Page 342 - Electrical Engineering Dictionary

heavy water water in which a heavy iso- heirarchical coding coding where im-
tope of hydrogen substitutes for the hydrogen age data are encoded to take care of different
atoms. resolutions and scales of the image. Addi-
tional data is transmitted from the coder to
Hebb’s principle to update the weights of refine the image search. See also progressive
the simple feedforward neural network (e.g., transmission.
perceptron) a very simple idea, called Hebb’s
principle, can be used. The principle states heirarchical interpolation a technique
the following: Apply a given pattern to the of forming image pyramids. In this method,
inputs and clamp the outputs to the desired pixels are interpolated to higher levels in the
response, then increase the weights between hierarchy. Thus, only the interpolated pixels
nodes that are simultaneously excited. This need a residual in order to be reconstructed
principle was formulated by Hebb in 1949, in exactly, since the subsampled pixels are al-
an attempt to model neural networks. Math- ready correct.
ematically it can be expressed as follows:
helical antenna a wire antenna that is he-
0
w ij (t + 1) = w ij + γu i (t) y (t) − y j (t)
j lical in shape. Typically, the helical antenna
0
where y is the desired response and y j is is fed by a coaxial cable, the outer conduc-
j tor of which is connected to a ground plane
the response of the network. By regarding
and the inner conductor of which protrudes
the weights as parameters, one may treat the
through the ground plane and is shaped in the
above formula as a gradient method for pa-
form of a helix. Since a helical antenna emits
rameters estimation.
an elliptically polarized wave when operat-
ing in axial mode, it is often used in satellite
Hebbian algorithm in general, a method
communication applications.
of updating the synaptic weight of a neuron
w i using the product of the value of the ith
helical beam tube a backward wave am-
input neuron, x i , with the output value of the
plifier based on interaction between a for-
neuron y. A simple example is:
ward helical electron beam launched into
w i (n + 1) = w i (n) + αy(n)x i (n) a waveguide and backward traveling mi-
where n represents the nth iteration and α is crowave in the waveguide.
a learning-rate parameter.
helical scan tape a magnetictapein which
Hebbian learning a method of modifying recording is carried out on a diagonal to the
synaptic weights such that the strength of a tape, by a head that spins faster than the tape
synapic weight is increased if both the presy- movement. Popular for VCRs, camcorders,
naptic neuron and postsynapic neuron are ac- etc., as it improves recording density and tape
tive simultaneously (synchronously) and de- speed.
creased if the activity is asynchronous. In ar-
tificial neural networks, the synaptic weight helium neon laser neon laser in which ex-
is modified by a function of the correlation citation of the neon atoms occurs primarily
between the two activities. by collisions with electron-impact-excited
helium atoms; the first gas laser, and espe-
hedge a special linguistic term by which cially important for its CW visible oscillation
other linguistic terms are modified. Exam- lines.
ples are very, fairly, more or less, and quite.
Helmholtz coils a pair of coaxial coils of
height defuzzification See centroid the same diameter that are spaced one radius
defuzzification. apart. They are used to generate a field of

c
2000 by CRC Press LLC