AI Cnns Tutorial: Filters, Pooling, Stride

TL;DRCNNs use filters, pooling, and stride to extract and downsample features.

Three knobs shape a CNN. Filters (kernels) are the learnable feature detectors. Stride is how far the filter jumps each step (bigger stride → smaller output). Pooling (max/average) downsamples feature maps, shrinking spatial size while keeping the strongest signals, which adds robustness and cuts compute. Together they progressively compress the image into rich, abstract features.

Key points

Filters: learnable feature detectors
Stride: filter step size (controls output size)
Pooling: downsample, keep strongest signals
Progressively compress into abstract features

Common mistakes

Confusing stride with pooling
Over-pooling and losing spatial detail
Too-large filters losing locality

Try it: Explain what max-pooling keeps and what it throws away.

Example code

<!doctype html><html><head><meta charset="utf-8"></head>
<body style="background:#06040d;color:#e6e0ff;font-family:monospace;padding:20px"><pre>filter → feature map
stride 2 → half-size output
max-pool 2×2 → keep strongest of each 2×2 block</pre></body></html>

Open the interactive lesson →

← Cnns · Convolutional Neural Networks Rnns · Recurrence · Vanishing Gradients →