Introduction to Deep Learning & Neural Networks

Course outline and learning expectations

This is a tutor-led course. The outline shows what your tutor will cover; teaching materials and examinations are provided directly to enrolled students.

What you will learn

• Explain the structure and function of artificial neurons and feedforward neural networks
• Select and apply common activation functions for neural-network tasks
• Describe forward propagation and backpropagation intuitively without heavy mathematics
• Build, compile, train and evaluate a neural network using TensorFlow and Keras
• Interpret loss functions, metrics, epochs, batches and learning curves
• Detect overfitting and apply dropout, regularisation and early stopping
• Compare deep learning with classical machine-learning approaches for tabular data
• Complete and communicate a responsible capstone model using anonymised or synthetic education data

Modules

1. 1

   Neurons and neural networks

   Understand artificial neurons, weights, biases, layers and network flow through visual examples

   Artificial neuronsWeights and biasesInput, hidden and output layersNetwork architecture
2. 2

   Activation functions and forward propagation

   Explore how activation functions transform signals and how a network produces predictions

   Linear combinationsReLUSigmoidSoftmaxForward pass
3. 3

   Backpropagation intuition

   Develop an intuitive understanding of how prediction errors guide weight updates without heavy mathematics

   Loss signalsGradient conceptChain-rule intuitionGradient descent
4. 4

   Build a first neural network with Keras and TensorFlow

   Create, compile, train and inspect a small Sequential neural network

   TensorFlow and Keras workflowDense layersModel compilationTrainingPrediction
5. 5

   Training, loss functions and epochs

   Train models and interpret learning behaviour using suitable losses, metrics and validation data

   Training and validation dataBinary and multiclass loss functionsOptimisersEpochs and batchesLearning curves
6. 6

   Overfitting and regularisation

   Recognise overfitting and apply practical techniques that improve generalisation

   Validation performanceDropoutL1 and L2 regularisationEarly stoppingData quality
7. 7

   Deep learning versus classical machine learning

   Decide when neural networks are appropriate and compare them with simpler models for tabular data

   Baseline modelsData size and complexityCompute requirementsInterpretabilityModel selection
8. 8

   Capstone: neural network for tabular education data

   Build and evaluate a simple model that estimates UNEB pass likelihood from anonymised or synthetic study and attendance data

   Data preparationTrain-validation-test splitModel buildingEvaluation metricsLimitations, privacy and responsible useProject presentation