Page 448 - Deep Learning

Notes to Pages 197–207 431

62. There are several different computational models of generalization of practical
knowledge; see, e.g., Anderson (1983, 1987), Lewis (1988), Mitchell (1982), Sun,
Merrill and Peterson (2001) and Sun, Sluzarz and Terry (2005).
63. Examples of such learning mechanisms are found in, among other works, Elio
and Scharf (1990), Jones and VanLehn (1994), Larkin (1981), Neches (1987),
Ohlsson (1987b), Ruiz and Newell (1993), Shrager and Siegler (1998) and Siegler
and Araya (2005).
64. Altmann and Bums (2005), Gray and Boehm-Davis (2000), Gray, Shelles and
Sims (2005) and Schooler and Hertzwig (2005). This form of learning was origi-
nally studied by the behaviorists under the label statistical learning theory (Estes,
1950) and the main phenomenon was called probability matching (Grant, Hake &
Hornseth, 1951).
65. Logan (1998).
66. There is a persistent itch among cognitive theorists to reach for a description of
cognition at a level of abstraction above that of process models. Indeed, there
is a need for a type of description that constrains and specifies the system that
a process model is to instantiate and hence explain. Such formulations include
Noam Chomsky’s distinction between linguistic competence and performance
(Chomsky, 1964; Pylyshyn, 1973); Zenon W. Pylyshyn’s distinction between the
functional architecture and cognitive processes (Pylyshyn, 1980, 1986); David Marr’s
distinction between the computational and the algorithmic levels in vision (Marr,
1982); Allen Newell’s distinction between the knowledge level and the symbol level
in the description of an information processing system (Newell, 1982); and John
R. Anderson’s rationality principle, which says that a first approximation model
of human cognition can assume that the latter is maximally efficient (Anderson,
1989, 1990). The Principle of Maximally Efficient Learning is a rationality prin-
ciple of this sort, and it was inspired by these prior formulations but differs from
them in its exclusive focus on skill acquisition.
67. Anderson (1989, 1990) and Newell (1990, p. 33).
68. Ohlsson and Jewett (1997).

Chapter 7. Error Correction: The Specialization Theory
1. Norman (1981, p. 3).
2. Bruner (1970, p. 67).
3. Cavalli-Sforza (2000), Olson (2002) and Stringer and McKie (1997).
4. These two proverbs regarding learning from error appear in European sources
from way back. The Web page http://www.answers.com/topic/a-burnt-child-
dreads-the-fire gives sources as far back as a.d. 1250 for the one about dreading
the fire, and the Web page http://wwww. answers.com/topic/once-bitten-twice-
shy gives multiple 19th-century sources for the second one. The two proverbs
appear to have been fused in the American variant once burned, twice shy.
5. Thorndike (1927).
6. Thorndike (1898, p. 45).
7. James (1890, vol. 1, pp. 24–27).
8. James (1890, vol. 1, p. 25).