Neural Network Playground

What the neural network playground does

Train a small neural network right in your browser: add layers and neurons, set the learning rate, and watch the decision boundary bend to separate the data as it learns. It turns "the network learns" into something you can watch happen, epoch by epoch.

How a neural network learns

Each neuron computes a weighted sum of its inputs, passes it through an activation function, and hands the result to the next layer. Training compares the network's output to the correct answer, then uses gradient descent and backpropagation to nudge every weight in the direction that reduces the error. The learning rate controls how big those nudges are — too small and it crawls, too big and it overshoots.

Why it matters

Every modern AI system — image models, recommendation engines, and the large language models behind today's assistants — is built from these same components at massive scale. Building intuition on a tiny network is the fastest way to demystify the big ones. Next, see how models represent meaning in the embedding explorer.

Frequently asked questions

What is the learning rate?

The step size for weight updates. A good value trains quickly and stably; too high causes the loss to bounce or diverge, too low makes training painfully slow.

What is a decision boundary?

The line (or surface) the network draws to separate classes. Watching it curve as training proceeds shows the model fitting the data.

Why add more layers or neurons?

More capacity lets the network learn more complex boundaries — but too much invites overfitting, where it memorizes the training data instead of generalizing.