276 CATALYZING INQUIRY
Neural networks are most useful for problems that are not amenable to computational approaches
and are constrained by strict assumptions of normality, linearity, variable independence, and so on.^85
That is, they work well in classifying objects, capturing associations, and discovering regularities within
a set of patterns where the volume, number of variables, or diversity of the data is very great; when the
relationships between variables are vaguely understood or the relationships are difficult to describe
adequately with conventional approaches; or when the problems in question are ill-posed and involve
high degrees of uncertainty.^86 In addition, they are well suited for problems that are subject to distor-
tions in the input data.
Neural networks have been applied to a large number of real-world problems of high complexity,
including the following.^87
- Optical character recognition. Commercial OCR (optical character recognition) software packages
have incorporated neural network technology since the mid-1980s, when it significantly increased their
ability to recognize unfamiliar fonts and noisy, degraded documents such as faxes.^88 Today, OCR
systems typically use a mix of neural network and rule-based approaches. - Finance and marketing. Neural networks’ ability to detect unanticipated patterns has made them a
favored tool for analyzing market trends, predicting risky loans, detecting credit card fraud, managing
risk, and many other such tasks in the financial sector.^89 - Security and law enforcement. Neural networks’ pattern-detection ability has likewise made them
a useful tool for fingerprint matching, face identification, and surveillance applications.^90 - Robot navigation. Neural networks’ ability to extract relevant features from noisy sensor data can
help autonomous robots do a better job of avoiding obstacles.^91 - Detection of medical phenomena. A variety of health-related indices (e.g., a combination of heart
rate, levels of various substances in the blood, respiration rate) can be monitored. The onset of a
particular medical condition could be associated with a very complex (e.g., nonlinear and interactive)
combination of changes on a subset of the variables being monitored. Neural networks have been used
to recognize this predictive pattern so that the appropriate treatment can be prescribed. - Stock market prediction. Fluctuation of stock prices and stock indices is another example of a
complex, multidimensional, but in some circumstances at least partially deterministic phenomenon.
Neural networks are being used by many technical analysts to make predictions about stock prices
based on a large number of factors such as past performance of other stocks and various economic
indicators. - Credit assignment. A variety of pieces of information are usually known about an applicant for a
loan. For instance, the applicant’s age, education, occupation, and many other facts may be available.
After training a neural network on historical data, neural network analysis can identify the most rel-
evant characteristics and use them to classify applicants as good or bad credit risks.
(^85) This material adapted from http://cfei.geomatics.ucalgary.ca/matlab/ann.html.
(^86) See http://www.cs.wisc.edu/~bolo/shipyard/neural/neural.html.
(^87) See http://www.emsl.pnl.gov:2080/proj/neuron/neural/what.html; see also http://neuralnetworks.ai-depot.com/
Applications.html. Examples in the list below for the topics “detection of medical phenomena” through “engine manage-
ment” are taken from http://www.statsoftinc.com/textbook/stneunet.html#apps.
(^88) See http://www.scansoft.com/omnipage/ocr/. At the time, the state of the art in commercial OCR software was the rule-
based approach, in which a system broke each character image into simple features and then identified the letters by reasoning
about curves, lines, and such. This approach worked well—but only if the fonts were known and the text was very clean.
(^89) See http://neuralnetworks.ai-depot.com/Applications.html; see also http://www.nd.com/ and http://www.walkrich.com/
value_investing/howdo.htm.
(^90) See http://www.neurodynamics.com/.
(^91) See http://ai-depot.com/BotNavigation/Obstacle-Introduction.html.