Creating Deep Convolutional Neural Networks for Image Classification

This lesson provides a beginner-friendly introduction to convolutional neural networks, which along with transformers, are frequently-used machine learning models for image classification. Neural networks develop their own idiosyncratic ways of seeing and often fail to separate features in the ways we might expect. Understanding how these networks operate provides us with a way to explore their limitations when programmed to identify images they have not been trained to recognize.

In this tutorial, we will train a convolutional neural network to classify paintings. As historians, we can use these models to analyze which topics recur most often over time, or automate the creation of metadata for a database. In contrast to other resources that focus on developing the most accurate model, the goal of this lesson is more modest. It is aimed at those wanting to gain an understanding of the basic terminology and makeup of neural networks so that they can expand their knowledge later, rather than those seeking to create production-level models from the outset.

Reviewed by:

• Fabian Offert
• Melvin Wevers

#Learning outcomes

After completing this lesson, you will be able to:

• Understand how convolutional neural networks function
• Train and use them to perform image classification on paintings
• Embed a model on a live website