Machine Learning in Elixir

Understanding the Need for Deep Learning

Imagine you’ve been asked to create an algorithm that automatically scrapes handwritten phone numbers from a field on a paper form. Your team has already automated the process of extracting individual digits from the form into a 28x28 grayscale image. How would you go about solving this problem using the approaches you’ve used in the previous chapters of this book?

Your initial thoughts might be that it’s easy enough to throw this problem at a logistic regression model or a decision tree. But what would you use for the feature space of your model? Would you treat each individual pixel value as a feature?

You could potentially divide the image into a grid, as shown in the following image.

From there, you ...