5 Artificial Neural Networks Supporting Machine Learning

ANNs: the backbone of artificially intelligent technology

The human brain is especially good at solving problems… So good, in fact, that in the 1940s various computer scientists began to build computation models in an attempt to replicate it. The Artificial Neural Networks (ANNs) that they created are responsible for major advancements in machine learning. By using nodes to imitate biological neurons, ANNs mimic the process of information exchange within our brains and provide the framework for algorithms to function. ANNs are used to find patterns in input data that humans might not detect due to complexity or volume. As data grows at an unprecedented rate, this has become all the more important. Today, ANNs enable computer vision, speech and image recognition, machine translation, medical diagnosis and gaming applications, with the potential to be used for so much more.

1) Convolutional Neural Network (CNN)

CNNs (also known as ConvNets) are a class of ANN commonly applied to the analysis of visual imagery. Like all neural networks, they are comprised of input and output layers, as well as a number of unspecified hidden layers. CNNs contain one or more convolutional layers, pertaining to their name. In mathematics, a convolution is an operation on two functions to produce a third function that shows how one is changed by the other. In comparison to other neural networks, they use relatively little processing power. CNNs can identify faces, objects, and traffic signs, and power the vision in robots and other autonomous systems like self driving cars.

2) Feedforward Neural Network

Feedforward Neural Networks are one of the oldest and simplest forms of ANN, which first came into practical use during the 1990s. In feedforward systems, information moves in one direction from the input nodes to the output nodes, travelling through any hidden nodes on the way. The connections between nodes don’t form a cycle which, as far as complex neural networks go, makes them very straightforward. Applications include pattern recognition and functional mapping problems.

3) Modular Neural Network

Modular Neural Networks are typically used in predictive analytics. Instead of functioning as one concentrated neural network, they are made up of a series of independent networks which are also known as modules, hence the name. These modules carry out specific subtasks to accomplish an overall aim. Each module exists independently and does not interact with other modules in the network, however they are moderated by an intermediary. The intermediary accepts the modules’ inputs to create the final output. Benefits include efficiency, as fewer connections means less processing time. While large neural networks can fail at any node, modular alternatives are more robust because nodes are confined and therefore don’t effect the performance of other modules.

4) Radical Basis Function Network (RBF)

RBFs were first devised in a 1988 paper by researchers at the Royal Signals and Radar Establishment. They are a form of supervised ANN that uses supervised machine learning as a nonlinear classifier. Linear classifiers are based on linear criteria and have a basic form, whereas nonlinear classifiers are not restricted to one form or constrained by rigid criteria. This makes RBFs far more intuitive, but their basic structure remains the same: information travels through input layers, output layers, and any hidden layers in between. RBFs are used in classification, function approximation and system control.

5) Recurrent Neural Network

Recurrent Neural Networks are a type of Recursive Neural Network. Unlike Feedforward Neural Networks, the connections between the nodes in a Recurrent Neural Network form a cycle. This allows for a bidirectional flow of data, which literally means that information can be transmitted in both directions. As such, Recurrent Neural Networks can achieve complex tasks. One of the most common applications for Recurrent Neural Networks is in Natural Language Processing (NLP).

To find out more about artificiall intelligence, sign up for our free newsletter.