Читать книгу Digital Forensics and Internet of Things - Группа авторов - Страница 16

It is a machine-based program which imitates the function of human intelligence. It can be considered as a subdivision of machine learning. As machine learning uses simpler concepts, and the deep learning makes used artificial neural networks in order to mimic how humans think and learn. This learning is categorized into supervised, semi-supervised, or unsupervised.

Deep learning can be constructed with the help of connected layers:

 • The foremost layer is known as the input layer.

 • The bottom-most layer is known as the output layer.

 • All the in between layers are known as the hidden layers. Here, the word deep indicates the connections between different the neurons.

Figure 1.2 depicts a neural network consisting of an input layer, a hidden layer, and an output layer. The hidden layers consist of neurons. Here, the neurons are interlinked with one another. The neurons help to proceed and transfer the given signal it accepts from the above layer. The stability of signal depends upon the factors of weight, bias, and the activation function.

A deep neural network produces accuracy in numerous tasks and might be from object detection to face recognition. This does not require any kind of predefined knowledge exclusively coded which indicates that it can learn automatically.

The Deep Learning process includes the following:

 • Understanding the problem

 • Identifying the data

 • Selecting the Deep Learning algorithm

 • Training the model

 • Testing the model

Deep neural network is a very strong tool in order to construct and predict an attainable result. It is an expert in pattern discovery and prediction that is knowledge-based. Deep learning algorithms are keen to provide 41% more accurate results when compared to machine learning algorithm in case classification of image and 27% better fit in case of recognizing of face and 25% in recognizing of voice.

Figure 1.2 Layers of the model.