Practical Challenges Training with Medical Data (AI For Medicine)

We'll talk about three key challenges for training algorithms on medical the class imbalance challenge, the multitask challenge, and the data set size challenge. For each challenge, we'll cover one or two techniques to tackle them. Let's start with the class imbalance challenge. So here's the challenge. There's not an equal number of examples of non disease and disease in medical data sets. This is a reflection reflection of the prevalence or the frequency of disease in the real world, where we see that there are a lot more examples of normals than of mass, especially if we're looking at X rays of a healthy population. In a medical data set, you might see 100 times as many normal examples as mass examples.