📊 Batch Normalization Visualizer
Scattered data snaps to zero mean, unit variance — then Gamma and Beta reshape it live.
📋 How To Use This Lab
Look at the raw batch — scattered points with a real (non-zero) mean and spread.
Press
▶ Normalize Batch
→
watch every point slide to zero mean, unit variance.
Drag
Gamma
(spread) and
Beta
(shift)
→
the normalized cluster rescales live.
Change
Batch Size
→
a fresh random batch appears, un-normalized again.
mean = —
std = —
RAW (not normalized)
Raw Batch (unnormalized)
Normalized Batch (mean=0, std=1) × Gamma + Beta
Press ▶ Normalize Batch to begin
▶ Normalize Batch
◀ Previous Preset
Next Preset ▶
Auto Run ▶▶
↺ Reset
📊 NORMA
This batch is scattered with a real mean and spread. Press ▶ Normalize Batch and watch it snap into place.
🧑🏫 PROF. TORCH
BatchNorm computes mean and variance ACROSS the batch, then rescales every point the same way.
🎛️ Batch Norm Parameters
Batch Size:
8
Gamma (Scale):
1.0
Beta (Shift):
0.0