Summary and course review

Congratulations on completing the course.

Before we finish, this article offers a wrap-up of the course, and some ideas for your next steps.

In this course we have presented a first introduction to machine learning, particularly thinking about image-derived data. We have thought about two main ways we can derive data from images: first, from image pixels themselves, often needing the intermediate step of creation of features from the pixels; and second, from measures which have already been derived from the images, such as widths, area, etc. This second set of measures could be manually produced from images by experts, or measured by intermediate image analysis tools, before being used as input to machine learning algorithms.

A key step in machine learning (versus deep machine learning) is the creation or selection of features from the data. Image pixels are too numerous to be used directly most of the time. Features allow us to set up a much more information-rich, and more compact dataset derived from the millions of pixels in the image. This feature selection process is an important expertise to acquire when using machine learning.

Algorithm selection, as we have seen, is also important – picking the right algorithm for the data and task at hand. There are many different machine learning algorithms; here we can only scratch the surface. Building up knowledge of different algorithms and when they should be used is a key skill for a machine learning engineer.

We have hopefully presented an accessible background to machine learning as a discipline, and covered the key ideas behind just a few example algorithms. Those interested in a more mathematical or technical background to the techniques can find more information online. A good place to start is looking at the webpages for the Python libraries we have used in the practicals – for convenience we have included a few links below. These pages will tell you how to use the techniques in Python, and also give you some background on their operation and assumptions about data. A good way to learn more is simply to try out more example in Python.

For learning more about deep learning with images, specifically using models such as convolutional neural networks, we have a follow-on FutureLearn course in preparation which will be available soon. This will dive into more detail about deep machine learning.