LO 2.2.1.A

Learning objective: Describe the use of a training set as an alternative to a rules-based program to recognize digits.

Review:

• The difficulty of visual pattern recognition becomes apparent when trying to write a computer program to recognize digits. What seems easy when we do it ourselves becomes suddenly extremely difficult. Simple intuitions about how we recognize shapes turn out to be not so easy to express algorithmically. If you try to make such rules precise, you quickly lose yourself in countless exceptions and caveats, and special cases.
• Neural networks approach the problem differently. The idea is to take a large number of handwritten digits, known as training examples (see figure below), and then develop a system that can learn from these training examples. In other words, the neural network uses examples to automatically infer rules to recognize handwritten digits. We could build better handwriting recognition by using thousands, or even millions, or billions of training examples.

Source: Assigned reading