Scalar Activation
Key concept: A neuron's output is not the raw sum z — it is a = φ(z), the sum passed through a nonlinear activation φ.
This demo extends the scalar-accumulation walk (Demo 1) with one more row: the pre-activation z = w₁x₁ + w₂x₂ + b is passed through a switchable activation function φ to produce the neuron’s actual output a = φ(z).
A selector in the header lets you flip between six activations (e.g. identity, sigmoid, tanh, ReLU, and others); the activation-map panel plots φ as a curve and marks exactly where the current z sits on it, while a grey hairline shows the displacement φ produces — how far a ends up from z.
What to look for: drag the weight sliders to change z, then switch activations to see how the same z can map to very different outputs a — this is the difference between a linear unit and a nonlinear one.
Interaction is easiest on a laptop/desktop screen. On a small screen, you can also open the demo in its own tab at full size.