Definition
CNNs use convolutional layers that share filters across spatial positions, capturing edges, textures, and objects efficiently. Pooling layers downsample; fully connected layers often finish classification.
They dominated computer vision before vision transformers became competitive.
In simple terms
A CNN is like scanning a photo with small magnifying glasses that each look for a specific pattern — stripes, corners, eyes — then combining those detections.
Where you see it
- Image classification and object detection.
- Medical X-ray and MRI analysis.
- Audio spectrogram models for speech tasks.
How it works
1.Convolve
Filters slide over the input producing feature maps.
2.Activate and pool
Nonlinearity plus downsampling.
3.Stack layers
Deeper layers see larger receptive fields.
4.Classify or detect
Head outputs labels or bounding boxes.
Why it matters
- CNNs remain efficient and strong for many vision and signal tasks.
Often confused
CNNs are obsolete because of transformers.
Transformers lead many benchmarks, but CNNs are still widely used for speed, size, and inductive bias.