A neural network uses nodes and edges to simulate how a biological brain "learns". Using gradient descent and linear algebra, we can create highly accurate real-world models with enough training data. The secret to neural networks is backpropagation, which is a mechanism to distribute the training errors across the nodes in the network. Below you can train a 3-layer Neural Network to recognize digits. Select the size of your training set, the number of cycles, and the nodes in the hidden layer. You can even input your own handwritten digit using the canvas and by clicking "Predict Digit". Try it out!
Processed Drawing Shown here
Click and drag to draw a digit
Tip: Try drawing a "2" or a "5"
Tip: Try drawing a "2" or a "5"
Training Set:
100 Data Points
Cycles:
1 Epoch
Nodes:
100 Nodes
Tip: 6000 digits, 150 nodes, and 1 Epoch gives good results with not too much training time