This repository consists the code for the implementation of U-Net architecture but with Kolmogorov-Arnold Convolutions instead of regular convolutions.
KANs are promising alternatives of Multi-Layer Perceptrons (MLPs). KANs have strong mathematical foundations just like MLPs: MLPs are based on the universal approximation theorem, while KANs are based on Kolmogorov-Arnold representation theorem. KANs and MLPs are dual: KANs have activation functions on edges, while MLPs have activation functions on nodes. KAN seems to be more parameter efficient than MLPs, but each KAN Layer has more parameters than a MLP layer.
![mlp_kan_compare](https://private-user-images.githubusercontent.com/23551623/326219527-695adc2d-0d0b-4e4b-bcff-db2c8070f841.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.ZugH9uwmC42osaBvx-RbiF-Fkfvp7gmt2D-zOFv2ryU)
For more information about this novel architecture please visit:
- The official Pytorch implementation of the architecture: https://github.com/KindXiaoming/pykan
- The research paper: https://arxiv.org/abs/2404.19756
git clone https://github.com/JaouadT/KANU_Net.git
cd KANU_Net
pip install -r requirements.txt
python train.py --model KANU_Net --dataset BUSI --gpu 0
The following are the results after training KANU_Net and regular U-Net with the same experimental setup:
Model | Accuracy | Dice | IoU | Sensitivity | Precision | Specificity |
---|---|---|---|---|---|---|
KANU_Net | 97.85 | 73.83 | 60.08 | 70.19 | 97.88 | 97.85 |
U-Net | 95.23 | 74.95 | 62.17 | 69.44 | 84.33 | 98.64 |
This model is built upon U-Net and the authors of the implementation of the KA-Conv. We extend our gratitude to the creators of the original KAN for their pioneering work in the field.