'Memo' Functions and Machine Learning"

From Nature, Vol 218, April 6 1968.

A Summary by Robin Popplestone

February 1998


Michie begins by setting the idea in the context of machine-learning, with particular reference to Samuel's checkers program. The most definitive text is

Michie elaborates on (4),

Which is, I think close to the current understanding of memoisation, but adds

So, from Michie's perspective, something like data-mining (at least as incorporated in Clementine) is also an instance of memoisation.

There follows a discussion of a humanised scenario in which a person, needing, on call, to evaluate a mathematical function (hcf) from a rule could improve his performance by constructing a card-index (the rote).

He then discusses the effect of integrating a rote into the evaluation of recursively defined functions.

[there are in fact two syntax-errors here - the "if" before the "not" is erroneous - the "else if" should be "elseif". In POP-2 "n.isinteger" was a syntactic variant of isinteger(n). "close" means "end if".]

Michie continues:

[nonop in POP-2 was the way of passing an infix as a parameter, and the "->" notation is an assignment value -> variable].