Académique Documents
Professionnel Documents
Culture Documents
Neurocomputing.
James Anderson and Edward Rosenfeld, editors.
1989 The MIT Press.
is provided in screen-viewable form for personal use only by members
of MIT CogNet.
Unauthorized use or dissemination of this information is expressly
forbidden.
If you have any questions about this material, please contact
cognetadmin@cognet.mit.edu.
4
Introduction
(1949)
DonaldO. Hebb
The Organization of Behavior, New York : Wiley , Introduction and Chapter 4,
"
" The first
stage of perception : growth of the assembly, pp . xi - xix , 60- 78
The introduction is also notable becausein it is one of the first usesof the word
" connectionisrn" in the context of a
cornplex brain rnodel. The final paragraph of the
"
The
these
lines:
Introduction contains
theory is evidently a fonn of connectionisrn,
one of the switchboard variety, though it does not deal in direct connectionsbetween
'
afferentand efferentpathways: not an ' S- R psychology, if R rneansa muscularresponse.
The connectionsserverather to establish autonornous central activities, which then
are the basisof further learning" (p. xix). Most rnodern day connectionistscould find
little to argue with in that surnrnation.
A rnore detailed description of Hebb' s physiological ideas is found in chapter 4. In
this chapter, Hebb has a detailed discussionof neurophysiology and neuroanatorny
as it relatesto his ideas. It is worth ernphasizing, if it is not obvious at this point , that
. Jarnes, McCulloch , and Hebb were
the early rnodelersreally knew their neuroscience
and
usedtheir knowledgeextensively
nervous
about
the
systern,
highly knowledgable
in their rnodels. Much rnodern work in neural networks has rnoved far away frorn its
roots in the study of the brain and psychology. This is a causefor concern, both because
the field is losing contact with its foundations and becauseit has lost a source of
valuable ideas.
Hebb suggestsseveralirnportant ideasin chapter 4. First , and rnost farnous, was the
'
"
"
clear staternentof what has becorneknown as the Hebb synapse. To restateHebb s
"
description, for the nth tirne, Whenan axon of cell A is near enoughto excite a cell B
and repeatedlyor persistentlytakespart in firing it , somegrowth processor metabolic
'
changetakesplacein oneor both cells suchthat A s efficiency, as oneof the cellsfiring
'
"
B, is increased (p. 50). This, like the other ideasin Hebb s book, is not a rnathernatical
staternent, though it is closeto one. For exarnple, Hebb does not discussthe various
possibleways inhibition rnight enter the picture, or the quantitative learning rule that
is being followed. This has rneant that a nurnber of sornetirnesquite different learning
." (Paper 6, an early cornputer sirnularules can legitirnately be called " Hebb synapses
tion of Hebb' s ideas, discusses the rnodifications one rnust rnake to this bare outline
to rnake the systernwork.)
44
Cbapter4
Second, Hebb is keenly aware of the " distributed" nature of the representationhe
is assumingthe nervous systemuses. The idea is that to representsomething, many
cells must participate in the representation. Hebb was aware of the work of Lashley
(paper 5), suggestingwidely distributed representations, and made some use of his
ideas, though not in the strongestform of complete " equipotentiality ."
"
Third , Hebb postulated the formation of what he called " cell assemblies
, which
were really the heart of the entire book. The basic idea was that there were interconnected
, self-reinforcing subsetsof neuronsthat formed the representationsof information
in the nervous system. Single cells might belong to more than one assembly,
depending on the context. Multiple cell assembliescould be active at once, corresponding
to complex perceptionsor thoughts. There was a distributed representation
at the functional level as well as at the anatomical level. Hebb devotesa good deal of
attention to the details of the neuroanatomy and physiology that might underlie cell
assemblies
. The later chapters in the book contain many discussionsof how cell
assembliescan be usedto help explain a number of psychologicalphenomena.
In retrospect, the idea that there exist temporarily stable, relatively long lasting
neural activity patterns that are important in mental activity has reappearedin the
various " attractor " models for brain activity (see, for example, Hopfield, paper 27;
Grossberg, paper 24; or Anderson et al., paper 22). Details of the observed and
predicted stability depend critically on learning assumptionsthat are nearly always
basedto somedegreeon Hebb synapses
. Someof the ideasdescribedin this book have
becomepart of the acceptedlore of the field.
(1949)
DonaldO. Hebb
The Organization of Behavior, New York : Wiley , Introduction and Chapter 4,
" The first
"
stage of perception : growth of the assembly, pp . xi - xix , 60- 78
Introd uction
46
Chapter 4
47
Hebb1949
48
Chapter
49
Hebb1949
50
Chapter4
'
7
r 18
51
Hebb1949
from Area 17
Conduction
'B
A
'
C
52
Chapter4
-."
53
Hebb1949
A
I..X
.
b
Figure8 CellsA andB lie in a regionof area17(shownby
excitedby an afferentstimulation.C
) whichis massively
hatching
is a cellin area18whichleadsbackinto 17. E is in area17but lies
outsidetheregionof activity. Seetext.
S4
Chapter 4
~
1
7.1~ .
"
"
Figure 10 Arrows representa simple assembly of neural
pathwaysor open multiple chains firing according to the numbers
on each(the pathway " I , 4" fires first and fourth , and so on),
"
"
illustrating the possibility of an alternating reverberation which
would not extinguish as readily as that in a simple closedcircuit.
Statistical Considerstions
55
Hebb1949
S6
Chapter4