A neural network with recursive expansion and drop path mechanism was constructed. Recursive Expansion used:
where DP is a drop path function that drops entire subpaths with probability p during training.
The CIFAR10 dataset was used. It has the classes: 'airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck'. The images in CIFAR-10 are of size 3x32x32, i.e. 3-channel color images of 32x32 pixels in size.
Accuracy of 35% was found on the testset which is better than chance baseline accuracy of 10% but worse than conventional convolutional network accuracy of 52%. Further study needs to be made to improve fractal architecture and drop path mechanism and tested with other datasets.