NeuroNomixer logoNeuroNomixer
  • Home
  • Blog
  • Visual Guides
  • Authors
  • Contact
Sign InSign Up
NeuroNomixer logo
HomeBlogAuthorsContactPrivacy Policy

© 2026 NeuroNomixer — Built with Next.js & Tailwind CSS

Visual Guides/Pooling Layers: Shrinking Without Losing
Deep Learning

Pooling Layers: Shrinking Without Losing

Watch max, average, min, and RMS pooling slide across a pixel grid in real time. Understand how CNNs reduce spatial dimensions while preserving the features that matter.

Exploration Progress1/3 pooling types explored  · 0/1 comparison viewed

Sign in to save progress

1

Choose Pooling Type

1/4 explored

Max Pooling: Takes the strongest signal

Picks the largest value in each window. Great for preserving the most prominent feature, such as the brightest edge or strongest activation.

2

Configure Window

Kernel Size
Stride
Output size:⌊(8 − 2) / 2⌋ + 1 = 4(8×8 input → 4×4 output)
3

Watch the Pooling

Input (8×8)
120
80
200
150
30
180
90
210
60
240
70
110
220
40
160
50
190
30
140
80
170
250
20
130
100
160
50
230
90
60
200
70
40
200
180
20
140
100
50
220
230
70
110
190
60
180
140
30
80
150
60
100
210
40
170
90
170
20
220
50
80
160
30
240
→↓
Window [0,0]
120
80
60
240

max([120, 80, 60, 240]) = 240

Output:240
→↓
Output (4×4)
240
Step 1 / 16
6%
4

Max vs Average: Side by Side

Same 8×8 input, same 2×2 kernel, stride 2 → 4×4 output. Max Pooling retains sharp features (higher peak values); Average Pooling is smoother (values blend together).

Max Pooling2×2, stride 2
240
200
220
210
190
230
250
200
230
190
180
220
170
220
210
240

Preserves strongest activations. Sharper contrast.

vs
Average Pooling2×2, stride 2
125
133
118
128
120
125
143
105
135
125
120
110
105
108
123
133

Blends values together. Smoother, lower peaks.

Max Pooling range

Min: 170Max: 250Mean: 213

Average Pooling range

Min: 105Max: 143Mean: 122
5

Why Pooling Matters

Why pooling is used in CNNs

Spatial Invariance

Small shifts or translations of an object in the input don't significantly change the pooled output. This helps CNNs recognise the same feature regardless of its exact position.

Dimensionality Reduction

Pooling shrinks the spatial size of feature maps, reducing the number of parameters and computation in subsequent layers. A 2x2 max pool with stride 2 cuts width and height by half.

Noise Suppression

Max pooling naturally ignores low activations by taking only the peak value in each region. This helps the network focus on strong, meaningful signals and ignore background noise.

Key Insight

In modern CNNs like ResNet, Global Average Pooling collapses each feature map to a single value per channel, replacing the huge flattened fully-connected stacks of older architectures like VGG; only one small final classifier layer remains. This dramatically reduces parameters and overfitting.

← CNNs: See What Filters SeeNext Guide →