ConceptPrimal-Dual Method / 原始对偶方法
Primal-Dual Method
An algorithmic framework for solving constrained optimization problems. Through Lagrange multipliers, a constrained problem is transformed into a minimax problem in which primal and dual variables are updated alternately, often using a two-timescale update strategy.
Timeline (last 90 days)
Oct 19
Primal-dual methods for constrained Markov decision processes (CMDPs) typically rely on a two-timescale assumption, but when the environment is slowly non-stationary, oscillations in the constraint level or sensitivity break convergence.
Unverified50%