Abstract:In this work, we introduce a novel strategy for tackling constrained optimization problems through a modified penalty method. Conventional penalty methods convert constrained problems into unconstrained ones by incorporating constraints into the loss function via a penalty term. However, selecting an optimal penalty parameter remains challenging; an improper choice, whether excessively high or low, can significantly impede the discovery of the true solution. This challenge is particularly evident when training neural networks for constrained optimization, where tuning parameters can become an extensive and laborious task. To overcome these issues, we propose an adversarial approach that redefines the conventional penalty method by simultaneously considering two competing penalty problems--a technique we term the penalty adversarial problem. Within linear settings, our method not only ensures the fulfillment of constraints but also guarantees solvability, leading to more precise solutions compared to traditional approaches. We further reveal that our method effectively performs an automatic adjustment of penalty parameters by leveraging the relationship between the objective and loss functions, thereby obviating the need for manual parameter tuning. Additionally, we extend this adversarial framework to develop a neural network-based solution for optimal control problems governed by linear or nonlinear partial differential equations. We demonstrate the efficacy of this innovative approach through a series of numerical examples.

Adaptive Penalty-Based Neurodynamic Approach for Nonsmooth Interval-Valued Optimization Problem

A Novel Penalty Approach for Nonlinear Dynamic Optimization Problems with Inequality Path Constraints

An Improved Smoothing Technique-Based Control Vector Parameterization Method for Optimal Control Problems with Inequality Path Constraints

A subgradient-based neurodynamic algorithm to constrained nonsmooth nonconvex interval-valued optimization

An Effective Simultaneous Approach with Variable Time Nodes for Dynamic Optimization Problems

A smoothing approximation-based adaptive neurodynamic approach for nonsmooth resource allocation problem

Enhancing neurodynamic approach with physics-informed neural networks for solving non-smooth convex optimization problems

A Zeroing Neurodynamic Approach for Solving Time-Varying Optimization Problem with Interval-Valued Cost Function

Distributed Neurodynamic Approaches to Nonsmooth Optimization Problems with Inequality and Set Constraints

A Novel Neurodynamic Approach to Constrained Complex-Variable Pseudoconvex Optimization.

A Penalty-Like Neurodynamic Approach to Convex Optimization Problems with Set Constraint

A Penalty-Like Neurodynamic Approach to Constrained Nonsmooth Distributed Convex Optimization

A neurodynamic approach to nonlinear optimization problems with affine equality and convex inequality constraints.

A New Methodology for Robust Optimizations of Optimal Design Problems under Interval Uncertainty

A Collective Neurodynamic Penalty Approach to Nonconvex Distributed Constrained Optimization

A Neurodynamic Approach to Nonsmooth Constrained Pseudoconvex Optimization Problem

Adaptive Dynamic Programming for a Class of Complex-Valued Nonlinear Systems

An adaptive exact penalty method for nonsmooth optimal control problems with nonsmooth nonconvex state and control constraints

Inverse Optimal Adaptive Neural Control for State-Constrained Nonlinear Systems

A Continuous-Time Neurodynamic Approach and Its Discretization for Distributed Convex Optimization over Multi-Agent Systems

Penalty Adversarial Network (PAN): A neural network-based method to solve PDE-constrained optimal control problems