In humans, the recognition of specific combinations of sights and sounds gives rise to our perceptions of the world in which we live. The machinery underlying our perceptions is our sense organs (eyes, ears, touch, taste and smell receptors) and our nervous systems, including our brains.

In order to survive, animals need to react to the environment in which they live. If they detect a predator they must flee, whereas if they recognise an attractive mate they must move closer. On a more mundane level, they must shelter from the winter cold and move into the shade in the heat of summer. How do they do this? Basically they require two devices: an array of sensors and a pattern recogniser. The sensors detect the patterns of light, sound, smell, mechanical contacts and temperature of the external world. These are converted into signals and transmitted to pattern recognition devices (nervous system and brain) which respond to specific combinations of these signals. These devices (neural networks) in turn trigger the actions which enable the creatures to survive.

Some machines also behave as though they experience perceptions. Simple devices such as thermostats turn heaters on or off as the environmental temperature changes. More sophisticated devices such as optical character readers (OCRs) react to lines and shapes in much the same way as humans do when reading digits and letters. Recently, more complex devices such as automatic speech recognition systems, which specifically attempt to mimic the behaviour of humans, have appeared on the market.

Courses in Machine Perception and Neurocomputing bring together relevant knowledge from diverse fields such as biology, psychology, mathematics and computer science to enable students to understand the methods used by natural systems of perception, and to apply this knowledge to the development of artificial systems which will be employed in the machines of the twenty-first century. Such courses take students with degrees in engineering, physics or any of the above subjects and teach them the relevant topics from the other subjects. Successful completion leads to the award of an MSc degree.

A typical course consists of lectures on the mathematical foundations (statistics, linear algebra, optimisation theory), computational methods (computer programming and signal processing), biology and psychology of perception (neuroanatomy, neurophysiology and psychophysics), pattern recognition theory and artificial neural networks. This is followed by lectures on applications such as machine vision and image processing, speech processing (speech synthesis and recognition) and knowledge representation in computers (artificial intelligence). Practical experience is gained by undertaking a research project on which a dissertation is written.

A wide range of projects can be envisaged. For example, in screening for breast cancer, a large number of mammograms are produced. Each of these must be classified as normal or abnormal. Once this has been done, however, the mammograms can be scanned into a computer and used to train a neural network to classify further examples. Another project involved the construction of a robot that had ultrasonic sensors connected to its motors via a neural network. The neural network learned to control the motors so that the robots avoided colliding with obstacles.

People who apply for loans are usually required to complete a form with their personal financial details. The forms are scrutinised by a manager who decides whether they constitute a good financial risk. Later, this financial data (together with the repayment record) can be used to train a neural network so that future loan applications can be scruinised and a risk assessment made automatically. Examples of other projects which have been carried out include the recognition of vowel sounds in a noisy environment, the monitoring of infants in a paediatric ward, the verification of numbers printed on lottery tickets and the development of a device for simulating the action of a nose for testing the freshness of foodstuffs.

These courses are intellectually stimulating and provide an excellent introduction for a career in academic research. They also prepare graduates for employment in a wide variety of industries, such as information technology, telecommunications, aerospace, vehicle manufacturing, education, health and finance. Graduates in Machine Perception and Neurocomputing will be at the forefront of research and developments in these employment sectors for many years to come.

Author:
William Ainsworth
School of Life Sciences
Keele University, Staffordshire, UK