Wnek, J. & Michalski, R.S. (1994c). Discovering Representation Space Transformations for Learning Concept Descriptions Combining DNF and M-of-N Rules. In Working Notes of the MLCOLT '94 Workshop on Constructive Induction and Change of Representation (pp. 61-68). New Brunswick, NJ.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....in the consumer problem, every variable value must be associated with equal numbers of each output value. This means that no variable value is of any use at all in the initial, splitting up of cases and each, successive, split produces the same, non uniform situation at a finer grained level. [6] The result is that ID3 necessarily builds a lookup table for the consumer problem, i.e. a decision tree that captures no generalisations whatsoever and which has one leaf node per case in the training set. The decision tree actually produced in shown in Figure 1. x1 computer x2 consumes x3 ....

Wnek, J. and Michalski, R. (1994). Discovering representation space transformations for learning concept descriptions combining DNF and m-of-n rules. Proceedings of ML-COLT'94.

....almost perfect theories. Therefore constructive induction is only occasionally able to improve the scores marginally. But in every test run the initially induced theory was rewritten into a concise and easily comprehensible form like exemplified by the above given sample rule set. reported in [Wnek Michalski 94] Still this approach does not seem to be able to handle considerable levels of noise. 6.4. EXPERIMENTS 70 Noise First Best 0 63.37 99.51 5 57.36 99.20 10 50.05 50.05 Table 6.4: PARITY 5: accuracies (percentages) for CiPF 2.0 after the first and after the best cycle of induction for various ....

Wnek J., Michalski R.S.: Discovering Representation Space Transformations for Learning Concept Descriptions Combining DNF and M-of-N Rules, Workshop on Constructive Induction and Change of Representation, ML-COLT94, Rutgers, 1994.

....for constructive induction. For instance, it allows to retain redundant rules (M 1) i.e. rules that generalize same example(s) The comparison of the premises of redundant rules gives hints about defining new descriptors (e.g. XOR or M of N expressions involving the initial attributes) [WM94]. This ability allows to successfully handle some cases of weak adequacy, such as those arising from learning symmetry or parity based concepts with, so to say, bare hands. 1.3 The Version Space The version space frame [Mit82] determines the upper and lower bounds of the learning search: it ....

....at once would of course, if by any means tractable, be preferred. ffl It is a real case of weak adequacy, to be contrasted with what would be illadequacy. As a matter of fact, computer science witnesses that any numeric expression can be computed (approximated) by means of logical expressions [WM94]. So, it is not hopeless that logical learners could reach some competence regarding numerical problems (without any pretension to winning over purposely devised algorithms) All of the reviewed learning approaches basically handle numerical data via selectors [Mic83] i.e. logical functions ....

J. Wnek and R.S. Michalski. Discovering representation space transformations for learning concept descriptions combining dnf and m-of-n rules. In T. Fawcett, editor, Workshop on Constructive Induction, ICML-94, 1994.

No context found.

Wnek, J. & Michalski, R.S. (1994c). Discovering Representation Space Transformations for Learning Concept Descriptions Combining DNF and M-of-N Rules. In Working Notes of the MLCOLT '94 Workshop on Constructive Induction and Change of Representation (pp. 61-68). New Brunswick, NJ.

....attributes exist attribute construction methods can be invoked which try to combine the given attributes in more problemrelevant manner. A number of systems have been developed with this goal. These systems can be classified into data driven, hypothesis driven, knowledge driven and multistrategy (Wnek and Michalski, 1994). Some representative of each of these types are: AQ17 DCI (Bloedorn and 6 Michalski, 1991) BLIP (Wrobel, 1989) CITRE (Matheus and Rendell, 1989) Pagallo and Haussler s FRINGE, GREEDY3 and GROVE (Pagallo and Haussler, 1990) MIRO (Drastal, Czako and Raatz, 1989) and STABB (Utgoff, 1986) 3. ....

....task. An constructive induction based learning agent is able to expand or contract the provided representation space either automatically, or based on knowledge provided by the user using one or more of the different types of CI: data driven, hypothesis driven, knowledge driven and multistrategy (Wnek and Michalski, 1994). An architecture for a constructive induction based learning agent is shown in Figure 3. In this architecture the agent acts as an assistant to the user in dealing with the environment. The user can access the environment directly or through the agent. In its passive monitor mode the agent ....

[Article contains additional citation context not shown here]

Wnek, J. and Michalski, R.S., "Discovering Representation Space Transformations for Learning Concept Descriptions Containing DNF and M-of-N Rules," Working Notes of the ML-COLT94 Workshop on Constructive Induction, New Brunswick, NJ, 1994.

No context found.

Wnek, J. & Michalski, R.S. (1994c). Discovering Representation Space Transformations for Learning Concept Descriptions Combining DNF and M-of-N Rules. In Working Notes of the ML-COLT'94 Workshop on Constructive Induction and Change of Representation (pp. 61-68). New Brunswick, NJ.

....that expand the space by adding new dimensions (attributes) and contractors that contract the space by removing less relevant attributes and or abstracting values of some attributes. Since the inception of this view of CI as a double search, several methods for CI have been proposed (Wnek and Michalski, 1994, Wnek and Michalski, 1994) However, a comparison of the relative strengths of various methods has not been performed. This paper performs a comparison for five different CI methods: AQ HCI(ADD) AQ HCI(REMOVE) AQ HCI (ADD and REMOVE combined) AQ DCI and AQ SCALE. This analysis shows a ....

....space by adding new dimensions (attributes) and contractors that contract the space by removing less relevant attributes and or abstracting values of some attributes. Since the inception of this view of CI as a double search, several methods for CI have been proposed (Wnek and Michalski, 1994, Wnek and Michalski, 1994) . However, a comparison of the relative strengths of various methods has not been performed. This paper performs a comparison for five different CI methods: AQ HCI(ADD) AQ HCI(REMOVE) AQ HCI (ADD and REMOVE combined) AQ DCI and AQ SCALE. This analysis shows a strong need for multistrategy CI ....

[Article contains additional citation context not shown here]

Wnek, J., and Michalski, R.S., "Discovering Representation Space Transformations for Learning Concept Descriptions Containing DNF and M-of-N Rules," Working Notes of the ML-COLT94 Workshop on Constructive Induction, pp. New Brunswick, NJ, 1994.

....This research concerns automatic determination of neural network topology, and is viewed as a constructive induction method. Constructive induction systems perform a double search, one for the most suitable representation space, and second for the most suitable concept description in this space (Wnek Michalski, 1994). They include mechanisms for generating new, more relevant descriptors, as well as modifying or removing less relevant ones from those initially provided. This paper describes an approach to constructive induction in which the desirable changes in the representation space are determined by ....

....changes in the representation space are determined by analyzing the hypotheses generated in each iteration of the learning process. For this reason, this approach is a hypothesis driven constructive induction (HCI) as opposed to a knowledge driven approach (KCI) or data driven approach (DCI) (Wnek Michalski, 1994). In the context of neural network knowledge representation, the KCI approach can be identified, for example, with the KBANN system (Towell et al. 1991) The DCI approach in neural networks is yet to be identified. 2 HCI APPROACH The HCI approach is based on repetitively detecting strong ....

[Article contains additional citation context not shown here]

Wnek, J. and Michalski, R.S., "Discovering Representation Space Transformations for Learning Concept Descriptions Combining DNF and M-of-N Rules," Working Notes of the ML'94 Workshop on Constructive Induction and Change of Representation, 1994.

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