selectivity by including correlations in the electri- 
cal activity between eyes. Corresponding points in 
the two eyes tend to be correlated because, on aver- 
age, they look at the same point in space. The devel- 
opment of disparity selectivity was simulated in two 
stages: prenatal, when the two retinas have essen- 
tially independent activities, and postnatal, when 
the eyes are open and have correlated activities. By 
varying the degree of development that occurred in 
the model before eye opening, a mixture of monocu- 
lar and binocular cells arose with the experimen- 
tally observed relationship to disparity. 
Disparity can be combined with other cues to pro- 
vide information about the distance of an object 
from the viewer. With grant support from the Office 
of Naval Research, Dr. Sejnowski's laboratory has 
shown how the vergence of the two eyes (angle be- 
tween the two lines of sight) and the binocular dis- 
parity could be combined to represent the distance 
to an object in a population of neurons. Single neu- 
rons in the visual cortex have disparity-tuning 
curves that are broad and overlapping. In the net- 
work model, distances were represented by a popu- 
lation of such neurons whose responses were modu- 
lated by the vergence angle. Neurons with such 
properties have recently been observed in the pri- 
mary visual cortex and the posterior parietal cortex, 
a region of the brain that is essential for the internal 
representation of external space. These new find- 
ings suggest that transformations from retinal to spa- 
tial representations could be initiated much earlier 
than previously thought in the visual system. Com- 
puter models are being developed to study the con- 
sequences of incremental spatial transformations in 
a feedforward hierarchy of cortical maps. 
Motion Processing in Visual Cortex 
The primate visual system is very good at tasks 
such as tracking a moving object against a textured 
background. To track a moving object, the visual 
system must integrate local motion estimates from 
many neurons, each with a limited spatial receptive 
field. The integration of information on motion can 
be affected by a variety of cues such as contrast, 
spatial frequency, binocular disparity, color, trans- 
parency, and occlusion. Thus the integration of mo- 
tion signals cannot be performed in a fixed manner 
but must be a dynamic process dependent on the 
properties of the visual stimulus. There is the addi- 
tional problem of segregating information from a 
single object when there are several objects. 
Dr. Sejnowski's laboratory has developed a simple 
model for motion processing in area MT, a region of 
the primate visual cortex that specializes in repre- 
senting motion. The model assumes two popula- 
tions of neurons at each position in the visual field: 
one population computes estimates of motion in a 
local region of the visual field while the second pop- 
ulation estimates the relevance or reliability of each 
local motion estimate. Outputs from the second 
population of neurons then gate the outputs from 
the first population of neurons through a gain- 
control mechanism, before the local motion esti- 
mates are integrated to form more global estimates. 
The proposed mechanism of gain control is consis- 
tent with measured responses of cortical cells under 
conditions of interfering motion of transparent stim- 
uli. In addition, predictions were made for the re- 
sponse properties of neurons in MT that may be esti- 
mating the reliability of the local motion estimates. 
Visual area MT provides the oculomotor system 
with information about the motion of moving ob- 
jects so that they can be smoothly tracked by eye 
movements. Dr. Sejnowski is collaborating with Dr. 
Stephen Lisberger (University of California, San 
Francisco) to develop models of the oculomotor 
system that will complement the models of visual 
processing. Their models of motor control are based 
on networks of neurons that include feedback con- 
nections, which makes them highly dynamic. 
Dr. Sejnowski is also Professor in the Computa- 
tional Neurobiology Laboratory at the Salk Insti- 
tute for Biological Studies and Professor of Biol- 
ogy and Adjunct Professor of Neuroscience, 
Physics, Psychology, Cognitive Science, Electrical 
and Computer Engineering, and Computer 
Science and Engineering at the University of Cali- 
fornia, San Diego. 
Books and Chapters of Books 
Churchland, P.S., and Sejnowski, T.J. 1992. The 
Computational Brain. Cambridge, MA: MIT 
Press. 
Lehky, S.R., and Sejnowski, T.J. 1991. Neural 
model of stereoacuity based on a distributed rep- 
resentation of binocular disparity. In Limits of Vi- 
sion: Vision and Visual Dysfunction (Kuli- 
kowski, JJ., Walsh, V., and Murray, IJ., Eds.). 
New York: MacMillan, vol 5, pp 133-146. 
Sejnowski, T.J. 1991. David Marr: a pioneer in 
computational neuroscience. In From the Retina 
to the Neocortex: Selected Papers of David Marr 
(Vaina, L.M., Ed.). Boston, MA: Birkhauser, pp 
297-301. 
Articles 
Bush, P.C., and Sejnowski, T.J. 1991. Simulations 
of a reconstructed cerebellar Purkinje cell based 
436 
