23.02.2015 Views

Machine Learning - DISCo

Machine Learning - DISCo

Machine Learning - DISCo

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

What is interesting about this chess-learning task is that humans appear to<br />

learn such target concepts from just a handful of training examples! In fact, after<br />

considering only the single example shown in Figure 11.1, most people would<br />

be willing to suggest a general hypothesis for the target concept, such as "board<br />

positions in which the black king and queen are simultaneously attacked," and<br />

would not even consider the (equally consistent) hypothesis "board positions in<br />

which four white pawns are still in their original locations." How is it that humans<br />

can generalize so successfully from just this one example?<br />

The answer appears to be that people rely heavily on explaining, or analyzing,<br />

the training example in terms of their prior knowledge about the legal moves<br />

of chess. If asked to explain why the training example of Figure 11.1 is a positive<br />

example of "positions in which the queen will be lost in two moves," most people<br />

would give an explanation similar to the following: "Because white's knight is<br />

attacking both the king and queen, black must move out of check, thereby allowing<br />

the knight to capture the queen." The importance of such explanations is<br />

that they provide the information needed to rationally generalize from the details<br />

of the training example to a correct general hypothesis. Features of the training<br />

example that are mentioned by the explanation (e.g., the position of the white<br />

knight, black king, and black queen) are relevant to the target concept and should<br />

be included in the general hypothesis. In contrast, features of the example that are<br />

not mentioned by the explanation (e.g., the fact that there are six black pawns on<br />

the board) can be assumed to be irrelevant details.<br />

What exactly is the prior knowledge needed by a learner to construct the<br />

explanation in this chess example? It is simply knowledge about the legal rules of<br />

chess: knowledge of which moves are legal for the knight and other pieces, the fact<br />

that players must alternate moves in the game, and the fact that to win the game one<br />

player must capture his opponent's king. Note that given just this prior knowledge<br />

it is possible in principle to calculate the optimal chess move for any board<br />

position. However, in practice this calculation can be frustratingly complex and<br />

despite the fact that we humans ourselves possess this complete, perfect knowledge<br />

of chess, we remain unable to play the game optimally. As a result, much of human<br />

learning in chess (and in other search-intensive problems such as scheduling and<br />

planning) involves a long process of uncovering the consequences of our prior<br />

knowledge, guided by specific training examples encountered as we play the game.<br />

This chapter describes learning algorithms that automatically construct and<br />

learn from such explanations. In the remainder of this section we define more<br />

precisely the analytical learning problem. The next section presents a particular<br />

explanation-based learning algorithm called PROLOG-EBG. Subsequent sections<br />

then examine the general properties of this algorithm and its relationship to inductive<br />

learning algorithms discussed in other chapters. The final section describes<br />

the application of explanation-based learning to improving performance at large<br />

state-space search problems. In this chapter we consider the special case in which<br />

explanations are generated from prior knowledge that is perfectly correct, as it is<br />

for us humans in the above chess example. In Chapter 12 we consider the more<br />

general case of learning when prior knowledge is only approximately correct.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!