Make Your Own Neural Network in Python Course Syllabus

Overview: This hands-on course guides you through building a neural network from scratch using only Python and NumPy—no external machine learning libraries. You'll gain a deep understanding of the mathematical and computational principles behind neural networks by implementing each component yourself. With approximately 12 hours of content across six modules, you'll progress from foundational concepts to training a working digit classifier on the MNIST dataset. Each module combines theory with practical coding exercises to solidify your understanding of forward propagation, backpropagation, and optimization.

Module 1: Introduction to Neural Networks

Estimated time: 1.5 hours

• Biological vs. artificial neurons
• History and significance of neural networks
• Basic structure of neural network layers
• Visualizing data transformation through network layers

Module 2: Math Behind Neural Nets

Estimated time: 2 hours

• Matrix operations and dot product in neural computations
• Implementing the sigmoid activation function
• Understanding gradient descent for optimization
• Manual computation of forward and backward passes using NumPy

Module 3: Forward Propagation

Estimated time: 1.5 hours

• Transforming input through hidden layers to output
• Applying activation functions in layer transitions
• Coding a single-layer neural network using Python arrays

Module 4: Backpropagation and Weight Updates

Estimated time: 2.5 hours

• Defining loss functions for network training
• Applying the delta rule for weight adjustment
• Computing partial derivatives for gradient calculation
• Implementing learning rate in weight update steps

Module 5: MNIST Dataset Classification

Estimated time: 3 hours

• Preprocessing handwritten digit images from MNIST
• Feeding image data into a custom neural network
• Training and testing a digit recognizer from scratch

Module 6: Tuning and Optimization

Estimated time: 2 hours

• Adjusting hyperparameters like learning rate and hidden units
• Tracking model performance across epochs
• Understanding basics of overfitting and generalization

Prerequisites

• Basic knowledge of Python programming
• Familiarity with fundamental algebra and calculus concepts
• Experience using NumPy for numerical computations

What You'll Be Able to Do After

• Explain the mathematical foundation of neural networks
• Implement a fully functional neural network using only Python and NumPy
• Train a model to classify handwritten digits from MNIST
• Apply forward propagation and backpropagation algorithms manually
• Optimize neural networks by tuning hyperparameters and monitoring performance