SNN - Simple Neural Network

Pure Python (+ numpy) implementation of chosen neural network components.
PyTorch only for MNIST dataset.

Components:

• Linear layer
• ReLU
• Dropout (implemented as inverted dropout)
• CrossEntropy loss
• Mini-batch SGD
• Xavier uniform initialization
• more to come :)

Results:

MNIST results (accuracy):

• train - 96.54 %
• test - 96.32 %