《Machine Learning（Tom M. Mitchell）》读书笔记——12、第十一章

1. Introduction (about machine learning)

2. Concept Learning and the General-to-Specific Ordering

3. Decision Tree Learning

4. Artificial Neural Networks

5. Evaluating Hypotheses

6. Bayesian Learning

7. Computational Learning Theory

8. Instance-Based Learning

9. Genetic Algorithms

10. Learning Sets of Rules

11. Analytical Learning

12. Combining Inductive and Analytical Learning

13. Reinforcement Learning

11. Analytical Learning

Inductive learning methods such as neural network and decision tree learning require a certain number of training examples to achieve a given level of generalization accuracy, as reflected in the theoretical bounds and experimental results discussed in earlier chapters. Analytical learning uses prior knowledge and deductive reasoning to augment the information provided by the training examples, so that it is not subject to these same bounds. This chapter considers an analytical learning method called explanation-based learning (EBL). In explanation-based learning, prior knowledge is used to analyze, or explain, how each observed training example satisfies the target concept. This explanation is then used to distinguish the relevant features of the training example from the irrelevant, so that examples can be generalized based on logical rather than statistical reasoning.

11.1 INTRODUCTION 

Previous chapters have considered a variety of inductive learning methods: that is, methods that generalize from observed training examples by identifying features that empirically distinguish positive from negative training examples. Decision tree learning, neural network learning, inductive logic programming, and genetic algorithms are all examples of inductive methods that operate in this fashion. The key practical limit on these inductive learners is that they perform poorly when insufficient data is available. In fact, as discussed in Chapter 7, theoretical analysis shows that there are fundamental bounds on the accuracy that can be achieved when learning inductively from a given number of training examples.

Can we develop learning methods that are not subject to these fundamental bounds on learning accuracy imposed by the amount of training data available? Yes, if we are willing to reconsider the formulation of the learning problem itself. One way is to develop learning algorithms that accept explicit prior knowledge as an input, in addition to the input training data. Explanation-based learning is one such approach. It uses prior knowledge to analyze, or explain, each training example in order to infer which example features are relevant to the target function and which are irrelevant. These explanations enable it to generalize more accurately than inductive systems that rely on the data alone.

As we saw in the previous chapter, inductive logic programming systems such as CIGOL also use prior background knowledge to guide learning. However, they use their background knowledge to infer features that augment the input descriptions of instances, thereby increasing the complexity of the hypothesis space to be searched. In contrast, explanation-based learning uses prior knowledge to reduce the complexity of the hypothesis space to be searched, thereby reducing sample complexity and improving generalization accuracy of the learner.

11.1.1 Inductive and Analytical Learning Problems

The full definition of this analytical learning task is given in Table 11 .1

11.2 LEARNING WITH PERFECT DOMAIN THEORIES: PROLOG-EBG

As stated earlier, in this chapter we consider explanation-based learning from domain theories that are perfect, that is, domain theories that are correct and complete(正确的和完备的). A domain theory is said to be correct if each of its assertions is a truthful statement about the world. A domain theory is said to be complete with respect to a given target concept and instance space, if the domain theory covers every positive example in the instance space. 

EXPLAIN THE TRAINING EXAMPLE:

ANALYZE THE EXPLANATION:

By collecting just the features mentioned in the leaf nodes of the explanation in Figure 11.2 and substituting variables x and y for Objl and Obj2, we can form a general rule that is justified by the domain theory: SafeToStack(x, y) <-- Volume(x, 2) ^ Density(x, 0.3) ^ Type(y, Endtable), The body of the above rule includes each leaf node in the proof tree, except for the leaf nodes "Equal(0.6, times(2,0.3)" and "LessThan(0.6,5)." We omit(遗漏; 省略; 删除) these two because they are by definition always satisfied, independent of x and y.

Although this explanation was formed to cover the observed training example, the same explanation will apply to any instance that matches this general rule.

The above rule constitutes a significant generalization of the training example, because it omits many properties of the example (e.g., the Color of the two objects) that are irrelevant to the target concept. However, an even more general rule can be obtained by more careful analysis of the explanation.

PROLOG-EBG computes the most general rule that can be justified by the explanation, by computing the weakest preimage of the explanation, defined as follows: Definition: The weakest preimage(最弱前像) of a conclusion C with respect to a proof P is the most general set of initial assertions A, such that A entails C according to P.

The heart of the regression procedure is the algorithm that at each step regresses the current frontier of expressions through a single Horn clause from the domain theory. This algorithm is described and illustrated in Table 11.3. The final Horn clause rule output by PROLOG-EBGis  formulated as follows: The clause body is defined to be the weakest preconditions calculated by the above procedure. The clause head is the target concept itself, with each substitution from each regression step (i.e., the substitution Oh[ in Table 11.3) applied to it.

REFINE THE CURRENT HYPOTHESIS:

...

11.3 REMARKS ON EXPLANATION-BASED LEARNING

11.3.1 Discovering New Feature

One interesting capability of PROLOG-EBGis  its ability to formulate new features that are not explicit in the description of the training examples, but that are needed to describe the general rule underlying the training example. Notice this learned "feature" is similar in kind to the types of features represented by the hidden units of neural networks; that is, this feature is one of a very large set of potential features that can be computed from the available instance attributes.

11.3.2 Deductive Learning

It is interesting to compare the PROLOG-EBG learning setting to the setting for inductive logic programming (ILP) discussed in Chapter 10. ILP is an inductive learning system, whereas PROLOG-EBG is deductive.

11.3.3 Inductive Bias in Explanation-Based Learning

Approximate inductive bias of PROLOG-EBG: The domain theory B, plus a preference for small sets of maximally general Horn clauses.

11.3.4 Knowledge Level Learning

 deductive closure: 演绎闭包

11.4 EXPLANATION-BASED LEARNING OF SEARCH CONTROL KNOWLEDGE(搜索控制知识的基于解释的学习)

As noted above, the practical applicability of the PROLOG-EBG algorithm is restricted by its requirement that the domain theory be correct and complete. One important class of learning problems where this requirement is easily satisfied is learning to speed up complex search programs. In fact, the largest scale attempts to apply explanation-based learning have addressed the problem of learning to control search, or what is sometimes called "speedup" learning.

In such problems the definitions of the legal search operators, together with the definition of the search objective, provide a complete and correct domain theory for learning search control knowledge. 

An example of a rule learned by PRODIGYfo r this target concept in a simple block-stacking problem domain is

In fact, there are significant practical problems with applying EBL to learning search control. First, in many cases the number of control rules that must be learned is very large (e.g., many thousands of rules). As the system learns more and more control rules to improve its search, it must pay a larger and larger cost at each step to match this set of rules against the current search state. Note this problem is not specific to explanation-based learning; it will occur for any system that represents its learned knowledge by a growing set of rules. A second practical problem with applying explanation-based learning to learning search control is that in many cases it is intractable even to construct the explanations for the desired target concept.

11.5 SUMMARY AND FURTHER READING

In contrast to purely inductive learning methods that seek a hypothesis to fit the training data, purely analytical learning methods seek a hypothesis that fits the learner's prior knowledge and covers the training examples. Humans often make use of prior knowledge to guide the formation of new hypotheses. This chapter examines purely analytical learning methods. The next chapter examines combined inductive-analytical learning.

Explanation-based learning is a form of analytical learning in which the learner processes each novel training example by (1) explaining the observed target value for this example in terms of the domain theory, (2) analyzing this explanation to determine the general conditions under which the explanation holds, and (3) refining its hypothesis to incorporate these general conditions.

PROLOG-EBG is  an explanation-based learning algorithm that uses first-order Horn clauses to represent both its domain theory and its learned hypotheses. In PROLOG-EBG an  explanation is a PROLOG proof, and the hypothesis extracted from the explanation is the weakest preimage of this proof. As a result, the hypotheses output by PROLOG-EBGfo llow deductively from its domain theory.

 Analytical learning methods such as PROLOG-EBG construct useful intermediate features as a side effect of analyzing individual training examples. This analytical approach to feature generation complements the statistically based generation of intermediate features (eg., hidden unit features) in inductive methods such as BACKPROPAGATION.

Although PROLOG-EBG does not produce hypotheses that extend the deductive closure of its domain theory, other deductive learning procedures can. For example, a domain theory containing determination assertions (e.g., "nationality determines language") can be used together with observed data to deductively infer hypotheses that go beyond the deductive closure of the domain theory.

One important class of problems for which a correct and complete domain theory can be found is the class of large state-space search problems. Systems such as PRODIGY and SOAR have demonstrated the utility of explanation-based learning methods for automatically acquiring effective search control knowledge that speeds up problem solving in subsequent cases. 

Despite the apparent usefulness of explanation-based learning methods in humans, purely deductive implementations such as PROLOG-EBG suffer the disadvantage that the output hypothesis is only as correct as the domain theory. In the next chapter we examine approaches that combine inductive and analytical learning methods in order to learn effectively from imperfect domain theories and limited training data.