|
|
|
|
|
|
 |
Applications of Neural NetworksNeural Networks in SciencePerhaps the most obvious use of neural networks is in better understanding the biological brains they are based on. Recently "computational neuroscience" has become an intense focus of research with many articles and books written on the topic. The networks used vary widely with some modelling the individual molecules and ion channels within neurones as accurately as possible, and others focusing on some of the biological learning outlined earlier. Other uses of neural networks in science and technology follow. Solar Flare Prediction with Neural Networks Researchers at Lund University in Sweden used neural networks to predict the effects of solar flares. Solar flares are known to disturb the earth's magnetic fields and this in turn can disrupt electronic products causing blackouts etc. The neural network used various related variables to predict the following days geomagnetic activity, and how severe such activity might be. After being trained on several months of data from various US databases. The network compared favourably to traditional prediction methods (NOAA/SEL) when tested in June 1990. During the testing period three major storms and one minor storm occurred. Traditional methods predicted only a single minor storm while the neural network did much better correctly predicting two of the major storms and the minor storm, and misclassifying the third major storm as another minor storm. Protein Sequence Recognition with Neural Networks Researchers at Wayne State University used neural networks to identify whether a particular amino-acid sequence of protein was trans-membrane or not. The network was trained on over 1500 sequences until an accuracy of about 98% was achieved. The neural network was then used to determine the position of bacteriohodopsins 7 trans-membrane helices. Mosquito Identification with Neural Networks Aubrey Moore working at the University of Hawaii, developed a neural network to distinguish between species and sexes of mosquitoes in flight. After a training set of over 400 samples (approximately 100 for each of sex, and each of the two species being assessed) he tested the network on 57 samples. The neural network correctly identified every species and sex of mosquito. Typical discriminate analysis had provided an accuracy of about 85%. Spectroscopy with Neural Networks StellarNet Inc. used neural networks to help identify substances using data derived from spectroscopy. The company has used this spectroscopy technology for a wide variety of purposes. In the agriculture area, for example, one client uses the technology to identify and assure proper hydration of recently harvested onions. SpectraNet used data from absorbance, transmittance, and reflectance etc. to correctly determine hydration. Weather Forecasting with Neural Networks Tony Hall, working in the National Weather Service in Texas, used neural networks to predict local rainfall to an accuracy of 85%. While the data used by these neural networks was available for some time, the relationships between these variables are complex and ever changing. The use of a neural network that learned rather than analysed these relationships showed great promise. Air Quality Testing with Neural Networks At Intel, neural networks were used to identify fabrication problems that caused failures in Intel's VLSI chips. Traditionally, expert systems had been used but they were found to be incapable of generalising their knowledge. So, Dr Dan Seligson PhD trained a neural network using the expert system and process data until it was 99.5% accurate in assessing data similar to that used in training. Quality Control with Neural Networks At Intel, neural networks were used to identify fabrication problems that caused failures in Intel's VLSI chips. Traditionally, expert systems had been used but they were found to be incapable of generalising their knowledge. So, Dr Dan Seligson PhD trained a neural network using the expert system and process data until it was 99.5% accurate in assessing data similar to that used in training. Another example of neural networks in industry is their implementation by the National Institute of Standards and Technology (NIST) to develop a non-destructive method for testing the internal structure of concrete. Steel balls are dropped onto the concrete causing a characteristic sound. By analysing the sound, neural networks determine the structure of the concrete and the probability of a flaw. Beer Testing with Neural Networks Anheuser-Busch use neural networks to determine the content of their own, and their competitors beer vapours. This allows them to ensure consistent quality of their own beer as well as keep an eye on any changes made in other beers. Speech Recognition with Neural Networks Neural networks have been used by companies for simple speech recognition allowing clients to reach the voicemail address they require using their voice rather than a touch-tone phone. Dr Mark Ortner trained a network to recognise 28 words, including the numbers 0 to 9, the words "yes" and "no" etc. with an accuracy of 90-95%. Classification of Text with Neural Networks Electronic Data Publishing incorporated a neural network in order to read journal articles and recognise data as the title, author, publisher, or date so that this information can be stored in a database for later retrieval. After about 100 training runs, the network had about 95% accuracy on examples. Chemical Drawing Recognition with Neural Networks Fein-Marquart Associates developed a neural network that reads printed chemical drawings and records them as connection tables in a database. Joe McDaniel, a Senior Staff Member at the company, said the system had a 98% recognition success rate. Previous | Next | Page 1 2 3 4 Index | Applications | Science References: © 2008 Marcus bros
|
|
|
|
