Quantum and complex-valued hybrid networks for multi-principal element alloys phase prediction

Area of Science:

• Materials Science
• Quantum Computing
• Artificial Intelligence

Background:

• Traditional phase classification relies on extensive feature engineering.
• Quantum and complex-valued neural networks offer potential for improved data processing.
• Integrating these approaches can enhance information dimensionality and reduce data loss.

Purpose of the Study:

• To develop a hybrid network model for material phase classification using elemental composition.
• To leverage quantum networks and complex-valued neural networks to eliminate feature engineering.
• To improve classification accuracy and robustness compared to existing methods.

Main Methods:

• A hybrid model combining parameterized quantum networks and complex-valued neural networks was developed.
• Elemental composition was used as the sole input, converting real-valued data to complex domains.
• Quantum networks managed sparse data and increased dimensionality, while complex-valued networks processed complex-domain data.

Main Results:

• The hybrid model achieved a phase classification accuracy of 94.93%.
• This accuracy surpassed the best machine learning model by 2.27% and the quantum model by 8.67%.
• Excellent precision (0.9494), recall (0.9493), and F1-score (0.9500) were recorded.

Conclusions:

• The hybrid network model offers a highly accurate and efficient method for material phase classification.
• The model demonstrates robust generalization capabilities, validated by phase transition analysis in alloys.
• This approach minimizes information loss and eliminates the need for complex feature engineering.