This module focuses on ready to use ML APIs and how to use them on unstructured data. When we say unstructured data, we are referring to the data that comes in the form of audio, video, images, freeform text, etc. In this module, we are going to start off by describing why unstructured data is so hard to work with. Then, we will discuss some of the products available on GCP for applying machine learning to unstructured data. As you can see on this slide, two images are shown. One, containing a newspaper page, and another containing a sporting event. There's a lot going on in these images, and presumably a lot of useful information can be extracted from them. For example, what language is the newspaper clip written in? What does the article say? In what year was it published? What sport is being played in the image on the right? What's the flag being waved? The important question is how can we extract this metadata? Do we have such technology? The answer is yes. And we are going to talk about it in this module. Let's give a few real-world example of businesses using unstructured data in their products. Can you differentiate between snow and cloud cover in these two images? One of GCP's customer, Airbus Defense and Space, works with satellite imagery such as this, and it's very important that they can detect and correct imperfections in the images, such as the presence of cloud formations. Historically, this imperfection correcting process was time-consuming, prone to error, and not scalable. Airubs solved these issues with machine learning. If you are stumped like I am, the clouds are in the upper-right part of the right image, highlighted in red. As another example, consider the realm of health and diagnostics. Diabetic retinopathy is a disease that can lead to irreversible blindness. Fortunately, the disease can be caught early and treated by inspecting retinal photographs like the ones here. Unfortunately, a specialist is required to inspect these photographs and make a diagnosis. Furthermore, such specialists are not very common in the part of the world where diabetes is most prevalent. Google worked with a team of specialists to build a deep learning algorithm to automate the process of diagnosing diabetic retinopathy where training on labeled images like the ones shown here. That's powerful. How can you map from images like this to a label using machine learning? Adding to the complexity, keep in mind that medical images are usually extremely high resolution, and so processing them takes a lot of compute. Google Cloud platform offers a number of products that can help businesses make sense of their unstructured data. These products, such as Cloud Vision API and Dialogflow, are based on both Google's data and models. You don't have to worry about training models with your data. You simply pass the products your data wire and API, and they will return predictions. It's really hard to train models on unstructured data. Consequently, developing something like vision recognition model is out of Reach for most businesses. On the downside, if your unstructured data is not within the scope of the data used to train Google's pre-trained models, the APIs won't give you good results. This is what we mean when we say common ML tasks.
