← ~/visualizations

neural-network

A 3-5-5-2 layer network with flowing pulses through bezier-curved connections. Nodes activate in sequence as data propagates forward.

canvasclick to interact
t=0s

practical uses

  • 01.Teaching neural network architecture to non-technical stakeholders
  • 02.Visualizing forward propagation in ML presentations
  • 03.Demonstrating layer-by-layer activation patterns
  • 04.Creating intuitive mental models of how signals flow through networks

technical notes

Uses cubic bezier curves for connections. Pulse position calculated via parametric cubic interpolation. Layer activation synchronized to global animation phase.

gradient-descent →