Track: Practitioner | Time: 12 hours | Prerequisites: Chapters 1, 3, 6
By the end of this chapter, you will be able to:
- Understand neural network architecture: neurons, layers, activations
- Implement a neural network from scratch using NumPy (forward pass, backpropagation)
- Train networks with gradient descent, mini-batch SGD, and momentum
- Apply regularization techniques: dropout, L2, early stopping, batch normalization
- Build and train models with PyTorch (nn.Module, DataLoader, optimizers)
- Understand and implement CNNs for image classification
- Understand and implement RNNs and LSTMs for sequence data
- Evaluate deep learning models and diagnose underfitting/overfitting
- Build a complete image classification pipeline end-to-end
chapter-09-deep-learning-fundamentals/
├── README.md
├── requirements.txt
├── notebooks/
│ ├── 01_introduction.ipynb # Neurons, forward pass, backpropagation from scratch
│ ├── 02_intermediate.ipynb # PyTorch basics, training loops, regularization
│ └── 03_advanced.ipynb # CNNs, RNNs, image classification capstone
├── scripts/
│ ├── deep_learning_toolkit.py # NeuralNetScratch, training utilities, plotting
│ └── utilities.py # Data loading helpers, synthetic data generators
├── exercises/
│ ├── exercises.py # 5 exercises
│ └── solutions/
│ └── solutions.py # Complete solutions
├── assets/diagrams/
│ ├── neural_network.svg # Feedforward network architecture
│ ├── backpropagation.svg # Computational graph and gradient flow
│ └── cnn_architecture.svg # CNN layers: conv, pool, fully-connected
├── datasets/
│ └── spirals.csv # Synthetic spiral classification data (500+ rows)
| Section | Time |
|---|---|
| Notebook 01: Introduction (Neural Networks from Scratch) | 4 hours |
| Notebook 02: Intermediate (PyTorch & Regularization) | 4 hours |
| Notebook 03: Advanced (CNNs, RNNs & Capstone) | 4 hours |
| Exercises | Included in notebooks |
| Total | 12 hours |
Generated by Berta AI | Created by Luigi Pascal Rondanini