Michalski, R. S., & Chilausky, R. L. (1980). Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. Policy Analysis and Information Systems, 4, 125-160.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....on classification. Inductive learning refers to learning from examples in which knowledge is acquired by drawing inductive inference from the examples given. Acquiring knowledge involves operations of generalizing, specializing, transforming, correcting and refining knowledge representations [42, 43]. Many of the tasks to which machine learning techniques are applied are tasks that humans can perform quite well. However, humans often cannot tell how they solve these tasks. Inductive supervised learning is able to exploit the human ability to assign labels to given instances without requiring ....

R.S. Michalski and R.L. Chilausky, Learning by Being Told and Learning from Examples: An Experimental Comparison of the Two methods of Knowledge Acquisition In the Context of Developing an Expert System for Soybean Disease Diagnosis, International Journal of policy Analysis and Information Systems, 4, 1980.

....During the past decade, expert systems have become the first practical application of artificial intelligence. However, many expert system applications are rule based systems, which require a time consuming and difficult knowledge acquisition process to extract expertise from domain experts [1 3]. Neurocomputing is the application of artificial neural networks to practical problems. An artificial neural network consists of processing elements (which is analogous to neurons in the biological neural system) in an interconnected network [4 11] Aprocessing element accepts inputs, ....

R.S. Michalski and R.L. Chilausky, Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean diagnosis, International Journal of Policy Analysis and Information Systems (1980)125 -- 61.

.... of this type include: learning a checkers evaluation function [29, 30] that assigns to a given class of board situations a certain value; learning descriptions of block structures [36] determining rules for interpreting mass spectrograms [5] formulating diagnostic rules for soybean diseases [25]; and discovering heuristics to guide the application of symbolic integration operators [27] In Samuel s checkers program, for example, each training instance was a board situation represented as a vector of 16 attributes. The learned concept was an evaluation function that computed the value ....

....plant in terms of 35 multi valued attributes. Each plant had one of 19 common soybean diseases. From several hundred training instances, the program inferred general diagnostic rules for these diseases. These rules outperformed rules developed through consultation with expert plant pathologists [25]. In all of the above problems, the system learns a general class description from instances of the class, and therefore, this type of inductive learning can be called instance to class generalization. A review of several methods for such instance to class generalization can be found in [8, 24] ....

Michalski, R.S. and Chilausky, R.L., Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis, Policy Analysis and Information Systems 4 (2) (i9S0).

....monk( red, monk( Fig. 2. Hypotheses for the MONK1 problem represented as atomic formulae built by the GSL algorithm A further illustration of the performance of the algorithm for a concept learning task is given in Figure 3. It shows the generalizations of the soybean data [7]. These are 47 examples of four soybean diseases described by 21 nominal attributes. The class attribute is skipped. Only the top part of the generalization tree is shown where only the most general hypotheses covering single class examples are included. The hypotheses are denoted by lists of the ....

R. Michalski and R. Chilausky. Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. International Journal of Policy Analysis and Information Systems, 4(2), 1980.

....this work on the general problem of reformulation are discussed in light of experience on real Cyc data. The algorithm is called MDL OC, for Minimum Description Length Orthogonal Clustering. A more detailed description, including results on a diagnosis problem also used by Michalski and Chilausky [MC80], real Cyc data, and an analogical mapping also used byFalkenhainer et al. [FFG89] is given in Derthick [Der90] 2 Problem Representation For the purposes of this paper, the goal of a learning knowledge representation system is to develop a domain theory from a set of ground atomic assertions ....

....Quinlan 96 . Since few algorithms have been applied to the family relations data, MDL OC was also run on the large soybean data in spite of the fact that it is not relational. Completion was 69 on the training set and 74 on the test set, in contrast to others who have attained close to 100 [MC80]. Primarily this poor completion performance is because the MDL principle limits discovered knowledgetothatwhich can be expressed more compactly than the data, and MDL OC s rule language is so simple that only a small class of knowledge can be captured concisely. Better completion was achieved ....

Ryszard S. Michalski and R. L. Chilausky. Learning by being told and learning from examples: An experimental comparison of the two methodsofknowledge acquisition in the context of developing an expert system for soybean disease diagnosis. International Journal of Policy Analysis and Information Systems, 4(2), 1980.

....description. Monks3 has 5 noise added. Breast Cancer (from University of Wisconsin Hospitals, Madison [MW90] Given some information about a patient, determine whether she su ers from a benign or malignant cancer. Contains some unknown attribute values. Soybean (small and large databases [Mic80] Characterize the soybean diseases. In the large dataset, we only use the rst 15 attributes. Contains some unknown attribute values. Votes (Congressional voting record database [Sch87] Predict the representative of a member of the House given its previous votes. Contains some unknown ....

R. Michalski. Learning by being told and learning from examples ... International Journal of Policy Analysis and Information Systems, 4(2):125-161, 1980. 32

....work of Buchanan, Feigenbaum, and Sridharan on MetaDendral [5] where human experts were crucial in (1) providing initial background knowledge and (2) evaluating the results of the automated system. The thread continued with the work of Michalski and Chilauski on soybean disease diagnosis in 1980 [27], and became particularly clear in the work on structured induction later in the 1980s by Shapiro, Niblett, and Leech [45, 24, 46] In structured induction, the expert and learning system construct a body of background knowledge together: the expert provides data for various target sub concepts, ....

R.S Michalski and R.L. Chilausky. Learning by being told and learning from examples: an experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. International Journal of Policy Analysis and Information Systems, 4(2):125{ 161, 1980.

....which are used for describing training data. This kind of induction is called selective induction. There are many learning systems of this type such as ID3 [Quinlan, 1983; 1986] Cart 8 [Breiman et al. 1984] C4.5 [Quinlan, 1993a] PLS1 [Rendell, 1983] CN2 [Clark and Niblett, 1989] and AQ11 [Michalski and Chilausky, 1980a] For example, decision tree learning algorithms ID3, Cart, and C4.5 partition the instance space defined by task supplied attributes into regions of locally invariant or the same class membership values by using axis orthogonal splits, each of which is based on a single attribute. If ....

....terms, such as inverse(Y ) sin(Y ) and log(Y ) using a simple minded generate and test strategy 50 [Langley et al. 1987a] To deal with noise, it requires a near constant differential instead of a strict constant to be found. The rule learning algorithm Induce [Larson and Michalski, 1977; Michalski, 1980a; 1983] also uses mathematical operators as constructive operators. It constructs new attributes (descriptors) by applying constructive generalization rules. The detecting descriptor interdependence rule can create new descriptors X=Y and X Theta Y using an idea similar to that of Bacon. It ....

R.S. Michalski and R.L. Chilausky, Learning by being told and learning from examples: an experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. International Journal of Policy Analysis and Information Systems, 4, 125-160.

....data sets. We checked several data repositories, but could not locate data sets for discrete variable clustering. So instead, we obtained classification data sets from the UCI Machine Learning Repository (Merz and Murphy, 1996) and discarded the known class information. We used the Small Soybean (Michalski and Chilausky, 1980), Standard Audiology (Bareiss and Porter, 1987) and Lung Cancer (Hong and Yang, 1994) databases. For the Audiology data set, where both training and test data were available, we merged these data sources. The results, shown in Figure 11 and Table 3, are similar to that for synthetic data. In ....

Michalski, R., & Chilausky, R. (1980). Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. International Journal of Policy Analysis and Information Systems, 4.

....and applied prediction systems to problems in such areas as agriculture[DM79] mathematics[Mit82] MUNB83] chemistry[BFL71] and industrial fuel production[Qui85] Unfortunately, the full range of applicability of these systems is unknown. Systematic evaluation. With few exceptions (e.g. Qui79] [MC80] [Qui85] CN86] QCHL86] Mic87] very little attention has been paid to the systematic evaluation of the systems using empirical data. Moreover, there has been very limited empirical validation of the effectiveness of proposed learning principles and approaches. Coupled with this is a ....

....Section 5.2 for explanation of the 9600 combinations) Although decision trees serve as an efficient representation in this study, we do not imply that this is true for all data sets and tree generation heuristics. Covers are an alternate representation mechanism and are used in the AQ system [MC80] [MMHL86] A comparative study observed 4 that generation of decision trees was more efficient than generation of covers, while covers tended to be more comprehensible [ORo82] In the decision tree approach, the members of a set of objects are classified as either positive or negative instances ....

[Article contains additional citation context not shown here]

R. S. Michalski and R. L. Chilausky. Learning by being told and learning from examples: an experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. Policy Analysis and Information Systems, 4(2), June 1980. 30

....There are 214 examples in this set, and 107 of them are used for training. MLPs with a single hidden layer of 15 units, and RBF networks with 20 kernels are selected for this data set. The SOYBEAN1 data set consists of 19 classes of soybean, which have to be classified using 82 input features [29]. There are 683 patterns in this set, of which 342 are used for training. MLPs with one hidden layer with 40 units, and RBF networks with 40 kernels are selected. Table 12: Combining Results for GENE1. Classifier(s) Ave Med Max Min N Error st. dev. Error st. dev. Error st. dev. Error st. dev. 3 ....

R.S. Michalski and R.L. Chilausky. Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. International Journal of Policy Analysis and Information Systems, 4(2), 1980.

....defines constructive induction as any process that creates new terms in order to improve future learning performance. Therefore, UNIMEM [Lebowitz, 1987] does constructive induction, because it forms new terms (or concepts) in order to improve performance on future classification tasks; AQ 11 [Michalski Chilausky, 1980], which is similar to UNIMEM, does not do constructive induction because the resulting concepts (or terms) are not used in future learning tasks. Even with this restriction, there are many systems containing some module or subsystem that fits the definition of constructive induction. Quite a ....

....utility allows one to argue that the hierarchies it produces are more valid psychologically than hierarchies produced by UNIMEM. However, like UNIMEM, COBWEB is affected by the order in which examples are presented. When COBWEB was run on Stepp s soybean data (a subset of Michalski s soybean data [Michalski Chilausky, 1980]) it required three iterations through the examples before converging on the desired concept hierarchy [Fisher, 1987] Rendell s PLS0 [Rendell, 1985] is a more complicated system that does not have this limitation. It is designed for domains like Checkers, where the structural attributes are ....

Michalski, R. S., & Chilausky, R. L. (1980). Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. Policy Analysis and Information Systems, 4, 125-160.

....on Domain Information 1 1 Introduction One of the original goals of Artificial Intelligence (AI) was to endow machines with the ability to learn from experience. Research toward this goal has produced a variety of inductive learning algorithms (e.g. the Perceptron [Minsky Papert, 1972] AQ11 [Michalski Chilausky, 1980], ID3 [Quinlan, 1983] and some impressive performance programs (e.g. Samuel s checker program [Samuel, 1959] MetaDendral [Buchanan Mitchell, 1978] LEX [Mitchell, Utgoff Banerji, 1983] However, in spite of these successes, few of today s problem solving systems have the ability to learn. ....

Michalski, R. S., & Chilausky, R. L. (1980). Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. Policy Analysis and Information Systems, 4, 125-160.

....A partial hypothesis covers an example if it includes the point defined by the example. None of these partial hypotheses cover all examples, so they must be combined in a decision list. The decision rules for the decision list are typically formed by a concept learner such as AQ, ID3, or PLS [Michalski and Chilausky, 1980; Quinlan, 1986; Rendell, 1986] Such a concept learner takes as input a set of positive and negative examples. It returns a decision rule that covers (that is, returns true when applied to) each positive example but does not cover any negative example. The assignment problem is thus two fold. ....

R. S. Michalski and R. L. Chilausky. Learning by being told and learning from examples: an experimental comparison of the two methods of knowledge acquisition in the context of developing and expert system for soybean disease diagnosis. Policy Analysis and Information Systems, 4(2), 1980.

....with an error probability similar to the error probability attached to each conjunct in our work. No error probability for the entire decision tree is estimated even though it could be added with an analysis similar to ours. It does not address the issue involved in induction by hierarchy. AQ11 (Michalski 1980) generates DNF rules in an incremental fashion from a preselected set of examples. It does not handle noisy data directly, and does not consider induction by hierarchy either (Forsyth 1986) ACKNOWLEDGEMENTS The authors are members of the Institute for Robotics and Intelligent Systems (IRIS) and ....

Michalski, R.S., and Chilausky, R.J. 1980. Learning by being told and learning from examples. Journal of Policy Analysis & Information Systems, 4.

....complicated than Either s [45] This is because Either s goal is to produce a minimally revised Horn clause theory and Kbann has no such bias. 7. 2 Soybean Results In order to demonstrate Either s ability to revise multiple category theories, Either was used to refine the expert rules given in [24]. This is a theory for diagnosing soybean 7 EXPERIMENTAL RESULTS 29 diseases that distinguishes between nineteen possible soybean diseases using examples that are described with thirty five features. The original experiments compared expertrules to induction from examples. By revising the ....

....and test examples drawn at random from the entire example population, with no overlap. Each point on the curves was computed from a 22 sample average. Note that even with the flexible tester, the accuracy of the original rules was only 51 , as compared to 73 for the original results presented in [24]. Overall, the accuracy of the initial rules is increased by 26 percentage points when Either is trained using 100 training examples. Compared to pure induction, Either maintains its initial performance advantage over the entire training interval. A one tailed Student t test on paired differences ....

R. S. Michalski and S. Chilausky. Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. Journal of Policy Analysis and Information Systems, 4(2):126--161, 1980.

....The data was provided by Gail Gong from Carnegie Mellon University. We also compared the performance of the algorithms on the following non medical real world data sets (SOYB, IRIS, and VOTE are obtained from the Irvine database (Murphy Aha, 1991) SOYB: The famous soybean data set used by Michalski Chilausky (1980). IRIS: The well known Fisher s problem of determining the type of iris flower. MESH3,MESH15: The problem of determining the number of elements for each of the edges of an object in the finite element mesh design problem (Dolsak Muggleton, 1992) There are five objects for which experts have ....

Michalski, R.S. & Chilausky, R.L. (1980) Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. International Journal of Policy Analysis and Information Systems, 4:125-161.

....behavior, starting with Feigenbaum s EPAM model of verbal learning [18] There is also considerable interest in applied machine learning research, focusing on the automatic construction of knowledge based systems. For instance, Michalski and Chilausky have worked on a knowledge base from examples [36]. Since learning is a central phenomenon in human cognition, the researchers evaluate machine learning methods in terms of their ability to explain human learning. A number of different themes can be identified within the machine learning community, each corresponding to central goals of its ....

R.S. Michalski and R.L. Chilausky, Learning by Being Told and Learning from Examples: An Experimental Comparison of the Two methods of Knowledge Acquisition In the Context of Developing an Expert System for Soybean Disease Diagnosis, International Journal of policy Analysis and Information Systems, 4, 1980.

No context found.

Michalski, R.S., and Chilausky, R.L., "Learning By Being Told and Learning From Examples: An Experimental Comparison of the Two Methods of Knowledge Acqusition in the Context of Developing an Expert System for Soybean Disease Diagnosis," Policy Analysis and Information Systems, Vol. 4, No. 2, 1980.

No context found.

Michalski, R. S., & Chilausky, R. L. (1980). Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. Policy Analysis and Information Systems, 4, 125-160.

No context found.

R.S. Michalski and R.L. Chilausky. Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing expert systems for soybean disease diagnosis. International Journal of Policy Analysis and Information Systems, 4(2), 1980.

No context found.

R. S. Michalski and R. L. Chilausky. Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. International Journal of Policy Analysis and Information Systems, 4(2):125--161, 1980.

No context found.

Michalski R.S., Chilausky, R.L., Learning by Being Told and Learning from Examples: An Experimental Comparison of the Two Methods of Knowledge Acquisition in the Context of Developing an Expert System for Soybean Disease Diagnosis, International Journal of Policy Analysis and Information Systems Vol. 4 No. 2 (1980)

No context found.

R.S. Michalski and R.L. Chilausky. Learning by being told and learning from examples: An experimental comparison of the two methods of knowledge acquisition in the context of developing an expert system for soybean disease diagnosis. International Journal of Policy Analysis and Information Systems, 4(2), 1980.

No context found.

Michalski, R.S. and Chilausky, R.L.: Learning by being told and learning from examples. Int J Pol Anal Info Sys, 4, pp. 125--160 (1980).

First 50 documents  Next 50

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