Mitchell, T.: 1977, Version spaces: A candidate elimination approach to rule learning, in: Proc. of the 5th IJCAI, MIT, Cambridge, MA, pp. 305--310.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....two subsets: the subset covered by the concept and the subset not covered. Any mechanism which produces such blackboxes can be construed as a concept learner. The class of all such mechanisms is large. It encompasses concept learners proper, e.g. symbolist methods such as Candidate Elimination [1, 2], Focussing [3,4] Classification [5] and Conceptual Clustering [6,Fisher, Learning from 7,8,9] It also encompasses mechanisms which do not have concept learning as an explicit goal but which nevertheless produce the requisite black boxes, e.g. connectionist mechanisms such as BackPropagation, ....

Mitchell, T. (1977). Version spaces: a candidate elimination approach to rule learning. Proceedings of the Fifth International Joint Conference on Artificial Intelligence (pp. 305-310). 9

....order to achieve specific knowledge acquisition objectives. 1 Introduction In many knowledge acquisition contexts there will be many classification rules that perform equally well on the training data. In the most clear cut example from machine learning we have the phenomena of the version space [1], the set of all rules that are consistent with a set of training data. However, even when we move away from a situation in which we are expecting to find rules that are strictly consistent with the training data, in other words, when we allow rules to misclassify some training cases, there will ....

Mitchell, T.M.: Version spaces: A candidate elimination approach to rule learning. In: Proceedings of the Fifth International Joint Conference on Artificial Intelligence. (1977) 305--310

....input and output [5] Concept learning can be regarded as search through a space of hypotheses implicitly de ned by the hypothesis representation. A number of algorithms for concept learning are known: the Find S algorithm, the List Then Eliminate algorithm, the Candidate Elimination algorithm [3, 5]. Here Mitchell s List Then Eliminate algorithm is used as the basis for an interactive constraint acquisition system. This algorithm outputs a description of the set of all hypothesis consistent with the training examples. An hypothesis is consistent with the training examples if it correctly ....

Tom Mitchell. Version spaces: A candidate elimination approach to rule learning. In Proceedings of the Fifth International Joint Conference in Arti cial Intelligence, 1977.

....of description logics, have both extensional and intensional parts. The extension is determined concretely from the intensional description and an information ordering is defined over the feature structures. Returning to machine learning, algorithms such as decision trees [5, 6] the version space [7, 8] algorithm and techniques of inductive logic programming [9] formulate intensional descriptions of classes of objects given a number of training examples. These algorithms are intensional because they characterise classes in terms of attribute and predicate expression based descriptions intended ....

T. Mitchell, "Version spaces: A candidate elimination approach to rule learning," in International Joint Conf. on Artificial Intelligence, pp. 305--310, MIT Press, 1979.

....Algorithm 3. Generate Projection Expression . for every i with entry (DNL node d(i) path i ) in the data mapping, create projection expression e i = r d(i) path i ) 11 3. Example Based Learning with the Version Space Model The versions spaces algorithm was first presented by Mitchell in [17]. We build a Version Space model for example based learning adapted to our application. Figure 6 illustrates the sequence of actions in the SPHINX learning algorithm. This algorithm is derived from the original version spaces algorithm, but adapted to handle Sample Selection and Active Learning in ....

....and SchemaLog [11] graph based ontology [18] and XML Query [20] Many XML query systems such as [7,14,21] fundamentally assume that data comes from heterogeneous sources, and can perform integration, provided the user can write view definitions in XML Query. The Version Spaces algorithm ([10,17]) has seen widespread use in machine learning applications. To our knowledge, we are the first to exploit it in the context of databases. 9. Conclusion We built a Version Space model for query discovery by example and developed the SPHINX learning algorithm, by adding a new kind of label and a ....

Mitchell T.M.: Version Spaces: A Candidate Elimination Approach to Rule Learning. IJCAI 1977: 305-310

....occur in one class but occur in the other class with only a few times (1 or 2) They are interesting as they can overcome problems caused by noisy data. Note that the patterns covered by epLp ; epRp and epLn ; epRn are significantly different from patterns covered by a version space (Mitchell, 1977; Mitchell, 1982; Hirsh, 1994) where each pattern 8 Table 3: Data reduction in the mushroom data set after intersecting with an instance. mushroom data volume dimension poisonous edible original 3525 3788 22 after reduction 9 7 11 must have a 100 frequency in the positive data. However, our ....

Mitchell, T. (1977). Version spaces: A candidate elimination approach to rule learning. Proceedings of the Fifth International Joint Conference on Artificial Intelligence (pp. 305--310). Cambridge, MA.

....particular, M82] presents a unifying approach to the problem of generalizing knowledge (that, for example, can be represented as rules) by viewing generalization as a search problem. Moreover, based on the specific search methods used, Mitchell [M82] also categorizes several rule learning systems [BM78, P70, W75, HRM7, V78, M77, MUB82] that deal with generalization. In particular, M82] deals with a broader set of objects (that can also include rules) and formulates the generalization problem as follows. Given a set of instances specified in an instance language, the generalization problem is formulated in [M82] as a search for ....

Mitchell, T.M., 1977. Version Spaces: A Candidate Elimination Approach to Rule Learning. In Proceedings of International Joint Conference on Artificial Intelligence-1977., pp. 305-310.

....preconditions that produce each conditional effect from different action observations. While learning, it is likely there will not be enough information to uniquely establish a set of preconditions for a given effect. To represent this uncertainty, the preconditions are stored in a version space (Mitchell 1977) that maintains sets of preconditions describing the most specific and general possible preconditions proven. This makes explicit both the known restrictions on what the preconditions may be as well as the remaining uncertainty at any point during learning. Linguistic Knowledge The intelligent ....

Mitchell, T. M. 1977. Version Spaces: A Candidate Elimination Approach to Rule Learning. Procedings of the 5 th International Joint Conference on Artificial Intelligence, Cambridge, MA, 305310.

....difficulties with them. Ultimately it appears that other languages will be more useful for inductive inference. Some alternatives are presented in [27] However the algebra of rational term expressions turns out to be useful in providing an algebraic formulation of Mitchell s version space method [30, 31] on the lattice of rational terms. 2 A critical decision problem is to determine whether a rational term expression is equivalent to a purely disjunctive rational term expression. This problem arises from a wish to infer disjunctive concepts. We show that the class of sets definable by ....

.... Gamma f(y; a) This is equivalent to f(a; b) f(a; c) f(a; f(x; y) Finally, the fifth technique produces the term expression x Gamma (f(y; a) Gamma f(a; z) which is equivalent to a b c f(a; a) f(y; b) f(y; c) f(y; f(u; v) 3 A second approach to this problem was developed by Mitchell [30, 31] in his work on version spaces and Young et al. 39] in their work on description spaces. This approach represents the set of all possible solutions, that is, all term expressions u describing a set U such that E U and C U = In this approach, the set of all solutions is represented by the ....

[Article contains additional citation context not shown here]

T.M. Mitchell, Version Spaces: A Candidate Elimination Approach to Rule Learning, Proceedings of IJCAI-77, 305--310, 1977.

....at the centre of the information flow, is so obvious but often overlooked in intelligent systems. There followed a series of incremental innovations as we came to better understand the value of FCA. The formal concept lattice will remind readers with an AI background of Mitchell s version spaces [20]. The space complexity of the lattice is similar and in general lattice completion is O(2 n ) where n is number of formal concepts. There are a number of ways of reducing the complexity of the lattice completion algorithm from O(2 n ) These are the subject of another paper [4] but involve the ....

T. Mitchell. Version spaces: A candidate elimination approach to rule learning. In Proceedings of Fifth International Conference on Artificial Intelligence. IJCAI, 1977.

....say production rules or a decision tree, that distinguish instances of each class from other classes. Attribute based induction algorithms (such as ID3 [18] C4.5 [19] AQ, CN2 [5] AE1 [12] and HCV [26] and incremental induction algorithms (such as ID5R [24] and the version space method [17]) fall into the supervised classification category. Unsupervised clustering (or concept formation [13] deals with discovery of new concepts from unclassified data. The data input for clustering is similar to that for classification, but the significant difference is that no class information is ....

T. Mitchell, Version Spaces: A Candidate Elimination Approach to Rule Learning, Proceedings of the Fifth International Joint Conference on Artificial Intelligence, Cambridge, Mass., 1977.

.... hypothesis selection [Mitchell 1980, Utgoff 1986] In the second meaning, bias refers to restrictions imposed on parts of the learning system, e.g. restrictions of the hypothesis language as in [De Raedt 1991] Thirdly, bias is viewed as the core of induction because it triggers the inductive leap [Mitchell 1977]. Without bias, a hypothesis need not to be a generalization of the input examples. This contradicts the common definition of induction. These different meanings cause a lot of confusion and discussion of what bias is and how it should be defined. In this paper, we give a more precise definition ....

....collection of all positive examples is sufficient to be accepted as hypothesis. Obviously, such a hypothesis lacks the predictive power as it will fail on explaining further positive examples. Given this definition, there is even no reason for an inductive system to make an inductive leap at all. [Mitchell 1977] was aware of this problem, and argued that generalization is due to the bias of the system. In retrospect, it is not surprising that an unbiased generalization system cannot make classifications of instances other than the training instances. An unbiased system is one whose inferences ....

Mitchell, T. (1977). Version spaces: A candidate elimination approach to rule learning. In Proc. of 5th International Joint Conference on AI. Morgan Kaufmann.

....include classification [Qui86] associations [AS94, HS95] clustering [Fis95] and sequential patterns [MTV95] The process is often very slow, particularly when databases are large. The success of the data mining process is critically dependent upon the availability of user insights and biases[Mit77] even though the process may use unsupervised learning algorithms [Lan96] User insights and biases includes abstract preferences for attributes and attribute sets that reflect the user s interests and purposes. It could also include more detailed guidance in terms of preferences on the partial ....

T. M. Mitchell. Version spaces: A candidate elimination approach to rule learning. In Proc. of IJCAI 77, 1977.

....possible to start with the most general case and, using a specialization operator, add or constrain the rule conditions. These approaches generally rely on the fact that pre classified training instances are (usually) partially ordered according to generality. The candidate elimination algorithm (Mitchell 1977) employs both methods to conduct a bidirectional exhaustive search to identify the conditions for classification rules. Unfortunately, the algorithm assumes that a single, conjunctive rule can describe each class and that the training set is free from noise. To compensate, another methodology (for ....

Mitchell, T. M. (1977), Version spaces: A candidate elimination approach to rule learning, in `Proceedings of the Fifth International Joint Conference on Artificial Intelligence', San-Mateo, California, pp. 305--310.

.... addition to traditional classification methods such as linear regression and logistic regression, several dozen classifier construction algorithms have been developed in the last few decades in the machine learning community, including various versions of perceptron (Nilsson 1965) version space (Mitchell 1977), decision tree (Quinlan 1986) instance based (Duda Hart 1973) and neural net al..gorithms (Rumelhart McClelland 1986) The results of empirical comparisons of existing algorithms illustrate that each algorithm has a selective superiority: it is best for some but not all classification tasks ....

Mitchell, T. M. 1977. Version spaces: A candidate elimination approach to rule learning. In Proceedings of the Fifth International Joint Conference on Artificial Intelligence, 305--310, Morgan Kaufmann.

No context found.

Mitchell, T.: 1977, Version spaces: A candidate elimination approach to rule learning, in: Proc. of the 5th IJCAI, MIT, Cambridge, MA, pp. 305--310.

No context found.

Mitchell, T.M. #1977#. Version spaces: A candidate elimination approach to rule learning. In: Proceedings of the Fifth International Joint Conference on Arti#cial Intelligence #IJCAI-77#. IJCAII. Cambridge, Massachusetts. pp. 305#310.

No context found.

T. Mitchell, "Version spaces: a candidate elimination approach to rule learning". In proceedings of I.J.C.A.I., vol. 5, pp 305-310, 1977.

No context found.

T. M. Mitchell. Version spaces: A candidate elimination approach to rule learning. In Proc. 5th IJCAI, pages 305--310, 1977.

No context found.

T. M. Mitchell. Version spaces: A candidate elimination approach to rule learning. In Proc. 5th IJCAI, pages 305--310, 1977.

No context found.

T.M. Mitchell. Version spaces: A candidate elimination approach to rule learning. In IJCAI-77, pages 305--310, Cambridge, MA, 1977.

No context found.

T. Mitchell. Version spaces: A candidate elimination approach to rule learning. IJCAI, pages 305--310, 1977.

No context found.

T. Mitchell. Version spaces: A candidate elimination approach to rule learning. IJCAI, pages 305--310, 1977.

No context found.

Mitchell, T. (1977). Version spaces: a candidate elimination approach to rule learning. Proceedings of the Fifth International Joint Conference on Artificial Intelligence (pp. 305-310).

No context found.

Mitchell, Tom, 'Version Spaces: A Candidate Elim- ination Approach to Rule Learning,' in I.ICAL77, 1977

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC