Designing neurons from scratch without using any libraries

In this assignment, I implement the gradient descent and backpropagation mathematically from scratch, and implement it in Python code. These form as the building blocks of the abstract neural network layers, using which I explore how models learn patterns within data.

I also test the effects of varying the following structural and algorithmic hyperparameters on the learning process

• Loss / Cost Function
• Activation Function
• Optimization algorithm
• Learning Iterations / Epochs
• Learning Rate
• Regularization
• Train-Test Splits
• Number of hidden units
• Initialization

Link to Notebook