Accuracy
I demonstrated that the FBM dropout achieves at least the same accuracy improvement as standard random dropout.
Convolutional Neural Network
with FBM Dropout Mechanism
I developed a CNN that classifies cats and dogs images to evaluate the effectiveness of a novel dropout mechanism based on fractional Brownian motion (FBM), which simulates the movement of stochastic axons in biological neural networks.
Visual Display Feature
I added a supplementary feature to visually display how the "axons" would extend under current FBM parameters.
Training and Testing Sets
I had 2,000 samples each training set and 500 samples in each testing set.