High School Lab

Neural Network Lab

Train a small model to classify shapes using two visual features: curves and corners. Adjust the network, inspect its weights, compare train and test accuracy, and trigger an overfitting scenario on purpose.

What you will learn

  • How hidden units and learning rate shape training.
  • Why high training accuracy does not automatically mean good generalization.
  • How weights, loss, and decision boundaries change during learning.

Missions

  • Reach solid training accuracy in the standard lab.
  • Reach strong test accuracy with a small generalization gap.
  • Switch to the overfit demo and observe the gap between train and test.

Interactive model studio

Neural Network Lab

Mode: Standard Lab
Epoch 0
Train accuracy 0%
Test accuracy 0%
Gap 0%

Feature space

Decision regions and data points

Filled circles are training points. Hollow circles are test points.

Model view

Live weight map

Training trace

Loss over time

Interpretation

What the model is doing

Train the network to see how the decision boundary starts to separate the three shape classes.