kan-rbf

This repository is a straightforward port of FastKAN, replacing the B-spline in KAN with a Radial Basis Function (RBF). The implementation is written in Flax NNX.

For comparison, efficient-kan is also included, providing an efficient implementation for benchmarking against RBF-based approach.

Key Files:

• Core Implementation: kanrbf.py
• Test for Correctness: test_kanrbf.py
• Latency Benchmark: test_running_time.ipynb
• MNIST Training Example: train_mnist.py

MNIST Training Performance:

If you find any bugs or issues, please report them.