Probability-Density-Based Deep Learning Paradigm for the Fuzzy Design of Functional Metastructures

Area of Science:

• Quantum mechanics
• Metamaterials science
• Artificial intelligence

Background:

• The probabilistic nature of quantum mechanics, represented by the norm-squared wave function as probability density, underpins the behavior of microscopic systems.
• Hybrid neural network architectures have gained significant traction, leading to advanced intelligent systems with broad applications.
• Inverse design methods for functional metastructures often face challenges in exploring high-dimensional parameter spaces efficiently.

Purpose of the Study:

• To develop a novel deep learning paradigm based on probability density for the inverse design of functional metastructures.
• To create a method capable of efficiently evaluating and capturing a wide range of plausible metastructure designs within a high-dimensional parameter space.
• To demonstrate the effectiveness and adaptability of this probability-density-based approach in designing metastructures with targeted properties.

Main Methods:

• A probability-density-based deep learning network was developed to analyze the parameter space of metastructures.
• The method leverages local maxima in probability density distributions to identify promising metastructure candidates.
• The approach was applied to acoustic metastructure design, targeting specific transmission spectra.

Main Results:

• The proposed neural network efficiently evaluated and captured numerous plausible metastructures in a high-dimensional parameter space.
• Local maxima in the probability density distribution successfully identified candidates with desired performances.
• Experimental verification confirmed the effectiveness and generalization of the inverse design method in acoustics.

Conclusions:

• The probability-density-based deep learning paradigm offers an efficient and accurate approach for the inverse design of functional metastructures.
• This universally adaptive method can identify optimal designs by analyzing probability density distributions.
• The approach demonstrates significant potential for designing advanced metastructures across various scientific and engineering domains.