Simulates a forward pass through a feedforward neural network: signals originate at the input layer and propagate layer-by-layer along weighted connections, each arriving pulse triggering the receiving node to "activate" and glow before the signal continues onward. Nodes are rendered as emissive sphere meshes paired with additive-blended glow sprites, connections are a single THREE.LineSegments BufferGeometry per layer-gap, and the traveling signals themselves are pooled sprite objects lerped along each connection over time rather than fired as literal particle systems.