Pedagogical Framework & Technical Manual v0.10.0 Production

AI for VET: 9 Interactive ML Simulations Guide

Comprehensive guide for vocational teachers and students on utilizing interactive machine learning simulations in Moodle, visual intuition building, and autograded parameter challenges.

🤖 tiny_aisim
TinyMCE Editor Plugin

Embeds interactive simulation cards into any Moodle course page, book, or forum post with a single click. Configure initial datasets, hyperparameter presets, and display modes.

🔬 filter_aisim
HTML5 & Canvas Filter

Renders high-performance interactive widgets (Canvas, SVG & WebGL) with real-time parameter tuning, curve animations, loss curves, confusion matrices, and live metrics.

🎯 qtype_aisim
Question Type

Autograded quiz questions where students interact with live simulations to meet engineering performance criteria (e.g. achieve recall > 90% without false alarm penalty).

1. The 9 Core Machine Learning Simulations

Simulation Core Scientific Concepts Interactive Hyperparameters & Student Tasks
1. Gradient Descent Loss surface J(w), learning rate (α), momentum, convergence speed, divergence, local minima traps. Adjust α from 0.01 (sluggish) to 0.85 (divergence). Observe ball roll down convex vs non-convex 3D loss valleys.
2. Confusion Matrix & Threshold Classification probability P(y=1|x), decision threshold (θ), TP/FP/TN/FN trade-offs, Precision, Recall, F1-score. Drag decision threshold slider. Observe how lowering θ boosts Recall in medical/safety checks while increasing False Positives.
3. Polynomial Regression Model complexity, underfitting vs overfitting, bias-variance trade-off, L2 Ridge regularization (λ). Increase polynomial degree from 1 to 12. Observe wild oscillation on noisy points. Tune λ penalty to smooth the curve.
4. Non-linear Decision Boundaries k-Nearest Neighbors (k-NN), Decision Trees, tree depth, Voronoi decision boundaries, decision regions. Switch between k-NN (tune k from 1 to 15) and Decision Trees (depth 1 to 6). Observe how boundaries morph from smooth to sharp partitions.
5. K-Means Clustering Unsupervised learning, centroid recalculation, Euclidean distance assignment, Voronoi partitions, convergence. Place k cluster seeds. Click Step to watch centroids migrate to cluster centers. Test on non-spherical data to see limitations.
6. Perceptrons & Neural Networks Linear separability (XOR dilemma), multi-layer perceptron (MLP), hidden layers, ReLU and Sigmoid activations. Build a network to solve circular and XOR datasets. Add hidden layers and observe non-linear decision boundary deformation.
7. Convolutions & Image Filters Convolution operation, 3×3 kernel matrices, feature maps, edge detection (Sobel), blurring, sharpening. Select sample images (digits, industrial parts). Apply Edge, Gaussian Blur, and Sharpen filters. Inspect mathematical dot products.
8. Reinforcement Learning: Q-Learning States, Actions, Rewards, Bellman equation, Q-table updates, ε-greedy exploration vs exploitation, discount (γ). Train an agent in Maze, Cliff, and Treasure worlds. Watch arrows emerge in cells as the Q-table discovers optimal navigation policies.
9. Data Representation & Bias Training vs test split, subgroup representation disparity, algorithmic fairness, minority population under-representation. Simulate factory part inspection with 95% Factory A and 5% Factory B data. Discover why 95% aggregate accuracy hides 50% failure on Factory B!

2. Teacher Guide: Authoring & Assessment

Creating Quiz Questions with qtype_aisim:

In standard quizzes, questions test passive memory. With qtype_aisim, students are given a simulated engineering scenario and must find parameters that achieve specific performance criteria:

  • Simulation Selection: Choose any of the 9 ML simulations.
  • Preset Initial State: Provide starter data, initial noisy weights, or default hyperparameter settings.
  • Evaluation Targets: Define target thresholds that student adjustments must satisfy (e.g. Accuracy ≥ 92%, Iterations ≤ 25, Loss ≤ 0.05).
  • Tolerance Bands: Set allowable error margins for floating-point metrics.
  • Instant Feedback & Hints: Provide contextual guidance when student convergence fails (e.g. "Learning rate too high; point diverged into infinity").

3. Pedagogical Methodology for Vocational Education (VET)

The AI for VET curriculum prioritizes visual, tactile intuition over dense multi-variable calculus proofs. By interacting directly with loss curves, decision boundaries, and simulation dynamics, students develop concrete practical instincts for tuning machine learning models in modern industry and technical workflows.