In data mining applications, the lack of labeled data makes supervised learning algorithms fail to build accurate classification models. Transfer learning has been developed to deal with such lack of label problem. It aims to improve the performance of learning by transferring knowledge from several source domains to a target domain. For example, image classification can be modeled as a target learning task where there are only a few labeled training images. Fortunately, it is possible to collect some texts related to images, such as image annotations or documents around images, so that the knowledge from text data (a source domain) can be transferred to classify images in a target domain.
In real-world applications, examples are described by different feature sets or different “views” due to the innate properties, or collecting from different sources. For instance, in multimedia content understanding, the multimedia segments can be simultaneously described by their video signals from visual camera and audio signals from voice recorder devices. The different views usually contain complementary information, and multi-view learning can exploit this information to learn representation that is more expressive than that of single-view learning methods.
The explosive growth of online content such as images and videos nowadays has made developing classification system a very challenging problem. Such new classification system is usually required to assign multiple labels to one single instance: an image might be annotated by many semantic tags in image classification; one article can focus on several topics for text mining. Most of the conventional classification techniques under the assumption that an object only refers to one single class fail to work in such scenario. Therefore, methods that are capable of accomplishing multi-label learning can be more and more important.
Deep learning is part of a broader family of machine learning methods based on learning data representations, which is known as deep structured learning or hierarchical learning. Deep learning architectures have been applied to fields including computer vision, speech recognition, natural language processing, etc, where they have produced results comparable to and in some cases superior to human experts.