to use a CNN: for data with spatial structure like images or time series

Why can't we just feed all data into one big neural net?

Imagine trying to understand a complex painting by looking at it from one spot at a time. You'd miss the details and the overall picture.

A CNN works like having multiple eyes scanning different parts of the painting at once, each focusing on different details. This way, you get a full understanding of the artwork.

Example

A CNN scans a 100x100 pixel image using 25 weights for a 5x5 tile, instead of 10,000 weights for a fully connected layer.

Remember this

A CNN efficiently processes spatial data by focusing on different parts simultaneously, avoiding the need for a massive number of connections.

Text adapted from Wikipedia, licensed under CC BY-SA 4.0.

Related concepts

GCN (Graph Convolutional Network) does: spectral convolution approximated by neighbor averaging

GNNs use pairwise message passing for node representation updates

to use an RNN/LSTM: for sequential data where order matters (mostly replaced by transformers)

Why do we remember stories better when they have a clear beginning, middle, and end?

message passing does in GNNs: each node aggregates features from its neighbors

Nodes in GNNs aggregate features from neighbors

data augmentation does for generalization: artificially expands training set

How can you teach a computer to see better?

Generative adversarial network

Ever wondered how computers can create new images that look real?

Machine learning

Can we teach computers to learn like humans?

Swipe through 100 ML concepts daily

Open Pocket Polymath