Scalar Accumulation
Key concept: What a single neuron computes: a weighted sum plus a bias, a = w₁x₁ + w₂x₂ + b.
This demo breaks one artificial neuron’s forward computation into its atomic pieces. Two sliders control the weights w₁, w₂ and one controls the bias b; a fixed input x = (2, −1) is shown alongside them.
Click Step through to watch the running total build up term by term — w₁x₁, then w₂x₂, then + b — with red bars pushing the total right (positive) and blue bars pushing it left (negative). The panel on the right shows the same computation geometrically, as the projection of w onto x.
What to look for: dragging any weight changes both a bar’s length in the accumulation view and the length/direction of w in the geometric view — the two panels are two readings of the exact same number.
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.