Biomedical Engineering Reference
In-Depth Information
TE
view independence
Layer 4
50
TEO
20
view dependent
configuration sensitive
Layer 3
combinations of features
V4
8.0
V2
3.2
larger receptive fields
Layer 2
V1
1.3
LGN
Layer 1
0
1.3
3.2
8.0 20
50
Eccentricity / deg
Figure 16.12
Convergence in the visual system. Right - as it occurs in the brain. V1: visual cortex
area V1; TEO: posterior inferior temporal cortex; TE: inferior temporal cortex (IT).
Left - as implemented in VisNet. Convergence through the network is designed to
provide fourth layer neurons with information from across the entire input retina.
16.5
Visual stimulus-reward association, emotion,
and motivation
Learning about which visual and other stimuli in the environment are rewarding pun-
ishing, or neutral is crucial for survival. For example, it takes just one trial to learn
if a seen object is hot when we touch it, and associating that visual stimulus with the
pain may help us to avoid serious injury in the future. Similarly, if we are given a
new food which has an excellent taste, we can learn in one trial to associate the sight
of it with its taste, so that we can select it in future. In these examples, the previously
neutral visual stimuli become conditioned reinforcers by their association with a pri-
mary (unlearned) reinforcer such as taste or pain. Our examples show that learning
about which stimuli are rewards and punishments is very important in the control
of motivational behaviour such as feeding and drinking, and in emotional such as
fear and pleasure. The type of learning involved is pattern association, between the
conditioned and the unconditioned stimulus. This type of learning provides a major
example of how the visual representations provided by the inferior temporal visual
cortex are used by the other parts of the brain [77, 80, 82]. In this section we con-
sider where in sensory processing this stimulus-reinforcement association learning
occurs, which brain structures are involved in this type of learning, how the neuronal
networks for pattern association learning may actually be implemented in these re-
gions, and how the distributed representation about objects provided by the inferior
temporal cortex output is suitable for this pattern association learning.
The crux of the answer to the last question is that the inferior temporal cortex
representation is ideal for this pattern association learning because it is a transform-
invariant representation of objects, and because the code can be read by a neuronal
Search WWH ::
Custom Search