Adaptive Moment (Adam) Estimation Optimization Applied to AVM-FEM for Rapid Convergence

Adaptive Moment Estimation (Adam) is a kind of gradient descent optimization algorithm and by far the most common way to optimize neural networks. This paper proposes a gradient-based EM optimization method that updates parameters by means of Adam, based on gradients calculated by adjoint variation method (AVM) applied in finite element method (FEM) solutions. A benchmark that optimizes the design of a wave impedance transformer compares Adam's performance with other commonly-used gradient-based and global algorithms. These concepts are then integrated into a representative example of optimization of a rectangular inset-fed patch antenna.