🧠 Neural Network Explorer

Train a real neural network in your browser — watch it learn to separate data point by point.

Dataset

Hidden Neurons

Learning Rate

0.10

Activation

Training

Epoch: 0
Loss: —
Accuracy: —

DATA POINTS

DECISION BOUNDARY

NETWORK WEIGHTS (line thickness = weight magnitude, red = negative)

TRAINING LOSS OVER TIME

What is a neural network?

A neural network is a set of numbers (weights) connected in layers. During training, each weight adjusts slightly to reduce the gap between the network's prediction and the correct answer. Repeat thousands of times and the network "learns" the pattern.

How does it learn?

Backpropagation computes how much each weight contributed to the error, then nudges it in the direction that reduces the loss. The learning rate controls how big each nudge is — too high and it overshoots, too low and it's slow.

Key terms

Neuron — computes a weighted sum of inputs, then applies an activation function Weight — a number that scales a connection; adjusted during training Bias — an extra weight that shifts the activation threshold Activation — non-linear function (tanh, ReLU, sigmoid) that lets the network learn complex shapes Loss — measures how wrong the network is; goal is to minimize it Backpropagation — calculates the gradient of loss w.r.t. each weight Learning rate — step size for weight updates; too large = unstable, too small = slow Epoch — one full pass through the training data

🎯 Try this challenge

Set the network to 2 hidden neurons and train on the XOR dataset. Can it solve XOR? Now try 4 neurons — what changes? Why does XOR require at least one hidden layer?

Continue Learning

AI for Students