How Different is Deep Learning from Other Machine Learning Models?

Deep Learning is one of the most successful areas of Machine Learning. It has now been applied to problems of very different types. It has also reached levels of accuracy in some of these problems that no other Machine Learning model has ever reached. But what is Deep Learning? Are there more than one type of Deep Learning model? How does Deep Learning differ from other areas of Machine Learning? In this post we answer these questions and more.

Deep Learning, as it is done nowadays, has really nothing to do with the actual simulation of the human neural network. Deep Learning is about getting data through different filters, the layers, to bring these data to the desired target representation. The Deep in Deep Learning is that there may exist many of these layers in a Deep Learning model. The Learning in Deep Learning is that these layers learn how to get to the desired target data representation. The first layer is the input layer, the last layer is the output layer and all the other layers between these two are called hidden layers.

Supervised learning represents the great majority of cases where Deep Learning is used. In this type of learning there is an input with variables that are properly called predictors or features and there is a target. The goal is to learn the relationship between the predictors and the target. Deep Learning achieves this by successive changes, through its layers, in the representation of the input data until the final, output representation is equal to the target values. These layers have parameters that are adjusted to make these successive data representations. This adjustment is the learning part.

Another type of learning is unsupervised learning where, contrary to supervised learning, there is no target value previously given to be associated with the features of the input data. In unsupervised learning the goal is different, for example to reduce the dimensions of the data towards data visualization or better understanding or to reveal hidden clusters. Note that Deep Learning is not a type of learning, as Supervised and Unsupervised Learning are. Deep Learning is a type of Machine Learning model with the characteristics that we explained above.

Deep Learning Models and Their Variety

There are different types of Deep Learning models, according to the type of input data and the type of the network architecture:

• Dense Neural Networks (Dense Nets) are used for structured data, that is, data that can be represented in a table format. This type of data are also handled well by other types of Machine Learning models.
• Convolutional Neural Networks (ConvNets) are used for mage data.
• Long Short Term Memory Networks (LSTMs), a special kind of Recurrent Neural Networks (RNNs), are used for learning long-term dependencies such as Time Series and Sequence Data.
• Generative Adversarial Networks (GANs) are used in Unsupervised Learning. They can generate images from descriptions, high resolution images from low resolution images, identify images that contain a given pattern, among other things.

Contrary to what many think, some types of Deep Learning models are not black boxes, but explanatory.