The importance of Artificial Neural Networks in realising Artificial Intelligence
An Artificial Neural Network (ANN) is key in understanding how Artificial Intelligence works. Artificial intelligence was built to mimic aspects of human intelligence, such as learning and decision making. The need for implementation of artificial intelligence drove the development of Artificial Neural Networks.
What are Artificial Neural Networks?
ANNs are biologically inspired computational networks. They are built like the human brain, with neuron nodes interconnected like a web, in order to enable the computer to learn and make decisions like a human. ANNs are created by programming regular computers to behave as though they are interconnected brain cells. In the words of Dr. Robert Hecht-Nielsen, inventor of the first neurocomputer, a neural network is defined as — “…a computing system made up of a number of simple, highly interconnected processing elements, which process information by their dynamic state response to external inputs.”
Basic Structure of an Artificial Neural Network
ANNs are inspired by the biological nervous system. The human brain is composed of 86 billion nerve cells, also called neurons. They are connected to thousand other cells by axons. Stimuli from the external environment or inputs from sensory organs are received by dendrites. These inputs are then transmitted through the neural network using electric impulses. The output from each neuron is in the form of response / action to stimuli.
The basic architecture of an ANN is divided into layers categorized as input, hidden and output layers. An ANN can have 1 or more hidden layers according to its function. Each layer consists of 1 or more artificial neurons.
The connection between the neuron inputs and neuron outputs are represented by the directed edges with weights. The neuron sums all the signals it receives, with each signal being multiplied with its associated weights on the connection. If the weighted sum equals zero, a bias is added to make the output non-zero or else to scale up to the system’s response. Bias has a weight and the input to it is always equal to 1. The weighted sum output is then passed through an activation function, that is usually non-linear, to give the final output. The activation function, can be described as the set of transfer functions used to get the desired response from the ANN.
Types of Artificial Neural Networks and their applications
1. Feedforward Neural Network
The information in the neural network travels only in the forward direction. This is the simplest form of an ANN and does not have the provision for back propagation.
Applications: Speech Recognition, Computer Vision
2. Modular Neural Network
In this type of neural network, many independent neural networks perform sub-tasks and contribute to the results collectively. The complexity of the problem is thus reduced and the computing speed is faster.
Applications: Function approximation, Stock market prediction
3. Radial Neural Network
It has two functional layers that are used to estimate the distance of the centre with respect to a point. In the first layer, features in the inner layer are united with the Radial Basis Function. In the next step, the output from this layer is considered for computing the same output in the next iteration.
Application: Power Restoration Systems
4. Konohen Self Organising Neural Network
In this neural network, vectors and small weights are given as input to each neuron in a discrete map from an arbitrary dimension. The neuron’s location does not change while training the map. A winning neuron is a neuron that is closest to a particular point. Other neurons then start to moving towards the point along with the winning neuron in the second phase. Each neuron represents each kind of cluster, and the clustering of all the points happen through the iterations.
Application: Pattern Recognition
5. Recurrent Neural Network
In this Neural Network, the output layer is provided as feedback to the input. During the process of computation, each neuron acts as a memory cell. RNNs work by evaluating sections of an input in comparison with the sections both before and after the section being classified through the use of weighted memory and feedback loops. This serves the purpose of error correction for optimal output.
Application: Text to speech conversion
6. Convolutional Neural Network
It is a multi-layer neural network used to analyse images for image classification, segmentation or object detection. CNNs work by reducing an image to its key features and using the combined probabilities of the identified features appearing together to determine a classification.
Application: Computer Vision
7. Long / Short Term Memory
The main goal of this neural network is to remember data for a long time in a memory cell that is explicitly defined. The memory can be erased using a “forget gate”. New data is added to the memory cell through the “input gate”, and is passed to the next hidden state from the cell along the vectors decided by the “output gate”.
Applications: Composition of primitive music, Learning complex sequences
Advances of Artificial Neural Networks
The applications of Artificial Intelligence and Neural Networks are constantly growing. Combining other algorithms and technologies with ANNs have expanded the fields of applications and usability. Popular commercial applications of ANNs are in chatbots and inventory management. Chatbots automate the process of customer support, eliminating the need for human presence round the clock. Other major applications include virtual assistants like Siri and Cortana, and face recognition. The healthcare sector can benefit greatly from the use of ANNs and artificial intelligence, detecting diseases sooner and suggesting effective treatments.
The future of Artificial Neural Networks is set to unlock multiple possibilities in various business sectors. Hence, both governments and businesses must be cognizant of advantages in the adoption of neural networks and devise strategies accordingly.
References:
https://www.investopedia.com/terms/a/artificial-neural-networks-ann.asp
https://www.tutorialspoint.com/artificial_intelligence/artificial_intelligence_neural_networks.htm
https://www.upgrad.com/blog/types-artificial-neural-networks-in-machine-language/
https://missinglink.ai/guides/neural-network-concepts/cnn-vs-rnn-neural-network-right/
https://www.allerin.com/blog/the-evolution-of-neural-networks
