Skip to content

Day 04 — Part 1: Intro to Deep Learning

Deep learning is the engine behind image recognition, language translation, and speech synthesis — but it is also one of the most misapplied tools in the field. Today you will build a mental model of how neural networks actually work, train one from scratch in Keras, and develop the practitioner instinct to know when deep learning is the right choice and when it is overkill.

Session Overview

Difficulty: Intermediate–Advanced Reading time: ~3 hours | Exercises: ~2 hours Prerequisites: Linear algebra (matrix multiplication) · Gradient descent concepts · Classification and regression fundamentals · NumPy vectorised operations

Learning Objectives

By the end of this session you will be able to:

  • Explain what a neuron computes and why activation functions are essential
  • Describe the forward pass, loss computation, and backpropagation in plain language
  • Build, compile, train, and evaluate a neural network using Keras
  • Diagnose overfitting from a training curve and apply at least two countermeasures
  • Decide when deep learning is worth its added complexity vs. when gradient boosting is the better call

Session Map

# Topic File Estimated Time
1 Neural Network Intuition 01-neural-network-intuition 25 min
2 Training Neural Networks 02-training-neural-networks 30 min
3 Keras Quickstart 03-keras-quickstart 30 min
4 Overfitting and Regularization 04-overfitting-regularization 25 min
5 Exercises 05-exercises 45 min

Total estimated time: ~2.5 hours

Prerequisites

Before this session, you should be comfortable with:

Info

This session uses TensorFlow 2.x with the Keras API. Install with pip install tensorflow. GPU is not required — all examples run on CPU in minutes.

Tip

Work through the files in order. The exercises in 05-exercises.md reference concepts from all four content files.

Start with 01-neural-network-intuition.