Introduction
The human brain has amazing
operative performance. It can work very fast, undertake tasks
in parallel, adapt to new situations, and deal with the unexpected.
It is also fault tolerant. Computers have become very fast and
can undertake some computations with great efficiency. However,
there are some areas, for example, visual recognition of objects,
where the human brain out performs the best computers.
Neuronal computing addresses
the development of computers and algorithms using the operation
and architecture of the human brain as a source of design inspiration.
The aim is to develop computers that are good at undertaking
tasks that are currently poorly performed by computers, but done
with great ease by people or other creatures. Another interest
driving research in this field is the development of better understandings
of how the brain works. In particular how the brain carries out
computations and information processing.
Main Issues
There is a need to understand
how the brain works and such work can benefit by bringing different
disciplinary perspectives to bear on this area of research. The
issues to be addressed by this research concern the identification
of the algorithms implemented in the brain and the way that neurones
actually produce these. Knowing how the brain works may be helpful
in understanding how to build better computers that have some
of those properties that the human brain possesses.
Some research is being undertaken
that combines insight from computational modelling and experimental
physiology, to understand the operation of neurones. Neurones
are in a state of continuous operation, emitting spikes spontaneously
and producing background noise. One of the key conclusions of
research in this area is that the emission of output spikes by
neurones to a meaningful input is enhanced by the addition of
noise. Without noise, the inputs can be filtered out resulting
in no output spike. The presence of noise tends to create sharper
outputs.
Reverse engineering in biology
is another filed of activity. Work in this area is concerned
with discovering the design principles of the brain in order
to design an equivalent solution in software. In this type of
research it is important to use knowledge from neuroscience,
psychology and computational modelling to achieve the objective.
Rapid visual processing, an activity where the brain outperforms
computers, is one application area that is being addressed. An
approach called spike based neural computing has been developed,
which is computationally efficient and can be implemented on
a personal computer. This approach uses a concept called temporal
coding, which basically means that the time taken to generate
a spike depends upon the stimulus intensity, with a larger stimulus
producing a spike faster than one of lower intensity. One of
the important conclusions about the potential of neuronal computing
is that it might lead to more natural interfaces for human interactions,
which would be better for people as the technology would work
in the same way that people operate.
Tools to help simulate the
brain's operation and to support co-operation between people
working in the field is a further area of research activity.
Computational models of the brain help with the understanding
of information processing taking place in the brain. Software
techniques however can provide people with the tools to collect,
compare and to analyse information from many different researchers.
Collaboration between different researchers is important. Data
from research is distributed and there is a need for tools to
support exchange, discussion and comparison of models and data.
Another perspective is the
engineering one. Computing implies a digital approach. Researchers
have been working on the development of a Neural Representation
Modeller, which provides modelling capabilities on different
levels: the digital neural cell; the neural module; and the neural
architecture. The architecture of the brain is a key area of
interest.
One question that should be
considered is the relevance of this sort of research. One of
the main application areas is in the solution of medical problems
and that industrial application might then spin-off from the
medical applications. Many people feel that this is a strange
situation as they expect to see medical applications arising
as a spin-off from industrial applications. Neuronal computing
however has the potential to change many things in industry in
the longer term.
Another question is whether
biological computers will ever become a reality. Some work has
already been undertaken in this area, where an analogue VLSI
based on neurones has been developed. However working with something
like a high performance personal computer has advantages since
they are rapidly evolving and no special hardware technologies
are required. This latter point means that more people could
be able to undertake work in this field. At the moment Pentium
microprocessors are ideal, but ultimately all technologies are
potentially interesting.
Conclusions
and Future Directions
A lot of further research is
needed in the area of neuronal computing. There is an expectation
that this research will eventually inspire new computer architectures.
There are things that can be learnt from biology that can be
transferred to engineering and vice versa. Cross fertilisation
between disciplines such as computer scientists and brain researchers
is important.
There are different ideas about
neuronal models. There are also different views about the level
of modelling and what constitutes the important units to study
in the brain and to take inspiration from. There is a need to
work at all levels and also to work across levels. Perhaps this
is where most work actually needs to be focused. The important
thing is to ensure that there is no particular bias in the research
supported in the future and that the different perspectives are
allowed to flourish. |