Deep Learning: A Recent Computing Platform for Multimedia Information Retrieval

Literature Review

Deep learning is defined as a technique that automatically detects significant patterns in the data. For deep learning, data is important, from which the learning algorithm discovers and learns the data properties. Both the quality and the quantity of the dataset tend to degrade the performance of prediction and learning. However, the techniques based on deep learning are proven to be promising for extracting learning patterns and features even if the data is complex. Deep learning can be considered as methods that are based on data-driven artificial intelligence and are used to frame the relationship between the input given and the output to be generated. Moreover, deep learning has unique features, like feature learning, model training, and model construction. Deep learning can also be considered as representation learning approaches, having several levels of representation that are attained through non-linear modules, where the representation at a level is transformed into a higher level or abstract level. Hence, it is possible to learn complex functions with the utilization of these transformations. Learning the representations of data via the abstract levels is the basic idea of DL. The major advantage of DL is its ability to extract the features directly from the original data such that the complexity of feature engineering can be avoided.

The two properties of DL are:

• •

  Numerous layers containing nonlinear processing units

• •

  Feature presentation on every layer throughthe learning carried out either in supervised or unsupervised manner