CIVE70111 Machine Learning

Explore how linear regression finds the best line through data points.

Step through gradient descent one iteration at a time.

Interactive demonstration of multi-linear regression with real-time feature selection.

Visualize the bias-variance tradeoff using real bike rental data. Adjust polynomial degree to see underfitting, optimal fit, and overfitting.

Explore how higher model capacity can cause overfitting in polynomial regression and the effects of L1 and L2 regularisation to mitigate this.

Understand how L1 (Lasso) and L2 (Ridge) regularisation prevent overfitting by penalizing model complexity. Interactive visualization of regularisation effects on polynomial regression.

Visualize how L1 and L2 regularization geometrically reshape the loss surface and affect gradient descent trajectories. Watch optimization paths converge to different solutions based on penalty strength.

The first neural network algorithm: step through perceptron learning epoch-by-epoch, watching the decision boundary adapt to separate two classes.

Interactive logistic regression with adjustable decision boundaries. See how the sigmoid function creates smooth classification boundaries.

Multi-class classification with various boundary types (linear, polynomial, radial basis). Experiment with different decision boundary shapes.

Adjust decision thresholds and explore sensitivity, specificity, precision, accuracy, F1-score, and balanced accuracy. See why accuracy fails with imbalanced data.

Build and train neural networks with adjustable architecture and parameters. Experiment with different layer sizes and activation functions.

Step-by-step visualization of the backpropagation algorithm. Watch how gradients flow backward through the network layers.

Train neural networks on real civil engineering datasets. Configure architecture, compare with linear regression, and make predictions in real-world units.

Interactive exploration of CNN operations with real-time filter visualization. See how convolution and pooling operations transform images.

Complete MNIST digit classification with real PyTorch backend and epoch-by-epoch training visualization.

Explore how text can be converted into numerical representations using the bag-of-words approach. See how documents become vectors based on word frequencies.

Discover how RNNs process sequential data step-by-step. Visualize hidden state evolution and understand the core mechanics of recurrent processing.

Explore Long Short-Term Memory networks with sophisticated gating mechanisms. See how forget, input, cell, and output gates work together to maintain long-term dependencies.

Explore how attention mechanisms work through Q, K, V computations and dot-product attention. Visualize how tokens attend to each other in sequence processing.

Interactive clustering demonstration with step-by-step k-means algorithm visualization. Add your own points and watch how centroids move to minimize within-cluster distances.

Explore agglomerative clustering with interactive dendrogram visualization. Compare different linkage criteria and discover nested cluster structures.

Investigate PCA on real civil engineering scenarios. Toggle standardisation, inspect explained variance, and compare feature loadings with saved snapshots.

Explore the exploration vs exploitation tradeoff with interactive epsilon-greedy algorithm. Watch an agent learn to balance trying new actions with exploiting known good ones.

Navigate a 5×5 grid world from start to goal while avoiding obstacles. Learn the fundamentals of state spaces, action spaces, and sequential decision making.

Watch Policy Iteration find the optimal policy through alternating evaluation and improvement. Visualize value functions and policies converging to the optimal solution.

Manual control of CartPole environment. Experience continuous state spaces and unstable dynamics before learning RL algorithms.

Compare three model-free RL algorithms side-by-side on CartPole. Watch Monte Carlo, SARSA, and Q-learning learn through 4 training checkpoints with Q-table visualizations.