Jaime G. Carbonell, editor. Machine Learning: Paradigms and Methods. MIT Press, Boston, MA, 1990.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....problem solving strategies. The studies in machine learning suggest computational algorithms and analyses of such algorithms that suggest explanations for capabilities and limitations of human cognition. Learning can be described as increasing the knowledge or skills in accomplishing certain tasks [13]. The learner applies inferences in order to construct an appropriate representation of some relevant reality. One of the fundamental research problems in machine learning is how to learn from examples since it it usually possible to obtain a set of examples to learn from. From a set of training ....

....the environment and making discoveries (learning from observation and discovery) In this thesis, we will concern with concept acquisition. Concept acquisition can be defined as the task of learning a description of a given concept from a set of examples and counterexamples of that concept [13, 43]. Examples are represented usually by input vectors of feature values and their corresponding class labels. Concept descriptions are then learned as relations among the given set of feature values and the class labels. The ability to classify is another important facet of intelligence. The task ....

[Article contains additional citation context not shown here]

J.G. Carbonell, editor, Machine Learning: Paradigms and Methods, The MIT Press, 1990.

.... examples using rough sets system [10] and algorithms developed by Stefanowski and Slowinski [11] 12] all based on the theory proposed by Pawlak [9] Examples of other algorithms and developments in learning and data mining can be found in the edited volumes by Lin and Cecerone [13] Carbonell [14], Michalski et al. 15] and the book by Mitchell [16] Lim et al. 17] presented an extensive survey of learning algorithms. III. EXAMPLES OF RULE EXTRACTION The example presented below illustrates the rules derived with different rule extraction algorithms. A. Example 1 Consider the data set ....

....= 3) THEN (Post Op Arrhy1 = IART) 1, 50.00 , 100.00 ] 21] Rule 2. IF (Days in ICU = 1) AND (Days before d c in [9, 11] AND (Pump Time = 44) THEN (Post Op Arrhy1 = IART) 1, 50.00 , 100.00 ] 11] Decision: Inducible IART = Y Rule 1. IF (Days in ICU = 2) AND (Days before d c in [14, 22]) THEN (Inducible IART = Y) 3, 20.00 , 100.00 ] 17, 23, 44] Rule 2. IF (Days before d c = 15) THEN (Inducible IART = Y) 3, 20.00 , 100.00 ] 26, 36, 55] Rule 3. IF (Days in ICU = 1) AND (Circ Arrest Time = 20) AND (Pump Time in [64, 70] THEN (Inducible IART = Y) 3, 20.00 , 100.00 ] 8, ....

[Article contains additional citation context not shown here]

J. G. Carbonell, Machine Learning: Paradigms and Methods. Cambridge, MA: MIT Press, 1990.

....Langley and Simon [2] grouped machine learning research into the following categories: a) neural networks; b) genetic algorithms; c) case based learning; d) rule induction; e) analytical learning. The major developments in learning and data mining are summarized in the edited volumes [2] [4], 5] and the book by Mitchell [6] For a survey of important applications of machine learning see [2] Other related topics include nonparametric regression, which is discussed in [7] and [8] and reinforcement learning covered in [9] To date, numerous data mining software products have been ....

J. G. Carbonell, Machine Learning: Paradigms and Methods, J. G. Carbonell, Ed. Cambridge, MA: MIT Press, 1990.

....Auer et al. 13] It minimizes the number of errors and discretizes continuous attributes. LERS: Learning from Examples using Rough Sets System [14] Examples of other algorithms and developments in learning and data mining can be found in the edited volumes by Lin and Cecerone [15] Carbonell [16], and Michalski et al. 17] and the book by Mitchell [3] For a survey of important applications of machine learning see [18] and [19] Most of the developed to date rule extraction algorithms fall into the following classes: 1) Decision tree algorithms, for example ID3 [4] and C4.5 [9] A ....

J. G. Carbonell, Ed., Machine Learning: Paradigms and Methods. Cambridge, MA: MIT Press, 1990.

....X Y Figure 1. Assembling some X and Y with a nut bolt combination. 2 C.T. CHARLTON AND K.M. WALLACE At the core of this method lies inductive learning, in which concept patterns are hypothesised by generalisation and specialisation from particular positive and negative examples of that concept (Carbonell, 1989; Mitchell, 1997) However, we do not use a teacher to label examples with their corresponding concepts. Instead, similar examples are first gathered together and their directed graph representations are aligned in a clustering process. 2. Structured Representation The representation of design ....

Carbonell, J.G. (ed): 1989, Machine Learning: Paradigms and Methods, Elsevier, Amsterdam, The Netherlands.

....raw text corpora, annotated text corpora, and existing lexical databases. 2 Association of Computational Linguistics Special Interest Group in Natural Language Learning; URL: http: signll.aclweb.org signll . il.tex; 8 01 1999; 16:43; p. 3 4 Walter Daelemans and Gert Durieux Langley (1996) and Carbonell (1990) for introductory material, Weiss and Kulikowski (1991) for methodological issues, and Natarajan (1991) for a formal theoretical approach) Conceptually, a learning system consists of a performance component which performs a specific task (given an input, it produces an output) and a learning ....

Carbonell, J. G.: 1990, Machine learning: paradigms and methods. Cambridge, MA: MIT Press.

....data where it is important to extract the relevant items quickly. The learning capability with the distributed representation (over the neurons) make them candidates for such problems. Learning capabilityisnot an exclusive feature of neural networks. Knowledge bases too have learning capability [4, 16, 11]. However, such capability is not enough in the contest of uncertain and noisy data such as object and speech recognition. Thus inductive logic programming is not appropriate for those problems. Another force of neural networks is their ability to generalize from examples. They will always give ....

J. G. Carbonell. Machine Learning: Paradigms and Methods. The MIT Press, 1990.

....rather than algorithmic ones. These dimensions separate artificial and cognitive science from mainstream computer science and pattern recognition, and machine learning is much more closely associated with the former two areas than with the latter two. Carbonell defines machine learning as follows [8]: Perhaps the tenacity of ML researchers in light of the undisputed difficulty of their ultimate objectives, and in light of early disappointments, is best explained by the very nature of the learning process. The ability to learn, to adapt, to modify behavior is an inalienable component of human ....

....from examples and usually known counterexamples of the concept. The task is to build a concept description from which all previous positive instances can be rederived by universal instantiation while none of the previous negative instances (counterexample) can be rederived by the same process [8]. Until recently learning referred almost exclusively to classification mechanisms, focusing on programs that learn concept descriptions from a series of examples and counterexamples. While learning now extends to include many other topics and types of systems, classification is still an active ....

J.G. Carbonell, editor, Machine Learning: Paradigms and Methods, The MIT Press, 1990. BIBLIOGRAPHY 84

.... particular situations; learning from examples, in which an environment contains specific examples which must be generalized to form more abstract concepts or rules; and learning by analogy, in which the learner must discover the similarity between trial situations and future situations [MCM86] [Car89]. There is a need not only for having machine learning strategies available but for guidelines deciding which ones to use and when to use them. Machine learning techniques are developed without regard to their eventual use, because the techniques are general enough to be applicable to innumerable ....

Jaime Carbonell. Machine Learning: Paradigms and Methods. MIT Press, 1989.

....its time is low order polynomial, it can be seen as one of the fastest learning algorithms to date. The rules produced by HCV for each concept from an input data set can be thought of as a disjunctive set of conjunctive terms. The partitioning technique adopted in HCV is a kind of greedy covering [Carbonell 90] So, HCV has attacked the best known problem (learning disjunctive concepts) in inductive learning by coupling one of the oldest and best known techniques, greedy covering, with the HFL algorithm which chooses the candidate conjuncts at each point in the run. The combination can guarantee that ....

J.G. Carbonell (Ed.), Machine Learning: Paradigms and Methods, The MIT Press, 1990.

....for a given problem among several hand coded representations. Many researchers have addressed the representation problem by designing learning algorithms that deduce important information from the domain description and use the deduced information to improve the representation [ Allen et al. 1992; Carbonell, 1990 ] Examples of these representation improvements include decomposing a problem into subproblems [ Newell et al. 1960 ] generating abstraction hierarchies [ Knoblock, 1994; Bacchus and Yang, 1994 ] replacing operators with macros [ Korf, 1985; Mooney, 1988; Shell and Carbonell, 1989 ] ....

Jaime G. Carbonell, editor. Machine Learning: Paradigms and Methods. MIT Press, Boston, MA, 1990.

....a model of a task domain, including the systematic patterns of interaction of an agent situated in a task environment. Learning of an agent involves both learning to solve new problems and learning better ways to solve previously solved problems. Carbonell describes machine learning as follows [6]: Perhaps the tenacity of ML researchers in light of the undisputed difficulty of their ultimate objectives, and in light of early disappointments, is best explained by the very nature of the learning process. The ability to learn, to adapt, to modify behavior is an inalienable component of human ....

....of the concept and (usually) known counterexamples of the concept. The task is to build a concept description from which all previous positive instances can be rederived by universal instantiation but none of the previous negative instance (counterexample) can be rederived by the same process [6]. Learning from examples has been one of the primary paradigms of ML research since the early days of Artificial Intelligence (AI) Many researchers have observed and documented the fact that human problem solving performance improves with experience. In some domains, the principal source of ....

J. G. Carbonell, editor. Machine Learning: Paradigms and Methods. The MIT Press, 1990.

....with all algorithms. The task of finding a good representation is usually left to the human user. Many researchers have addressed the representation problem by designing learning algorithms that deduce important information from the domain description [Newell et al. 1960; Allen et al. 1992; Carbonell, 1990] which includes learning control rules [Langley, 1983; Laird et al. 1986; Minton, 1988; Veloso and Borrajo, 1994] generating abstraction hierarchies [Knoblock, 1994] replacing operators with macros [Korf, 1985; Mooney, 1988] and reusing past problem solving cases [Carbonell, 1983; Hall, ....

Jaime G. Carbonell, editor. Machine Learning: Paradigms and Methods. MIT Press, Boston, MA, 1990.

....induction results from different data sets in a single rule base and perform rule based reasoning. The rules produced by HCV for each concept from an input data set can be thought of as a disjunctive set of conjunctive terms. The partitioning technique adopted in HCV is a kind of greedy covering [5]. So, HCV has attacked the best known problem (learning disjunctive concepts) in inductive learning by coupling one of the oldest and best known techniques, greedy covering, with the HFL algorithm which chooses the candidate conjuncts at each point in the run. The combination can guarantee that ....

J.G. Carbonell (Ed.), Machine Learning: Paradigms and Methods, The MIT Press, 1990.

....see, the results produced by each of C4.5, C4.5rules and HCV are more compact and more general than the original cases in Table 1. They take less space and can be used to predict or classify new examples. In contrast to credit assignment and generate and test processes in genetic algorithms (Carbonell, 1990) and numerical activity vectors based numerical computations in connectionist methods (Dayhoff, 1990) attribute based induction concentrates on symbolic and heuristic computations. These relate to models that operate at the level of symbols and operations that manipulate symbolic expressions with ....

....and generated in HCV take the form of variable valued logic rules rather than decision trees. The rules produced by HCV for each concept from an input database can be thought of as a disjunctive set of conjunctive terms. The partitioning technique adopted in HCV is a kind of greedy covering (Carbonell, 1990). So, HCV has attacked the best known problem (induction of disjunctive concepts) in induction by coupling one of the oldest and best known techniques, greedy covering, with the HFL algorithm which chooses the candidate conjuncts at each point in the run. The combination can guarantee that the ....

Carbonell, J.G. (Ed.). (1990). Machine Learning: Paradigms and Methods. The MIT Press.

....in the small 7 although theoretical gorithmic and implementational advances at the foundational level will continue to improve the basic building blocks in the field. Empirical induction methods have been developed [Michalski et al. 7 19837 Michalski et al. 7 19867 Michalski and Kodratoff 7 19907 Carbonell 7 1990] at the symbolic level and tested on standard (albeit small) test suites 7 I and occasionally they have been used externally 7 as in the case of decision tree induction [Quinlan 7 19837 Quinlan 7 19867 Nfifiez7 1991] Subsymbolic induction methods 7 including genetic algorithms [Holland 7 19867 ....

....and analytic performance improvements. In this chapter 7 1. The University of California at Irvine maintains informaJ]y a varied set of training and test data for inductive generalization deposited and accessible by researchers in machine learning. we address the latter in the context of PRODIGY [Carbonell eta] 1990, Minton eta] 1989b, Ve]oso, 1989] a general purpose complete plan ner that incorporates various learning techniques: explanation based learning (EBL) Minton, 1988] acquisition of control knowledge through static analysis [Etzioni, 1990] learning by analogy [Veloso, 1991] learning by ....

[Article contains additional citation context not shown here]

Carbonell, J. G., editor (1990). Machine Learning: Paradigms and Methods. MIT Press, Boston, MA.

....in the small, although theoretical, algorithmic and implementational advances at the foundational level will continue to improve the basic building blocks in the field. Empirical induction methods have been developed [ Michalski et al. 1983, Michalski et al. 1986, Michalski and Kodratoff, 1990, Carbonell, 1990 ] at the symbolic level and tested on standard (albeit small) test suites, 1 and occasionally they have been used externally, as in the case of decision tree induction [ Quinlan, 1983, Quinlan, 1986, N u nez, 1991 ] Subsymbolic induction methods, including genetic algorithms [ Holland, 1986, ....

....this chapter, 1. The University of California at Irvine maintains informally a varied set of training and test data for inductive generalization deposited and accessible by researchers in machine learning. 2 Towards Scaling Up Machine Learning we address the latter in the context of prodigy [ Carbonell et al. 1990, Minton et al. 1989b, Veloso, 1989 ] a general purpose complete planner that incorporates various learning techniques: explanation based learning (EBL) Minton, 1988 ] acquisition of control knowledge through static analysis [ Etzioni, 1990 ] learning by analogy [ Veloso, 1991 ] ....

[Article contains additional citation context not shown here]

Carbonell, J. G., editor (1990). Machine Learning: Paradigms and Methods. MIT Press, Boston, MA.

No context found.

Jaime G. Carbonell, editor. Machine Learning: Paradigms and Methods. MIT Press, Boston, MA, 1990.

No context found.

J. G. Carbonell, Ed., Machine Learning: Paradigms and Methods. Cambridge, MA: MIT Press, 1990.

No context found.

Carbonell, J. G. Machine Learning: Paradigms and Methods. MIT Press/Bradford Books; Cambridge, MA, 1992.

No context found.

Carbonell, editor, Machine Learning: Paradigms and Methods, MIT Press, Cambridge, MA. Schmidt, R.A. (1968). Anticipation and Timing in Human Motor Performance. Psychological Bulletin, 70:631-646.

No context found.

Carbonell, J., editor (1990). Machine Learning: Paradigms and Methods. MIT Press, Cambridge, Mass.

No context found.

Carbonell (ed.), Machine Learning: Paradigms and methods. MIT Press. Cambridge, MA. Schank, R. C., and Osgood, R. (1990). A Content Theory of Memory Indexing. Technical Report 2. Institute for the Learning Sciences, Northwestern University, Evanston, IL.

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