Artificial Neural Networks

Image Credit: Photos.com

An artificial neural network is a mathematical model inspired by the complex network of neurons in the brain. This system includes an interconnected group of artificial neurons that processes information using a connectionist computation approach.

A neural network is an adaptive system whose structure is changed during a learning phase. Neural networks are used for modeling complicated relations between outputs or inputs or to determine patterns in data.

Research

Eng. Abdelfattah A. Ahmed et al in 2011 used an approach based on neural networks to identify nuclear research reactor accidents.

A neural network is trained and designed, first without noise so as to identify the pattern of accidents in nuclear research reactors.

The simulation of the neural network response resulted in an output value almost equal to 1 indicating the network is well-trained. Another copy of the neural network was made by training this system with noisy accident patterns.

Training of the network on this noisy input vector reduces its errors drastically and causes the output to be almost equal as when training is done without noise input vectors. This system was also trained on accident patterns without noise to obtain high consistency of the network.

Experiments have shown good results at places where the network did not make any errors for input patterns with noise levels from 0 to 0.14. When a noise level of more than 0.15 was added to the pattern, both neural networks began making errors.

Jayanta Kumar Basu et al (2010) conducted research on the use of an artificial neural network in pattern recognition. Among the conventional approaches of pattern recognition, the statistical approach has been studied the most and also used in practice. A recognition system needs to be designed for the following:

• Defining sensing environment, pattern classes, pattern representation
• Performing feature extraction and selection, classifier design, cluster analysis, selection of test and training samples, and performance evaluation.

New and evolving applications such as web searching, data mining, face recognition, retrieval of multimedia data, and cursive writing recognition need effective and robust pattern recognition techniques.

The research study reviewed and summarized some popular methods in various stages of a pattern recognition system using ANN and determined new research areas and applications in this field.

Dong, Xiu-Cheng et al (2008) conducted research on bus protection based on artificial neural networks. A bus is highly important in transformer substations and power plants.

Research on high-intelligence and high-reliability bus protection will be significant theoretically and practically. An artificial neural network is used instead of physical bus protection, presenting a novel bus protection method based on approximation ability of function on ANN.

Current Applications

Fuzzy logic is a kind of logic that understands more than true and false values and simulates the real world around us. For instance, the statement today is cloudy maybe 100% true, 80% true if it is a bit sunny and probably 50% true if it is quite sunny and there are a few clouds. So instead of just binary values, fuzzy logic incorporates concepts such as usually, somewhat, and sometimes.

The combination of neural networks and fuzzy logic are used for applicant screening for jobs, crane control, automotive engineering, and glaucoma monitoring.

Some other applications of neural networks are:

• Character Recognition – Portable devices like the palm pilot are gaining popularity as character recognition is slowing becoming more significant. Neural networks can be used to identify handwritten characters.
• Image Compression – Neural networks are capable of receiving and processing large amounts of data at once making them useful in image compression.
• Stock Market Prediction – The daily business of stock markets is highly complex. Several factors determine whether stocks will rise or fall on a particular day. As neural networks are capable of analyzing a lot of information quickly and sorting the same, stock prices can be predicted.
• Medicine, Electronic Nose, Security, and Loan Applications – These applications are still at the proof-of-concept stage; however, it is believed that they will soon be in vogue.

Future Developments

All present neural networks may be very much improved in the future. Everything from speech and handwriting recognition to stock market prediction will be more advanced as researchers are developing better network architectures and training methods.

Future neural networks may enable:

• Robots that can see, feel, and perceive the world around them.
• Better stock prediction.
• Self-driving cars will become a reality.
• Music composition, handwritten documents automatically changed to formatted word processing documents.
• Self-diagnosis of medical problems and many more.

Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.