BIOLOGICAL INSPIRATION FOR COMPUTING 277
- Monitoring the condition of machinery. Neural networks can be instrumental in cutting costs by
bringing additional expertise to scheduling the preventive maintenance of machines. A neural network
can be trained to distinguish between the sounds a machine makes when it is running normally (“false
alarms”) versus those it makes when it is on the verge of a problem. After this training period, the
expertise of the network can be used to warn a technician of an upcoming breakdown, before it occurs
and causes costly unforeseen “downtime.” - Engine management. Neural networks have been used to analyze the input of sensors from an
engine. The neural network controls the various parameters within which the engine functions, in order
to achieve a particular goal, such as minimizing fuel consumption.
8.3.3.3 Neurally Inspired Sensors,
One of the first attempts to draw on the principles underlying biological sensors occurred in the
mid-1980s, when researchers such as Carver Mead and his coworkers at Caltech made their first at-
tempts to create artificial retinas using VLSI technology,^92 with hoped-for applications that ranged from
artificial eyes for the blind to better sensors for robots. A second, more recent example of a neurally
inspired sensor is the computational sensor of Brajovic and Kanade.^93 Many approaches toward im-
proving machine vision have been based on better cameras with higher resolution and sensitivity, new
sensors such as uncooled infrared cameras, and new recognition algorithms. But standard vision sys-
tems typically have high latency (a long time between registration of the image on the vision system’s
sensors and image recognition), induced by the requirements of transferring large amounts of data from
the sensor to the processor and processing those large amounts of data quickly. In addition, latency
increases more or less linearly with image size. Standard vision systems can also be very sensitive to
small details in the appearance of an object in sensor images. A number of processor-based algorithms
have been developed that adjust for such variations, but they are often complex and ad hoc, and hence
unreliable.
The computational sensor approach borrows biological architectural principles to use low-latency
processing and top-down sensory adaptation as techniques for speeding up vision processes. Computa-
tional sensors are (usually) VLSI circuits that include on-chip processing elements tightly coupled with
on-chip sensors, exploit unique optical design or geometrical arrangement of elements, and use the
physics of the underlying material for computation. The integration of sensor and processor elements
on a VLSI chip enables latency to be reduced by a considerable factor and provides opportunities for
fast processor-sensor feedback in service of top-down adaptation—and computational sensors have
produced an order-of-magnitude improvement in sensing and information processing itself, such as
range sensing, sorting, high-dynamic range imaging, and display.
8.3.4 Ant Algorithms,
Ant colonies depend on workers that can collectively build nests, find food, and carry out a multi-
tude of other complex tasks while having little or no intelligence of their own. Further, they must do so
without the benefit of a leader to organize their efforts. They also continue to do so even in the face of
outside disruptions, or the failure and death of individual members, thereby exhibiting a high degree of
flexibility and robustness.
(^92) M.A. Sivilotti, M.A. Mahowald, and C. Mead, “Real-time Visual Computations Using Analog CMOS Processing Arrays,” pp.
295-312 in Advanced Research in VLSI: Proceedings of the 1987 Stanford Conference, P. Losleben, ed., MIT Press, Cambridge, MA,
1987.
(^93) V. Brajovic, “Computational Sensor for Global Operations in Vision,” Ph.D. Thesis, Carnegie Mellon University, Pittsburgh,
PA, 1996.