network architecture
loading...
edge width and opacity: relative RMS of full exported edge samples, not feature importance. moving particles illustrate connectivity only.
learnable Gaussian RBF edge functions
function details
live inference
node labels are actual activations; node size scales with |activation|. edge width and opacity scale with |full edge contribution| at the current input; blue is nonnegative and red is negative.
inputs
current input: [0]
output
0.000
target: 0.000
mean absolute actual activation
training progress
recorded training loss by epoch; validation loss is shown only when exported