Deep learning is a subset of machine learning. It is defined as a neural network with three or more layers. A layer simply refers to a collection nodes at a specific depth in a neural network.
Deep neural networks, or artificial neural networks, try to simulate the human brain. Neural networks consist of nodes. The input nodes take in a range of input data, whilst the outputs nodes output the result, either a classification or prediction. The middle nodes all have weights and biases, and use other techniques, such as forward propagation, back propagation, hidden layers, and dropout, to carry processing on the data. The several elements of a neural network complement each other to classify or make predictions based upon the given data.
A neural network only requires one layer to make predictions, however, increasing the number of these layers can aid in improving the accuracy of these predictions. On the other hand, increasing the number of layers, or modifying some of its features, increases the time taken for the neural network to be trained.
Through modifying the many features of a neural network, optimal deep neural networks can be constructed to efficiently and accurately solve problems.
The multiple layers of interconnected nodes use previous layers to make predictions. The process of computation through the network is known as forward propagation.
In deep learning, there are two main types of neural networks. Convolutional Neural Networks (CNNs) are mainly used in computer vision and in classifying images. Recurrent Neural Networks (RNNs) are used in natural language processing applications.
Deep learning, due to the neural networks having several layers, requires a huge amount of computing power. Graphical processing units (GPUs) are often used.
In today’s world, deep learning is widely used, for example in fraud detection in financial services, in chatbots to improve customer service, and in healthcare.