Neural Network Playground

KiwiBee’s classroom edition retains the original Neural Network Playground training engine by Daniel Smilkov and Shan Carter. Guided experiments compare a circle dataset with two hidden layers, one neuron, or noisy data, and include an XOR dataset. A fixed random seed makes preset comparisons reproducible. Accessible controls and live epoch, training-loss and test-loss readouts support prediction and explanation. Original source is Apache-2.0; upstream analytics is removed. The small synthetic-data model does not represent a human brain or a large language model.

• Press Play and watch the epoch counter and loss values change.
• Compare one hidden layer with two hidden layers on the same dataset.
• Add noise or reduce the training-data share and compare training loss with test loss.

1. 1Describe how a decision boundary changes during learning.
2. 2Compare training performance with performance on held-out data.

Pertanyaan yang sering diajukan guru