Deep Learning is a subset of Machine Learning. ML algorithms are designed to perform a given tasks without providing explicit instructions. A few types of these algorithms are Linear Regressions, Random Forest, Decision Trees, Support Vector Machines. Deep Learning, specifically, use Artificial Neural Networks as the architecture for their algorithms.
Machine revolution began back in 1940 and has since been called by many names. Deep Learning had a resurgence when Alex Krizhevsky released his monumental algorithm AlexNet on 10 September, 2012.
Understanding Artificial Neural Network
Artificial Neural Network (ANNs) are also called Connectionist Systems and are loosely inspired by the biological brain.
To make an ANN, a computer program is used to create virtual neurons or nodes. These nodes are connected with together. The connections serve the purpose of information or data transmission between these nodes. Each node is initially assigned a random value. This value is called weight. Each node is capable of performing simple algebraic operations. The algebraic operation uses weight and data received from the previous nodes.
Universal Approximation Theorem states that by modifying the weights of the nodes, ANNs are capable of learning a relationship between input and output, if it exists.
What is the Input & Output for ANNs?
This is dependent on the task required to be performed. AlexNet was designed to recognize objects in the images (Computer Vision). In this case the input was images and the output was the name of the object. The AlexNet ANN was able to approximate the functions required to identify patterns of different objects.
At Yobee Research, the objective is to decode the complexities of Stock Markets. For this applications, the input is financial data such as GDP, currency exchange rates, stock prices and the output is the movement of the market.