AI Simulations Viewer
Thirteen interactive machine learning simulations that run in the browser inside any Moodle text.
Live demo: try the plugins on aisim.tsp.edu.rs. Log in as a guest, or use the demo accounts teacher and student, password Demo-2026. Everything is reset every night. The simulations alone, without Moodle: sim.aiforvet.eu.
What's new in 1.0.0
- Stable release: thirteen simulations, all texts translatable, all layouts in Mustache templates.
- Four new simulations: logistic regression, cross-validation, data preparation and the recurrent neural network.
- Teacher guide 1.5: all thirteen modules, in English, Serbian, Slovak and Portuguese.
The simulations
- Gradient descent: the learning rate and the starting weight on a convex, non-convex or plateau cost curve; convergence, oscillation and divergence.
- Confusion matrix: move the decision threshold and watch precision, recall, F1 and the ROC curve in three scenarios.
- Polynomial regression: the degree and the regularisation, training and test error, underfitting and overfitting.
- k-NN and decision tree: add points and compare the decision boundaries for k and the tree depth.
- K-means: assignment and update steps, the number of clusters and the elbow chart.
- Perceptron and neural network: hidden layers, neurons, activation and learning rate; train and watch the boundary, the network and the loss curve.
- Convolution: a 3x3 kernel on small images, presets such as blur, sharpen and edges, the calculation of a pixel, and a target filter to find.
- Q-learning: an agent on a grid map, the learning rate, discount and exploration, the Q-values of each cell and the reward curve.
- Recurrent neural network: the memory, input and output weights of one recurrent neuron, the network unrolled in time, forgetting and the vanishing gradient on a noisy thermometer and a memory task.
- Data split and bias: how a biased sample and a small test set change the accuracy for two groups.
- Logistic regression: weights, bias and a decision threshold on two input variables; the straight decision boundary, the sigmoid, precision and recall, and a ring that no straight line can separate.
- Data preparation: statistics and histograms of raw data, missing values, a unit mistake and scaling, and their effect on a k-NN model.
- Cross-validation: the degree of a polynomial as a hyperparameter, k folds, the U-curve of the training and validation error, and a test set used only once.
How it works
- In any text: pages, books, lessons, labels, forum posts and quiz questions; a simulation is an element with a
data-aisim attribute, inserted most easily with the TinyMCE button.
- In the browser: everything runs on the student's device with HTML5 canvas; no server load, no external services, works on phones.
- Explains itself: every simulation shows its numbers and gives advice about what the student sees.
- Translatable: all texts are language strings and can be translated in AMOS.
The AI simulation plugins
Requirements
- Moodle: 4.5 LTS to 5.3.
- Setup: enable the filter in Site administration, Plugins, Filters, Manage filters.
Privacy
The plugin stores no personal data and sends nothing to external services; it implements the Moodle Privacy API.
About
Developed in the Erasmus+ project AI for VET for vocational schools. A teacher guide in English, Serbian, Slovak and Portuguese, the source code, issues and the changelog are in the GitHub repository.
