⚡ Dropout Visualizer
Watch neurons spark, then randomly die — and see dropout vanish completely in eval mode.
📋 How To Use This Lab
Set
Dropout Probability (p)
— that's each neuron's chance of dying on any given pass.
Press the big
▶ Run Forward Pass
button
→
every hidden neuron sparks, then survives or dies.
Watch survivors get
scaled up
(divided by 1−p) to keep the layer's total output balanced.
Flip to
Eval Mode
→
press it again — nothing dies anymore, no matter what p is set to.
p = 0.5
TRAIN MODE
Input Layer
Hidden Layer — Dropout Applied Here
Output Layer
▶ Run Forward Pass
◀ Previous Pass
Next Pass ▶
Auto Run (5 passes) ▶▶
↺ Reset
Train Mode
Eval Mode
⚡ SPARKY
Press ▶ Run Forward Pass — I'll spark every hidden neuron, then randomly decide who survives.
🧑🏫 PROF. TORCH
Dropout is a TRAINING-only trick. Call model.eval() and every neuron survives, every single time.
🎛️ Dropout Probability
p =
0.5
— chance each hidden neuron is dropped (train mode only)