Perception and Attention - Computational Explorations in Cognitive Neuroscience - page 263

ferent objects. If the spatial distances associated with
these operations are the same in the two conditions,
any difference in reaction time would indicate a cost
for switching attention between the two objects. Such
an object cost has been found in a number of studies
(Duncan, 1984; Vecera & Farah, 1994; Mozer et al.,
1992).
In the simulator, we can run the simpler cuing ex-
periment analogous to the Posner task because we have
transparent access to the internal representations, and
can measure the object processing facilitation directly.

Go to the PDP++Root window. To continue on to
the next simulation, close this project first by selecting
.projects/Remove/Project_0 . Or, if you wish to
stop now, quit by selecting Object/Quit .

8.5.3 Summary and Discussion

Perhaps the most important lesson from this simulation
is that attentional effects are the direct consequence of
some of the basic principles developed in the first part
of this topic: inhibitory competition, bidirectional in-
teractive processing, and multiple constraint satisfac-
tion. This has some important implications. First, it
suggests that we should expect to find attention-like ef-
fects throughout the cortex, consistent with some re-
cent views (Desimone & Duncan, 1995; Allport, 1989).
Second, as emphasized by Cohen et al. (1994), under-
standing attention in terms of more basic neural infor-
mation processing principles can result in a simpler con-
ceptual model that avoids the need for things like a dis-
engage mechanism. Third, it suggests that by studying
attentional effects, we are observing these fundamental
principles in action, and we could potentially use de-
tailed behavioral data to further constrain our model's
implementation of these principles. For example, the
detailed time-course of cue presentation, delay periods,
and the measured attentional cuing and inhibition of re-
turn effects should provide specific constraints.
There are a number of important spatial attention ef-
fects that we have not addressed in our model, but that
are addressed by other models based on similar prin-
ciples (e.g., Mozer & Sitton, 1998). For example, it
appears that spatial attention can be expressed in a va-
riety of different shapes and sizes. Using the popular
“spotlight” analogy, the spotlight of attention can range
from narrow and intense to wide and diffuse, and it can
also be discontinuous. These kinds of phenomena could
potentially be addressed in an expanded version of the
present model by scaling up the spatial and object rep-
resentations to allow for much more complex patterns
of activity to develop.
There are also detailed neurophysiological record-
ings of attentional phenomena in monkeys that could
potentially be simulated using something like the model
we explored here

Set env_type to OBJ_ATTN . Then do View ,
EVENTS to see what the network will experience.
The first event is a control condition, where we
present two objects without any prior cuing. Note that,
as in the MULTI_OBJ case, the target object is more
strongly activated than the cue object, so the network
will process the target object. The next event is a cuing
event, where the cue object is presented in the central lo-
cation. Then, within the same group so that activations
persist, the next event presents the two objects just as in
the first event. Thus, if the prior object cue is effective,
it should be able to overcome the relatively small differ-
ence in bottom-up salience between the two objects, so
that the network processes the cue object and not the tar-
get. Finally, the next two events are for the case where
the two objects appear in the same location. Recall that
before, the network was unable to select either object
for processing in this case, because they are spatially
overlapping. Perhaps now, with the object-based atten-
tional cue, the network will be able to focus on the cue
object.

Press Defaults . Then Step (you may also need
to turn the network display on). Note how the network
responds to each of the three task conditions as you
continue to Step through these cases.
You should observe that the prior object cue is indeed
capable of influencing subsequent processing in favor
of the same object. Note also that the spatial system re-
sponds to this in the appropriate manner — it activates
the spatial location associated with the cued object. Fi-
nally, note that the top-down object cue is sufficient to
enable the system to select one object (even the less ac-
tive one) when the two objects are presented overlap-
ping in the same location.

(e.g., Moran & Desimone, 1985;