The MNIST Dataset

MNIST dataset of handwritten numbers

Recognizing human handwriting is an ideal challenge for testing artificial intelligence because the problem is sufficiently hard and fuzzy. It’s not clear and is defined like multiplying lots of lots of numbers.

Getting computers to correctly classify what an image contains, sometimes called the image recognition problem, has withstood decades of attack. Only recently has good progress been made, and methods like neural networks have been a crucial part of these leaps forward.

To give us a sense of how hard the problem of image recognition is, we sometimes disagree on what an image contains. For example, we’ll disagree about what a handwritten character actually is, particularly if the character was ...