You are on page 1of 4

538 Perception: The Ecological Approach

Perception: The Ecological Introductory article


Approach
M T Turvey, University of Connecticut, Storrs, Connecticut, USA

CONTENTS
Gibson's Contribution Affordances
Optic Flow and the Visual Guidance of Locomotion

The ecological approach to perception is a theoret- inference, about what caused the input in the first
ical perspective on how animals, including humans, place. This traditional view is labeled as indirect
can be aware of their surroundings. It emphasizes perception because the animal's awareness of the
the relevance of activity to defining the environment world is a result of these intermediary steps. A
to be perceived.
theory of direct perception, in contrast, argues
that the intermediary steps are only needed if the
scientist has described the input incorrectly. In-
INTRODUCTION cluding the environment and activity in the theory
Perception refers to how animals, including of perception allows a better description of the
humans, can be aware of their surroundings. The input, a description that shows the input to be
ecological approach to perception refers to a par- richly structured by the environment and the
ticular idea of how perception works and how it animal's own activities. This means that the inter-
should be studied. The label `ecological' reflects mediary steps are not needed and perception is
two main themes that distinguish this approach direct.
from the establishment view. First, perception is
an achievement of animal±environment systems,
not simply animals (or their brains). What makes
GIBSON'S CONTRIBUTION
up the environment of a particular animal ± cliffs, The ecological approach to perception originated
caves or crowds ± is part of this theory of percep- in the work of the American psychologist James
tion. Second, perception's main purpose is to guide J. Gibson (1904±1979). Two biographical facts
activity, so a theory of perception cannot ignore were important to shaping his theory. As the
what animals do. The kinds of activities that a young son of a railroad employee in America's
particular animal does ± how it eats, moves and Midwest, he had spent many hours on trains
mates ± are part of this theory of perception. watching the world flow by. He noticed that from
Including the environment and behavior as im- a vantage point in the locomotive looking forwards
portant parts of perceptual theory, rather than as the flow was outwards; from the caboose looking
afterthoughts, is clearly different from theories that rearwards the flow was inwards. Here was the seed
start inside the eye (or the ear or the skin); but it is of his notion that the light that comes to our eyes is
not necessarily controversial. None the less, the reliably structured by activity ± structured light can
ecological approach is considered controversial be- be rich and meaningful. Many years later, after he
cause of one central claim: perception is direct. To had established a career as a perceptual psycholo-
understand the claim, and why some might con- gist, Gibson took time away from his college teach-
sider it troubling, we have to contrast it with the ing position to spend 4 years as a scientist with the
more traditional view. Most scientists believe that aviation psychology program during the Second
perception begins with faulty input. For example, World War. There he realized that the practical
when objects in the world reflect light, the pattern problems of takeoffs and landings, and pursuit
of light that reaches the back of the eye (the retina) and evasion, which could not only be mastered by
has lost and distorted much of the detail. The job of 18-year-old pilot trainees but also performed rou-
perception, then, becomes one of correcting the tinely by birds and bees, had little to do with the
input and adding meaningful interpretations to it physiology of the eyeball. Here was the seed of his
so that the brain can make an educated guess, an notion that perceptual theory should try to explain
Perception: The Ecological Approach 539

real-world behaviors (and not simply human be- whether any muscle commands had been issued to
haviors). (See Gibson, James J.) move the eyes, the head or the legs. In the absence
Upon resuming his job as a college professor, of counteracting motor commands, object motion is
Gibson set about challenging the assumptions that concluded; in the presence of such commands, the
he thought sat unexamined in most laboratory retinal signals would be counteracted, allowing
work, including his own. He argued that perceivers the alternative conclusion of self-motion. Another
are aware of the world, not of their own sensations, possibility is that the observer is moved passively
and perception theory should respect that. Gib- under somebody else's power (as in a train) so that
son's reformulation of the problem of distance per- other input and even knowledge must be taken into
ception illustrates the essence of what developed account. Gibson suggested an elegant alternative
into his ecological approach. For centuries, scien- solution: overall, or global, change in the pattern
tists believed that `distance is not perceivable by of light is specific to self-motion; local change
eye alone'. Indeed, if objects are treated as isolated against a stationary background is specific to object
points in otherwise empty space, then their dis- motion. This simple insight (echoing the experience
tances on a line projecting to the eye are indistinct: of the young Gibson riding the rails) opened a
each stimulates the same retinal location. Gibson new field of research devoted to uncovering the
dubbed this formulation `air theory' (A in Figure 1) structure in changing patterns of light: optic flow.
and argued that it was inappropriate for ad-
dressing how we see. His alternative was `ground
OPTIC FLOW AND THE VISUAL
theory', which emphasized the contribution of a
continuous background surface to providing rich
GUIDANCE OF LOCOMOTION
visual structure (G in Figure 1). The simple step of Optic flow refers to the patterns of light, structured
acknowledging that points do not float in the air by particular animal±environment settings, avail-
but are attached to a surface introduces a higher- able to a point of observation. The goal of optic flow
order property, the gradient, which opened up the research is to discover particular reliable patterns
new possibility that perception might be veridical, of optical structure, called invariants, relevant to
that is, about facts of the world. guiding activity. Outflow and inflow are distinct
Gibson began to emphasize the enriching role of forms of optic flow (distinct flow morphologies)
movement in perception. Once more, an ecological that tell us whether we are moving forwards or
solution to an old problem is instructive. This prob- backwards. As scientists consider how that flow is
lem concerned how a perceiver could distinguish structured by the variety of clutter that we encoun-
object motion from his or her own motion. The ter as we move around ± doorways, hillsides and
puzzle arises because of the traditional assumption the like ± they discover invariants specific to those
that the cue to perceived motion is the stimulation facts as well (Figure 2).
of successive retinal locations. An object moving In order to effectively guide their activities,
from left to right fixated by a stationary eye will animals need to know more than simply what
stimulate retinal receptors A, B, C, then D. That they are approaching. They also need to know
same object fixated by an eye moving from right how they are approaching ± are they moving too
to left will also stimulate retinal receptors A, B, C, fast? ± and whether they need to adjust that ap-
then D. Since the retinal input is ambiguous, it must proach: should they slow down? Turn? As a busy
be compared with other input having to do with waiter rushes towards the swinging door of the
restaurant kitchen, he makes subtle adjustments
to his behavior in order to control his collision. He
A
needs to maintain enough speed to push through
the door but not so much that he crashes into it.
Effective behavior requires that he should know
when a collision will happen (so he does not slow
down too early) and how hard the collision will be
(so that he slows down enough). Optical structure
G
relevant to these facts has been identified and pro-
Figure 1. The distances of four points in the air (A) are vides examples of quantitative invariants. The op-
indistinct because their projections are onto the same tical quantity called t is specific to when a point of
retinal location. As texture elements on the ground (G), observation will contact an upcoming surface.
the gradient of their projections distinguishes their rela- Consider the rectangular contour in Figure 2(a).
tive depth. As the waiter approaches it, its optical projection
540 Perception: The Ecological Approach

(a) (b) (c) (d)

Figure 2. (a) Global optical expansion, which accompanies forward locomotion, is represented as a velocity vector
field radiating from a focus of expansion. Smooth flow specifies approach to a flat surface. (b) Faster expansion inside a
contour specifies an obstacle. (c) Faster expansion outside a contour specifies an opening. (d) Curvilinear flow specifies
direction of heading.

magnifies. The speed of approach affects the rate


of expansion, that is, the change in optical area
per unit time. The quantity t is given by the inverse
of the relative rate of this expansion ± how long will
it take until there are no units of time left. As he
slows down (or speeds up), the rate at which t
approaches zero changes. The rate of this change
(that is, the derivative of t) is specific to how severe
the collision will be. It essentially quantifies for a
(a) (b)
moving observer (the waiter) whether his kinetic
energy is being reduced (by braking) at a rate suffi- Figure 3. (a) A brink in the path of locomotion presents
cient to stop movement before contact occurs. different behavioral possibilities to animals of different
The preceding descriptions of global optical sizes and locomotory styles. (b) For an animal with legs,
structure refer to situations in which the observer the ratio of flow above the horizontal contour of the brink
is approaching a surface. However, they are also relative to the rate of flow below scales the brink to the
relevant to a surface, such as a projectile, approach- animal's eye height, specifying whether it can be des-
ing the point of observation. Local disturbances of cended safely.
optical structure relevant to the guidance of inter-
ceptive behavior can also be described in terms of t
and its derivative, specific to when and how hard a is `personalized' is addressed in Gibson's notion of
collision will be. Other optical quantities are rele- affordances.
vant to moving the perceiver (or the perceiver's
hand or racket) into a position to intercept (or
avoid) the projectile. Moreover, the same invariants
AFFORDANCES
are available to the family dog catching a ball, a In highlighting the relevance of optical structure
chickadee landing on a bird feeder, and a bumble- to whether an activity is possible (`Is the ball
bee searching for pollen. Although these creatures catchable?') the preceding examples introduce
have obviously different visual systems and brains, what might be the most radical contribution of
the information relevant to guiding their behaviors this theory: behavioral possibilities are perceived.
is the same. Invariants of tissue deformation and Gibson coined the word `affordances' to refer to
sound compression waves reveal the same richness the possibilities for action of a particular animal±
that has been found in optical structure. Of course, environment setting. They are what an arrange-
while all creatures need to perceive things such as ment of surfaces means to an animal. Affordances
openings and obstacles to locomotion, what counts are usually described as `-ables', as in `catchable',
as an opening necessarily differs. How perception `pass throughable', `climbable', and so on. Whether
Perceptron 541

a ledge, for example, is a stepping-down place or a people of different sizes, the functional boundary ±
falling-off place is not determined by its absolute scaled by something like the person's eye height ±
size or shape but how it relates to a particular animal, is the same. If the eye height is manipulated in
including that animal's size and agility, and style some way (e.g. by having the person wear platform
of locomotion (Figure 3(a)). Gibson proposed that shoes or by surreptitiously raising the floor of the
such activity-relevant animal±environment rela- room they view through a window) it has predict-
tions are specified in the optics (Figure 3(b)). able consequences for their affordance evaluations.
Through their research on affordances, Gibson In summary, the theory of affordances is radical
and his followers `ecologize' the traditional prob- because it treats relational properties as objective:
lems of size, distance and shape perception. They the relation between the ledge size and animal size
manipulate aspects of the environment (such as the in Figure 3 exists even when it is not being per-
height of a flat surface, or the distance of a target) ceived. Moreover, such relational properties are
and ask people to evaluate whether a certain be- considered more fundamental to describing en-
havior would be possible. Instead of asking for vironments and understanding perception than
estimates of size or distance in absolute units, the objective primary properties studied so far by
they ask (for example) whether the surface could physics. According to this ecological theory, mean-
be climbed or the target could be reached. Ideally, ings are directly perceptible. The wide-ranging
experimenters would prefer that people's responses species inhabiting our planet, from simplest bacter-
be as natural as possible (that is, without mak- ium to human, do not invent meanings (by concep-
ing conscious decisions). When the technology tion), they discover them (by perception).
allows, therefore, active behavioral adjustments
are recorded. Either way, we have learned that
children, adults and the elderly all perceive possib- Further Reading
ilities for, and restrictions on, activity. They are well Gibson JJ (1986) The Ecological Approach to Visual
aware of behavioral category boundaries (this is Perception. Mahwah, NJ: Erlbaum.
climbable; that is not). Even though the absolute Reed ES (1988) James J. Gibson and the Psychology of
boundary ± measured in centimeters ± differs for Perception. New Haven, CT: Yale University Press.

Perceptron Intermediate article


Simon Haykin, McMaster University, Hamilton, Ontario, Canada

CONTENTS
Introduction Backpropagation algorithm
Perceptron convergence theorem Design principles for multi-layer perceptrons
Least-mean-square algorithm Conclusion

The perceptron is a neural network designed to the free parameters of this neural network first
simulate the ability of the brain to recognize and appeared in a learning procedure developed by
discriminate. Rosenblatt in 1958 for his perceptron brain model.
The generic `perceptron' was originally defined by
Rosenblatt as a set of input units called `S-units'
INTRODUCTION
that are connected to a second set of units called `A-
The single-layer perceptron is the simplest form of a units'. The definition of a `single-layer perceptron'
neural network used for the classification of pat- introduced in this article is different from that of
terns said to be linearly separable (i.e., patterns that Rosenblatt.
lie on opposite sides of a hyperplane). Basically, it If the patterns used to train the perceptron are
consists of a single neuron with adjustable synaptic drawn from two linearly separable classes, then the
weights and bias. The algorithm used to adjust perceptron algorithm converges and positions the

You might also like