“In an unsupervised model, the system typically has a large number of features, but the outputs are unknown or unavailable. However, gathering training (x,y) pairs can be costly. If only a small amount of x-y pairs exists, we move from supervised learning to semi-supervised learning. Without any x-y pairs, we need to rely on unsupervised learning,” he says. A good way to understand the two different models, according to Dr Lin, is to consider the way we learn language. As children, learning our first language, on a daily basis we are exposed to thousands of inputs, such as new words, idioms, imprecise sentences and other parts of our first language’s syntax without having a teacher to specifically teach us what is correct and what is not. This is a good example of unsupervised l

Join for free to read