The toy examples of KAN ( Kolmogorov-Arnold Network )

Support MLP, KAN, and KAN derivatives

KAN paper: Kolmogorov-Arnold Networks <img width="1163" alt="326219527-695adc2d-0d0b-4e4b-bcff-db2c8070f841" src="https://github.com/cheng-haha/KANs/assets/54107313/70baa18c-5289-48bc-bfd2-50fbff89ba15">

KANs:

• KANs
• Efficient-KAN
• FourierKAN
• Two-layer MLP (Toy Version)

Run

python examples/mnist.py --model MLP
python examples/mnist.py --model KAN
python examples/mnist.py --model MNISTFourierKAN

Experiential Settings

• For KAN, a large initial learning rate may be more effective. (You can try lr = 1e-2)
• In my experiments, KAN does converge faster than MLP.