Generative Deep Neural Networks for Inverse Materials Design Using Backpropagation and Active Learning.

Area of Science:

• Materials Science
• Computational Chemistry
• Artificial Intelligence

Background:

• Machine learning (ML) shows promise for materials discovery and design.
• Current inverse design methods face bottlenecks due to inefficient exploration of the design space.

Purpose of the Study:

• To present a general-purpose inverse design approach using generative inverse design networks.
• To enhance material discovery by overcoming limitations of existing methods.

Main Methods:

• Utilized generative inverse design networks with backpropagation for gradient calculations.
• Implemented an active learning strategy to optimize candidate materials and minimize training data.
• Compared the approach with gradient-based topology optimization and genetic algorithms.

Main Results:

• The ML-based inverse design approach effectively overcomes local minima traps.
• Active learning reduced training data requirements by an order of magnitude in composite materials.
• Demonstrated superior performance and efficiency compared to conventional methods in case studies.

Conclusions:

• The proposed generative inverse design network offers a robust and efficient solution for discovering and designing novel materials.
• Active learning integration significantly enhances the performance and data efficiency of the inverse design process.
• This approach provides a valuable tool for accelerating materials innovation.