Sliding RBF networks
Convolutional neural networks work by sliding a filter over an image and taking dot products. This will modify that idea, using radial basis functions instead of dot products.
Results
On the simple MNIST experiment, the network achieved a 97.63% success rate.