(x − x
∗
) ≥ 0 , ∀x ∈ X . (MVI)
This formulation is stronger in the sense that if (MVI) holds for some x
∗ ∈ X , then (VIP) holds
too. A merit function useful for our analysis can be derived from this formulation. Roughly, a merit
function is a convergence measure. More formally, a function g : X → R is called a merit function
if g is non-negative such that g(x) = 0 ⇔ x ∈ X ∗
(Larsson and Patriksson, 1994). A way to derive
a merit function from (MVI) would be to use g(x
∗
) = supx∈X F(x)
(x
∗ − x) which is zero if and
only if (MVI) holds for x
∗
. To deal with unbounded constraint sets (leading to a potentially infinite
valued function outside of the optimal set), we use the restricted merit function (Nesterov, 2007):
ErrR(xt)
def
= max
x∈X ,kx−x0k≤R
F(x)
(xt − x). (9)
This function acts as merit function for (VIP) on the interior of the open ball of radius R around x0,
as shown in Lemma 1 of Nesterov (2007). That is, let XR := X ∩ {x : kx − x0k < R}. Then for
any point xˆ ∈ XR, we have:
ErrR(xˆ) = 0 ⇔ xˆ ∈ X ∗ ∩ XR. (10)
The reference point x0 is arbitrary, but in practice it is usually the initialization point of the algorithm.
R has to be big enough to ensure that XR contains a solution. ErrR measures how much (MVI) is
violated on the restriction XR. Such merit function is standard in the variational inequality literature.
A similar one is used in (Nemirovski, 2004; Juditsky et al., 2011). When F is derived from the
gradients (6) of a zero-sum game (SP), we can define a more interpretable merit function (see
App. B.1). One has to be careful though when extending properties from the minimization setting to
the saddle point setting (e.g. the merit function used by Yadav et al. (2018) is vacuous for a bilinear
game as explained in App B.2).
Ниже краткое объяснение того, что описывается в приведённом фрагменте:
Задача вариационного неравенства (VIP)
Монотонность оператора
Стандартное и «укреплённое» вариационное неравенство
Мерит-функции и их роль
Почему это нужно
Связь с играми
Итого, в тексте описывается: