The Role of Constraints in Hebbian Learning
Kenneth D. Miller and David J.C. MacKay
Models of unsupervised correlation-based (Hebbian) synaptic plasticity are
typically unstable: either all synapses grow until each reaches the maximum
allowed strength, or all synapses decay to zero strength. A common method
of avoiding these outcomes is to use a constraint that conserves or limits
the total synaptic strength over a cell. We study the dynamical effects of
such constraints. Two methods of enforcing a constraint are distinguished,
multiplicative and subtractive. For otherwise linear learning rules,
multiplicative enforcement of a constraint results in dynamics that converge
to the principal eigenvector of the operator determining unconstrained
synaptic development. Subtractive enforcement, in contrast, leads to a final
state in which almost all synaptic strengths reach either the maximum or
minimum allowed value. This final state may be dominated by weight
configurations other than the principal eigenvector of the unconstrained
operator. Thus, multiplicative constraints yield a ``graded" receptive
field in which all mutually correlated inputs typically retain some
representation, whereas subtractive constraints yield a receptive field that
is ``sharpened" to a few maximally-correlated inputs. If two equivalent
input populations (e.g. two eyes) innervate a common target, multiplicative
constraints prevent their segregation (ocular dominance segregation) when
the two populations are weakly correlated; whereas subtractive constraints
allow segregation under these circumstances. An approach to understanding
constraints over input and over output cells is suggested, and some
biological implementations are discussed.
postscript (UCSF,USA) and postscript (Cambridge,UK) as
appeared in
Neural Computation (1994).
technical report version (USA) and technical report version (UK) (Longer, with a few minor corrections)
@ARTICLE{MM94:nc,
KEY ="",
AUTHOR ="K. D. Miller and D. J. C. MacKay",
TITLE ="The role of constraints in {H}ebbian learning",
JOURNAL ="Neural Computation",
VOLUME ="6",
NUMBER ="1",
PAGES ="98-124",
YEAR ="1994",
ANNOTE ="Date submitted: 9 Oct 1992; Date accepted: 13 May 1993;
Collaborating institutes:
California Institute of Technology"}
David MacKay's:
home page,
publications,
BibTeX file.