🌊 PESLA: Physics-informed Energy Self-supervised Landscape Analysis

This repository provides the official Python implementation of our ICLR 2025 manuscript: 📄 "Predicting the Energy Landscape of Stochastic Dynamical Systems via Physics-informed Self-supervised Learning".

PESLA introduces a novel framework for learning and predicting the energy landscape of stochastic dynamical systems, leveraging physics-informed self-supervised learning techniques.

🚀 Quick Start

📦 Required Dependencies

To get started, ensure the following dependencies are installed:

# platform: linux-64 (Ubuntu 11.4.0)
numpy==1.22.4
python==3.10.12
scikit-learn==1.3.0
scipy==1.11.2
torch==2.0.1+cu118
torchdiffeq==0.2.3
fast_pagerank==1.0.0

🏃 Running the Model

1️⃣ Prepare Datasets:

Download or simulate the datasets for the system of interest. Below are some examples:

• 2D Prinz Potential (Deeptime)
• Ecological Evolution (Defining Coarse-Grainability in a Model of Structured Microbial Ecosystems)
• Protein Folding (How Fast-Folding Proteins Fold)

2️⃣ Run the model:

• Specify the dataset by modifying the system field in config.yaml

• Execute the following command:

  python main.py

📁 Repository Structure

.
├── README.md
├── asset
│   └── image.png
├── config.yaml
├── data
│   ├── __init__.py
│   └── dataset.py
├── main.py
├── model
│   ├── __init__.py
│   ├── ae.py
│   ├── codebook.py
│   ├── dynamics.py
│   ├── model_4well.py
│   ├── model_homeodomain.py
│   └── model_sswm.py
└── utils.py

📜 Citation

@misc{li2025landscape,
      title={Predicting the Energy Landscape of Stochastic Dynamical System via Physics-informed Self-supervised Learning}, 
      author={Ruikun Li and Huandong Wang and Qingmin Liao and Yong Li},
      year={2025},
      eprint={2502.16828},
      archivePrefix={arXiv},
      primaryClass={cs.CE},
      url={https://arxiv.org/abs/2502.16828}, 
}

📝 License

Released under the MIT License.