Proposes LCG method achieving O(ε^{-2} log(ε^{-1})) iteration complexity for convex functional constrained optimization and IPP-LCG with O(ε^{-3} log(ε^{-1})) for nonconvex cases, without dependence on optimal dual multiplier magnitude.
Let (x∗,y ∗) ∈ Rn × Rm + be the saddle point of the convex constrained problem ( 1.1)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.OC 1years
2022 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Projection-Free Functional Constrained Optimization for Risk Aversion and Sparsity Control
Proposes LCG method achieving O(ε^{-2} log(ε^{-1})) iteration complexity for convex functional constrained optimization and IPP-LCG with O(ε^{-3} log(ε^{-1})) for nonconvex cases, without dependence on optimal dual multiplier magnitude.