Tandem Generalized Variational Autoencoder Network for Multi-Solutions Inverse Design

Inspired by the variational autoencoder (VAE), we propose a novel multi-solutions strategy combined with a tandem neural network for inverse design and optimization problems. In the training, a latent vector <inline-formula><tex-math notation="LaTeX">z</tex-math></inli...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE antennas and wireless propagation letters S. 1 - 5
Hauptverfasser: Yuan, Tianguo, Li, Yang, Yang, Xiaolin
Format: Journal Article
Sprache:Englisch
Veröffentlicht: IEEE 2025
Schlagworte:
Antenna
Convergence
Decoding
Inverse design
multi-solutions
Neural networks
Optimization
Perturbation methods
Slot antennas
tandem neural network
Training
variational autoencoder
Vectors
ISSN:1536-1225, 1548-5757
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Inspired by the variational autoencoder (VAE), we propose a novel multi-solutions strategy combined with a tandem neural network for inverse design and optimization problems. In the training, a latent vector <inline-formula><tex-math notation="LaTeX">z</tex-math></inline-formula> is concatenated to perturb the <inline-formula><tex-math notation="LaTeX">S_{11}</tex-math></inline-formula> as input to the decoder. Within this framework, a multi-band slotted dipole antenna is employed as an example to validate the feasibility of the proposed network. Experimental results demonstrate that the proposed method not only breaks the one-to-one limitation between inputs and outputs in neural networks to generate diverse solutions, but also significantly accelerates model convergence through the introduction of a tandem architecture, compared with conventional generative adversarial approaches. The network potentially provides an efficient framework for solving inverse problems involving multi-solutions and optimization.
ISSN:1536-1225
1548-5757
DOI:10.1109/LAWP.2025.3626410