Train a network to recognize handwritten digits and explore how different architectures learn
Draw a digit from 0-9 in the canvas below:
The processed 28×28 image appears in the bottom-right corner of the canvas.
This demo trains a neural network on the MNIST dataset, which contains 60,000 training images and 10,000 test images of handwritten digits.
You can adjust the network architecture by changing the size of the hidden layers. A larger network might learn more complex patterns, but could also overfit or take longer to train.
After training, you can draw your own digits and see how well the network recognizes them. The confidence bars show the network's estimated probability for each digit.