Neural Network

This simulation shows a real three-five-five-two neural network doing a forward pass. Each neuron adds its weighted inputs and a bias, then squashes the result with the tanh function. Students move three input sliders and see the signals flow through, with cyan connections for positive weights, orange for negative, and neuron color and size showing activation. A button randomizes the weights. The weights are not trained, so the focus is on how the arithmetic works, which makes it an approachable first look at how machine learning models compute.

• Move Input x1, x2 and x3 and watch Output y1 and Output y2 change.
• Look at the connection colors to find which weights are positive and which are negative.
• Press Randomise weights and see how the same inputs give different outputs.
• Set all three inputs to 0 and see what the outputs and Mean hidden activity do.
• Push the inputs to their limits and notice how the tanh squashing keeps values between minus 1 and 1.

1. 1Describe a neuron as a weighted sum plus a bias passed through an activation function.
2. 2Trace how information flows forward through the layers of a network.
3. 3Explain what positive and negative weights do to a signal.
4. 4Recognize that the outputs depend on both the inputs and the weights.

Questions teachers ask