HATNet: Hierarchical Attention Transformer Network

Official implementation of "HATNet: Hierarchical Attention Transformer Network for Material Synthesis Optimization", published in Advanced Engineering Informatics (Elsevier).

📄 Paper: https://www.sciencedirect.com/science/article/pii/S1474034625003556

---

Overview

HATNet is a deep learning framework specifically developed to optimize the synthesis of organic and inorganic materials, including molybdenum disulfide (MoS₂), and to estimate the photoluminescent quantum yield (PLQY). By leveraging the power of the multi-head attention (MHA) mechanism, HATNet captures complex dependencies within feature spaces, offering a significant advancement over traditional models like XGBoost and Support Vector Machines (SVMs). This unified framework, designed for classification and regression tasks, achieved state-of-the-art performance in material synthesis optimization for MoS₂ and PLQY.

---

Key Features

Feature	Description
🔄 Unified Framework	Combines classification and regression tasks using a shared attention-based architecture
🎯 State-of-the-Art Performance	Achieves 95% classification accuracy for MoS₂ synthesis and lower MSE values for PLQY estimation
🧠 Automated Feature Learning	Eliminates manual feature engineering by capturing intricate feature interactions
⚡ Multi-Head Attention	Leverages transformer-based attention mechanisms for complex dependency modeling

---

Installation

Prerequisites

• Python 3.8+
• PyTorch 2.0+
• CUDA (optional, for GPU acceleration)

Setup

# Clone the repository
git clone https://github.com/munsif200/HATNet.git
cd HATNet

# Install dependencies
pip install torch numpy pandas scikit-learn matplotlib seaborn scipy

---

Usage

MoS₂ Synthesis Classification

python Mos2_classification.py

PLQY Regression (Carbon Quantum Dots)

python CQDs_regression.py

---

Citation

If you find this work useful for your research, please cite our paper:

@article{hatnet2025,
  title={HATNet: Hierarchical Attention Transformer Network for Material Synthesis Optimization},
  journal={Advanced Engineering Informatics},
  year={2025},
  publisher={Elsevier},
  url={https://www.sciencedirect.com/science/article/pii/S1474034625003556}
}

---

Applications

HATNet addresses critical challenges in materials science, with applications in:

• 🔬 Advanced Material Synthesis: Optimization for electronics, optical, and other devices
• 📊 Quantum Yield Prediction: Estimation of photoluminescent properties
• 🧪 Process Optimization: Data-driven synthesis parameter tuning

---

Acknowledgments

This research was supported by the Nano & Material Technology Development Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT.

---

License

This project is released under the MIT License, ensuring accessibility to the research and development community.

---

Contact

For questions or collaborations, please email to munsif@sju.ac.kr.