View source for 10.4.4 Convolutional Neural Networks (CNNs)

=== 10.4.4 Convolutional Neural Networks (CNNs) ===
Imagine you're looking for your friend in a crowded school hallway. You're not just looking at the whole crowd at once; you're probably scanning for their face, their hair color, or the specific backpack they carry. You're looking for ''patterns'' or ''features'' in different parts of the scene.

'''Convolutional Neural Networks (CNNs)''' are a special type of digital "brain" that are really, really good at doing something similar, but with images! They are mostly used for tasks like:

* '''Recognizing objects in pictures:''' Is this a car, a bicycle, or a tree?
* '''Finding faces:''' Where are the faces in this photo?
* '''Understanding what's in a video:''' What's happening in this movie scene?

Here's a super simplified way a CNN works:

# '''Scanning for Features (The "Convolution" Part):''' Instead of looking at an entire image at once, a CNN uses special "filters" (think of them like tiny magnifying glasses looking for specific things). These filters slide over small parts of the image, one after another. One filter might be looking for straight lines, another for curves, and another for corners. When a filter finds what it's looking for (like a sharp edge), it "activates" or sends a strong signal.
# '''Building Up Complexity:''' After finding simple things like lines and curves, the CNN then puts these simple "features" together to find more complex ones. For example, lines and curves put together might form an eye. Eyes and noses put together might form a face!
# '''Making a Decision:''' Finally, after it has found all these different features at different levels (from simple lines to whole faces), it uses all that information to make a decision about what's in the picture. "Aha! I see two eyes, a nose, and ears. This must be a dog!"

So, CNNs are like master detectives for images, breaking them down into tiny clues and then putting those clues together to understand the bigger picture.