This is the same feedforward network as the 3D version, drawn flat instead of in a rotating 3D scene. Each patient signal enters the input layer; every hidden layer computes a weighted sum of the previous layer's activations plus a bias, then squashes it with a sigmoid; the output layer turns its raw scores into probabilities with softmax.
a(l) = σ(W(l)·a(l-1) + b(l))
P = softmax(z_out)
- Wiring diagram (left pane) — node brightness is the activation strength; only connections whose |weight|×activation clears a strength threshold are drawn, so the diagram thins out as signals fade.
- Probability bars (top right) — the three softmax outputs (Healthy / At-risk / Disease) as live bars, with a marker showing where the decision threshold sits.
- Confidence history (bottom right) — a scrolling strip of the winning class's probability over the last few seconds, so you can see how a slider drag reshapes the decision, not just its current value.
- Decision threshold — the probability the At-risk/Disease class needs to clear before it becomes the predicted diagnosis; raising it makes the network more conservative.
Same three inputs, same two-hidden-layer topology, same fixed random weights (seeded, so both the 2D and 3D versions agree exactly) as the 3D companion — this view is built for reading the forward pass rather than orbiting a scene.