Lab · c20.s03.L1
Normalisation effect
Normalisation removes the size of the activations and keeps their shape.
Question: Why does rescaling the activations keep the next layer's input range fixed?
Frame 1 of 10: The four activationsFour activations, 1.00, 1.00, 3.00 and 3.00, enter the layer. Their mean is 2.00 and their standard deviation is 1.00. Normalisation turns them into -1.00, -1.00, 1.00 and 1.00; the learned scale 2 and shift 5 then give 3.00, 3.00, 7.00 and 7.00.
The four activations. Four activations, 1.00, 1.00, 3.00 and 3.00, enter the layer. Their mean is 2.00 and their standard deviation is 1.00. Normalisation turns them into -1.00, -1.00, 1.00 and 1.00; the learned scale 2 and shift 5 then give 3.00, 3.00, 7.00 and 7.00.
Keyboard: Space plays or pauses, Left and Right step, Home goes to the first frame, End goes to the last frame.
Text description of this lab
Two horizontal number lines, one above the other. Four filled dots enter on the top line. A spread control pulls them apart, from a tight cluster near 2 to a wide range from -8 to 12, while the bottom line keeps four rings at 3, 3, 7 and 7. The normalised values stay at -1, -1, 1 and 1, so the outgoing range stays 4.00 until normalisation is switched off.
Return to the section that hosts this lab
Caption status: schematic. Every lab declares one of published, re-simulated, schematic, synthetic or interpolated.