A scheme for organizing and controlling sensor webs is based on artificial neural networks. [Sensor webs were described in “Sensor Webs” (NPO-20616), NASA Tech Briefs, Vol. 23, No. 10 (October 1999), page 80. To recapitulate: Sensor webs are collections of sensor pods that could be scattered over land or water areas or other regions of interest to gather data on spatial and temporal patterns of relatively slowly changing physical, chemical, or biological phenomena in those regions. Each sensor pod is a node in a data-gathering/ data-communication network that spans a region of interest.] The present scheme would exploit communication and information-processing concepts that have enabled biological neural networks to organize and control large numbers of biological sensors, as proven in nature during the last billion years or longer.

One Example of Center-Surround Architecture is that of a retinal neuron and its nearest neighbors. The balance between the stimulatory effect of light and the inhibitory inputs from the nearest neighbors is such that the central neuron is active only when the edge of the shadow crosses the central neuron sufficiently close to the center. In a field of center-surround neurons, only those along the edge of the shadow are active.

From one perspective, the scheme could be characterized as one of designing artificial neural networks to have architectures approximating those of biological neural networks that perform specific functions. The following are examples of three such architectures:

  • In the center-surround architecture, neurons are arranged in one- layer sheets, and each neuron is connected to its immediate neighbors only (see figure). In nature, this architecture occurs most notably in the retina of the eye; it is also used to organize information coming from the ears and tactile sensors. In artificial neural networks, this architecture is most often found in those of the cellular-neural-network type. In both natural and artificial implementations, the basic functional topology is the same: the neural sheet is exposed to some input, and each neuron is prevented from firing by the inhibiting effects of approximately half of its neighbors.

    One of the many phenomena detectable by use of the center-surround architecture is the location and movement of edges across receptive fields — for example, the edge of a shadow on a retina. The concept of edges could be generalized to include isotherms and isobars on an area spanned by a web of weather sensors and to enable the use of sensor webs to detect such phenomena as the spread of radiation or toxic chemicals, the spread of seismic activity, the spread of a traffic jam (in the case of a sensor web that spans a city), or the movement of an intruder against a background of starlight.

  • The second architecture is that of autoassociative neural networks, which, in nature, enable organisms to recall old memories from partial or noisy stimuli. In an autoassociative network, each neuron is connected to every other neuron in the network. The excitability of any such neuron is determined by the state of all the other neurons and the strengths (weights) of the interconnections between neurons. For any state, the pattern of activation at the next instant in time is completely determined by the preset weights. The number of different firing patterns is 2n, where n is the number of neurons in the net; because this number is finite, as the sequence of firing patterns proceeds, cycles are inevitable. Cycles that consist of repeated instances of the same pattern are called fixed points, and these fixed points can be chosen by setting the weights to make the points attract nearby patterns. The fixed points represent memories. An autoassociative sensor web could be used to search for known gaseous, biological, or geological signatures, for example.
  • The third architecture is that of hypernetworks, which are groups of neural networks that can cooperate on vaguely defined tasks. Hypernetworks occur naturally in bee and ant colonies, schools of fish, and flocks of birds. For example, in a flock of birds, each bird functions, basically, as a single neuron connected only to its nearest neighbors. Each bird simply matches the speed and direction of nearest neighbors and keeps itself an equal distance between them. The entire flock seems to move as a single organism.

Swarms of neural networks could accomplish tasks that would be impossible for a single large neural network. For example, a swarm could spread out to cover a large area or move in single file to go through a small opening. Swarms of flying sensor pods, organized with simple hyperneural rules similar to those of flocks of birds, could perform a wide variety of exploratory and data-collection tasks.

This work was done by Charles Hand of Caltech for NASA’s Jet Propulsion Laboratory.

This Brief includes a Technical Support Package (TSP).
Document cover
Artificial Nueral Networks for Organizing Sensor Webs

(reference NPO-30317) is currently available for download from the TSP library.

Don't have an account? Sign up here.