Thoughtful Machine Learning

Chapter 7. Neural Networks

Neural Networks (or Nets) are effective at mapping observed data to a function. Researchers have been able to use Neural Networks for things like handwriting detection, computer vision, and speech recognition with breakthrough results.

Essentially, Neural Networks are an effective way of learning from data and have a long history dating back to the 1800s. In this chapter, we’re going to discuss how the Neural Networks algorithm came to be, what goes into it, and how it works, as well as a practical example of classifying languages based on character frequencies.

Note

The Neural Networks algorithm is excellent at function approximation and supervised learning problems. It has little restrictions on what it can do and has proven quite successful in practice. However, it is limited to operating on binary inputs and can present challenges from a complexity and speed standpoint.

History of Neural Networks

When introduced, Neural Networks were about studying how the brain operates. Neurons in our brains work together in a network to process and make sense of inputs and stimuli. Alexander Bain and William James both proposed that brains operated in a network that could process lots of information. This network of neurons has the ability to recognize patterns and learn from previous data. For instance, if a child is shown a picture of eight dogs, she starts to understand what a dog looks like.

This research was expanded to include a more artificial bent when ...