LeNet Convolutional Network with KAN as classifier
This python script is derived from the efficient_kan MNIST example.
The KAN classifier receives the flatten data from convolution layers, the accuracy is around 98.7 %.
At the end of the script, it will convert the model in ONNX.
- You need to download efficient_kan and put it in the same folder of the script
- You need PyTorch CUDA for training with GPU
I'm using HandOCRPaint as handwritten OCR tester.
- Training on the entire alphabet
- Training on different fonts
- Benchmark with MLP equivalent