Page 419 - Deep Learning

402 Notes to Pages 30–32

demonstrating this and other memory systems, a moment in history when the
choice of mental representation might have had a decisive effect on world events
(Spence, 1985). For experimental confirmation of the effectiveness of visual mne-
monics, see, e.g., Bower (1972), Bower and Reitman (1972), Crovitz (1971), Peters
and Levine (1986) and Richardson (1978). The effects of visual mnemonics are
robust enough to have educational applications; see Higbee (1979) and Peters and
Levin (1986). For a wider perspective on imagery beyond mnemonics, see Finke
(1989), Kosslyn (1980) and Richardson (1969).
16. I am exercising a certain amount of poetic license in my portrayal of the tragedy
of the moth. Although everyone agrees that (some species of) moths fly toward
lights (under certain conditions), even when this leads to their demise, there is
no consensus about the mechanism behind this behavior. The most commonly
stated idea is that moths evolved to navigate at night by moonlight and that they
confuse artificial lights with the moon. By trying to keep a constant angle between
the light rays and their flight path, they trick themselves into constant course cor-
rections that result in a spiral path that ends at the light. This view is often stated in
online sources; see, e.g., http://e.wikipedia.org/wiki/Moth#Attraction_to_light.
But, as James K. Adams at Dalton State College points out, there is no experi-
mental proof of this explanation, and it does not explain why some moths fly
straight at a light; see http://butterflies.freeservers.com/moth_light.html.
Field observations are necessary to resolve this puzzle, but there appears to
have been no program of experimental work devoted to ascertaining moth flight
paths under controlled conditions since Hsiao (1972, 1973). Direct neural path-
ways between eyes and the wings should not be taken too literally; the moth ner-
vous system, although small in size, is nevertheless quite complicated. However,
in locusts, photoreceptors that are sensitive to the polarization of light feed into
a central neural complex that in turn has direct connections to the motor centers
that control flight (Homberg, 2004); if so in locusts, perhaps so in moths.
17. See Hebb (1949, 1963) and Bruner (1966). “Growth is characterized by increas-
ing independence of response from the immediate nature of the stimulus.” (Bruner,
1966, p. 5, italics in original).
18. Although abstractions have interested philosophers for a long time, there is no
widely accepted theory of exactly how abstractions are represented. The most
developed candidate is schema theory. The concept of a schema has its origins
in British neuroscience between the world wars but was brought into cog-
nitive psychology by Frederic C. Bartlett, who used it as the cornerstone of a
theory of long-term declarative memory recall (Bartlett, 1932, pp. 198–202). The
concept was picked up or perhaps independently reinvented 30 years later by
D. Rumelhart in psychology and M. Minsky in Artificial Intelligence; see
Thorndyke (1984) for a history of the concept. It has been employed in research
on text comprehension (Schank, 1986) and problem solving (Marshall, 1995), as
well as in social psychology; see Brewer and Nakamura (1984) for a review.
19. Markman (1999).
20. Kirkham (1992), O’Connor (1975) and Walker (1989). See Furth (1968, 1969) for
the claim that Piaget’s theory of knowledge offers a concept of representation that
resolves some of the classical issues with respect to representational theories of
knowledge.