NCWNO

Neural Combinatorial Wavelet Neural Operator for catastrophic forgetting free in-context operator learning of multiple partial differential equations

What are we trying to do?

NCWNO architecture in a glimpse.

Simultaneously Predicting solutions of multiple PDEs using a Pre-Trained NCWNO.

1D problems:

2D problems:

File description

📂 1d continual learning          # Contains files of the 1d continual learning.
  |_📂 data                       # Folder for storing DATA and generating data.
    |_📁 model                    # Folder for storing trained models.
      |_📄 pre-trained model               # location of model.
    |_📄 data file                         # location of data.
    |_📄 ...                     
  |_📁 results                               # location of results.
📂 1d multiple learning           # Contains files of the 1d multiple physics training.
  |_📂 data                       
    |_📁 model                    
  |_📁 results                    
📂 2d continual learning          # Contains files of the 2d continual learning.
  |_📂 data                       
    |_📁 model                    
  |_📁 results                    
📂 2d multiple learning           # Contains files of the 2d multiple physics training.
  |_📂 data                       
    |_📁 model                    
  |_📁 results                    

Essential Python Libraries

Following packages are required to be installed to run the above codes:

• Numpy
• Matplotlib
• PyTorch
• PyWavelets - Wavelet Transforms in Python
• Wavelet Transforms in Pytorch
• Wavelet Transform Toolbox

Dataset

• The testing datasets are available at the following link:

  Dataset The datasets must be placed inside the corresponding data folder

Cite us at

@article{tripura2025neural,
  title={Neural Combinatorial Wavelet Neural Operator for catastrophic forgetting free in-context operator learning of multiple partial differential equations},
  author={Tripura, Tapas and Chakraborty, Souvik},
  journal={Computer Physics Communications},
  pages={109882},
  year={2025},
  publisher={Elsevier}
}

@article{tripura2023foundational,
  title={A foundational neural operator that continuously learns without forgetting},
  author={Tripura, Tapas and Chakraborty, Souvik},
  journal={arXiv preprint arXiv:2310.18885},
  year={2023}
}