Explore how neural networks classify digits and images
Hover over pixels to see values
The network transforms pixel data through multiple layers
The 28ร28 pixel grid is converted into a single row of 784 numbers. Each number represents a pixel's brightness (0 = black, 1 = white).
Each input neuron connects to every hidden neuron. Each connection has a "weight" that determines its importance.
The sum passes through ReLU: if positive, keep it; if negative, make it zero. This adds "thinking ability" to the network.
Steps B and C repeat for each hidden layer. Each layer learns to detect more complex patterns.
Choose one of the 10 animals the model is trained to recognize
Watch the image flow through convolutional layers
A comprehensive visual guide to how CNNs process and understand images
A Convolutional Neural Network (CNN) is a specialized type of neural network designed specifically for processing visual data like images and videos.
Unlike regular neural networks that treat each pixel independently, CNNs understand spatial relationships between pixels, making them incredibly effective at recognizing patterns, objects, and features in images.
A small filter (kernel) slides across the image, computing dot products at each position to detect specific patterns.
Pooling reduces spatial dimensions while keeping the most important features, making the network more efficient and robust.
ReLU (Rectified Linear Unit) adds non-linearity by keeping positive values and zeroing out negatives.
Softmax converts raw scores into probabilities that sum to 1, giving us confidence levels for each class.
๐ก CNNs learn these filter values automatically during training!
Filters only look at small regions, detecting local patterns efficiently
Same filter applied everywhere, drastically reducing parameters
Simple features combine into complex concepts layer by layer
Can detect objects regardless of their position in the image