20-09-2017, 02:45 PM
Case-based reasoning (CBR) widely interpreted. It is the process of solving new problems based on solutions of similar previous problems. An auto mechanic fixing a motor recalling another car that exhibited similar symptoms is using case-based reasoning. An advocate advocating a particular outcome in a lawsuit based on legal precedents or a judge creating the case law is using case-based reasoning. So too, an engineer copying elements of nature (practicing the biomimetic), is treating nature as a database of solutions to problems. Case-based reasoning is a prominent type of elaboration of an analog solution.
It has been argued that case-based reasoning is not only a powerful method for computer reasoning, but also omnipresent behavior in solving daily human problems; or, more radically, that all reasoning is based on past cases personally experienced. This view is related to prototype theory, which is more deeply explored in cognitive science.
At first glance, CBR may seem similar to the rule induction algorithms of automatic learning. Like a rule-inducing algorithm, CBR begins with a set of training cases or examples; forms generalizations of these examples, although implicit, identifying the points in common between a recovered case and the objective problem.
If for example a simple pancake procedure is assigned to blueberry pancakes, the decision is made to use the same basic method of whipping and frying, thus generalizing implicitly the set of situations under which the whipping and frying method can be used. The key difference, however, between the implicit generalization in CBR and generalization in rule induction lies in generalization when done. A rule-inducing algorithm extracts its generalizations from a set of training examples before the objective problem is known; that is, it performs an impatient generalization.
For example, if a rule-inducing algorithm were to recite simple pancake recipes, Dutch apple pancakes and banana pancakes as training examples, it would have to derive at the time of training a set of general rules for making all kinds of pancakes . It would not be until the time of the test that would be given, for example, the task of cooking blueberry pancakes. The difficulty for the rules induction algorithm is to anticipate the different directions in which you should try to generalise your training examples. This contrasts with the CBR, which delays the (implicit) generalization of their cases to trial time - a lazy generalization strategy. In the pancake example, CBR has already given the objective problem of cooking blueberry pancakes; so you can generalise your cases exactly as needed to cover this situation. CBR therefore tends to be a good approach for rich and complex domains in which there are thousands of ways to generalise a case.
It has been argued that case-based reasoning is not only a powerful method for computer reasoning, but also omnipresent behavior in solving daily human problems; or, more radically, that all reasoning is based on past cases personally experienced. This view is related to prototype theory, which is more deeply explored in cognitive science.
At first glance, CBR may seem similar to the rule induction algorithms of automatic learning. Like a rule-inducing algorithm, CBR begins with a set of training cases or examples; forms generalizations of these examples, although implicit, identifying the points in common between a recovered case and the objective problem.
If for example a simple pancake procedure is assigned to blueberry pancakes, the decision is made to use the same basic method of whipping and frying, thus generalizing implicitly the set of situations under which the whipping and frying method can be used. The key difference, however, between the implicit generalization in CBR and generalization in rule induction lies in generalization when done. A rule-inducing algorithm extracts its generalizations from a set of training examples before the objective problem is known; that is, it performs an impatient generalization.
For example, if a rule-inducing algorithm were to recite simple pancake recipes, Dutch apple pancakes and banana pancakes as training examples, it would have to derive at the time of training a set of general rules for making all kinds of pancakes . It would not be until the time of the test that would be given, for example, the task of cooking blueberry pancakes. The difficulty for the rules induction algorithm is to anticipate the different directions in which you should try to generalise your training examples. This contrasts with the CBR, which delays the (implicit) generalization of their cases to trial time - a lazy generalization strategy. In the pancake example, CBR has already given the objective problem of cooking blueberry pancakes; so you can generalise your cases exactly as needed to cover this situation. CBR therefore tends to be a good approach for rich and complex domains in which there are thousands of ways to generalise a case.