J. R. Quinlan. Combining instance-based and model-based learning. In Proceedings of the Tenth International Conference on Machine Learning, pages 236--243, Amherst, Massachusetts, 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....regression based machine learning algorithm designed to learn a mapping from discrete valued inputs to continuous valued outputs. It operates by searching for useful conjuctive queries that form indicator functions for a linear regression. It is very similar to GMDH modeling [6] Model Trees [11] or even stepwise polynomial regression. In a dataset with two input attributes Gender and HairColor and one output attribute Age , a typical RADREG model might decide to use the following features: Gender=Female AND HairColor=Grey) HairColor=Red) Gender=Male AND HairColor=Red) in ....

J. R. Quinlan. Combining Instance-Based and Model-Based Learning. In Machine Learning: Proceedings of the Tenth International Conference, 1993.

....asymptotically in the number of iterations. This result was further strengthened in [HdMR02] that show that the CE converges with probability one to the so called optimal (global) solution after a nitely many iterations. The CE method can be viewed as a model based optimisation technique [Qui93], which involves two phases: 1. Generation of a sample of random data (trajectories, vectors, etc. according to a speci ed random mechanism. 2. Updating the parameters of the random mechanism, on the basis of the data, in order to produce a better sample in the next iteration. The signi ....

J. Quinlan. Combining instance-based and model-based learning. In Proceedings of the Tenth International Conference on Machine Learning, pages 236 - 243, San Mateo, CA, 1993. Morgan Kaufmann.

....components can be passed. Automobile Example The first example uses automobile data that can be found at the University of California Irvine, Repository of Machine Learning Database [18] The dataset was first used in the 1983 American Statistical Association Exposition and later used by Quinlan [19] to predict automobile gas mileage. The data set has information from 392 automobiles with seven variables of interest provided in Table 1. In this example we will use the first six variables to predict the seventh variable: the car s acceleration. Table 1. Automobile Data Set Variable Type 1 ....

R. Quinlan, Combining instance-based and model-based learning, proceedings on the Tenth International Conference of Machine Learning, University of Massachusetts, Amherst, Morgan Kaufmann, pp. 236-243, (1993)

....these experiments it is not to compare RECLA with these alternative methods. RECLA is not a learning system. As a pre processing tool the resulting accuracy is highly dependent on the classification system after the discretisation takes place. The first column of Table 10 presents the results M5 [11, 13]. This regression system is able to learn tree based models with linear regression equations in the leaves (also known as model trees) By default this system makes the prediction for each testing instance by combining the prediction of a model tree with a 3 nearest neighbour [13] In the second ....

.... results M5 [11, 13] This regression system is able to learn tree based models with linear regression equations in the leaves (also known as model trees) By default this system makes the prediction for each testing instance by combining the prediction of a model tree with a 3 nearest neighbour [13]. In the second column we give the result when this combination is disabled thus using only model trees. The third column of the table gives the results obtained by a standard 3 nearest neighbour algorithm. The fourth column shows the results using a least squares linear regression model. We then ....

Quinlan,J.R., Combining Instance-based and Model-based Learning, in Proceedings of the 10th ICML, Morgan Kaufmann, 1993.

....problems, for which the existence of efficient exact algorithms is highly unlikely, has led to a wide range of heuristic algorithms that implement some sort of search in the solution space. These heuristic algorithms can be classified, similarly to what is done in the machine learning field [15] , as being either instance based or model based. Most of the classical search methods may be considered instance based, since they generate new candidate solutions using solely the current solution or the current population of solutions. Typical representatives of this class are genetic ....

J. Quinlan. Combining instance-based and model-based learning. In Proceedings of the Twelfth International Conference on Machine Learning (ML-93), pages 236--243. Morgan Kaufmann Publishers, San Mateo, CA, 1993.

....directly to the task of process prediction, and the tightness of the confidence intervals is competitive with the accuracy of purely predictive approaches, which we consider an very positive result. An alternate approach to combining model based and instance based learning can be found in [ 9 ] . Quinlan uses a quantitative model to correct the values retrieved by a normal case based system, by applying the model both to the target instance and to the retrieved instance. His idea is that one can use the model to calculate a corrective factor, which is applied to the retrieved instance. ....

J. R. Quinlan: "Combining instance-based and modelbased learning," Proceedings of the 10th International Conference on Machine Learning, 1993.

....This may introduce certain information loss. It would thus seem that a method that does not require prior discretization should be better suited to this task. We have decided to verify this hypothesis and evaluated the capability of linear regression models, piecewise linear models (model trees [9]) and instance based models [1] to capture the information concerning applicability. This paper describes the results. The rest of the paper is organized as follows. Section 2 describes the metadata considered in our study and some preprocessing steps carried out. Section 3 describes several ....

....determine the degree to which the class concept is satis ed, and in e ect, enable to turn a categorical concept into a fuzzy one. 3. 3 Generation of Piecewise Linear Models We have decided to investigate the applicability of a particular class of piecewise linear models represented by model trees [9]. These can be seen as generalization of both linear regression models and decision trees. Whereas a leaf of a decision tree contains just a class name, the leaf of a model tree can contain a linear model relating a class value to corresponding attribute values. We have used Quinlan s M5.1 to ....

Quinlan R. (1993): \Combining Instance-Based and Model-Based Learning", in ML93, Machine Learning, Proceedings of the 10th International Conference, P.Utgo (ed.), Morgan Kaufmann.

.... where the sampling is adaptive, concentrating on misclassified training instances [ Freund and Schapire, 1997 ] Voting methods have also been applied to combining multiple neural networks trained on the same data [ Perrone, 1993 ] and applying di#erent types of classifiers to the same problem [ Quinlan, 1993 ] Why consensus algorithms work so well in practice is still an open question. As a step in that direction, theoretical work has recently established that combining multiple runs of a classification algorithm can reduce its variance [ Breiman, 1996b ] Unlike most voting algorithms, the ....

J. Quinlan. Combining instance-based and model-based learning. In Proceedings of the International Conference on Machine Learning. Morgan Kaufman, 1993.

....Harrison and Rubin eld (1978) is a well known benchmark test dataset. We use it to demonstrate that generalized ridge methods can again prove quick alternatives to BMA with good predictive power. We tested the predictive accuracy of the methods with ten fold cross validation, using the splits of Quinlan (1993). In each case we tted a linear model to the each training dataset and then determined the SSE over the corresponding test sets. 15 CV Set BMA GRN BGR1 BGR2 STR LSE 1 20.42 20.70 20.61 20.45 20.16 20.12 2 22.61 21.42 21.48 21.64 22.27 22.12 3 25.63 25.01 25.10 25.18 25.37 25.48 4 30.47 ....

Quinlan, R. (1993) Combining instance-based and model-based learning. Machine Learning: Proc. 10th Int. Conf., Amherst, MA, 1993. Morgan Kaufmann.

....studies the prediction of physico chemical parameters from biological parameters. In this work the same data set was used as the one we are using now, except for the water ow values. A di erent regression tree was built for each of the 16 physico chemical parameters; the model tree learner M5 [13] was employed. The regression trees turn out to have a predictive accuracy comparable to that of a nearest neighbor method and better than that of a linear regression method that were used as well in the article. Moreover the induced trees are relatively small, hence interpretable, and ....

J.R. Quinlan. Combining instance-based and model-based learning. In Proceedings of the 10th International Workshop on Machine Learning. Morgan Kaufmann, 1993.

....from biological ones [9] Dzeroski et al. 9] discuss the construction of predictive models that allow prediction of a specific physico chemical parameter from biological data. A different predictive model is built for each parameter. The models, which are constructed using Quinlan s M5 system [17], are in the form of regression trees. This approach is compared with nearest neighbour and linear regression methods; the authors conclude that the induction of regression trees is competitive with the other approaches as far as predictive accuracy is concerned, and moreover has the advantage of ....

....regression mode employing as heuristic the variance as described above. The system seemed fit for our experiments because of the following reasons: Most machine learning and data mining systems that induce predictive models can handle only single target variables (e.g. C4.5 [15] CART [5] M5 [17], Building a predictive model for a multi dimensional prediction space can be done using clustering systems, but most clustering systems consider clustering as a descriptive technique, where evaluation criteria are still slightly different from the ones we have here. Using terminology ....

J.R. Quinlan. Combining instance-based and model-based learning. Proc. 10th Int'l Workshop on Machine Learning. Morgan Kaufmann, 1993.

....EM OFF Figure 5: Petri net model of the paper feed system. used to distinguish, for example, between a slow motor and a stalled motor. For real time, embedded applications, the fault symptom table can be compactly represented by a corresponding decision tree using, for example, the ID3 algorithm [Quinlan, 1993] . In our diagnosis system we have two types of sensors, builtin sensors that are always accessible with a low cost and virtual sensors that cannot be used directly in the diagnoser but require the invocation of the mode estimation algorithm. Thus, the built in sensors can be used for fault ....

J.R. Quinlan. Combining instance-based and model-based learning. In Proceedings of the 10th International Conference on Machine Learning, 1993.

....as cache memory size and cycle time. ffl Servo. J. R. Quinlan says that It covers an extremely non linear phenomenon predicting the rise time of a servomechanism in terms of two (continuous) gain settings and two (discrete) choices of mechanical linkages. There are 167 examples. Following (Quinlan, 1993), each dataset was randomly split into approximately 10 equal parts. A kernel, kernel parameter and ridge factor for dual form ridge regression estimation were chosen for each dataset by cross validating across the different parts and taking the values giving the smallest absolute error. The ....

....parameter 1.5 2.5 3 2.5 a 0.1 0.001 0.1 0.001 Dual form ridge regression has been shown to perform well on benchmark datasets before, here the method shows comparable performance to techniques previously used on the datasets. All results given here other than for ridge regression are taken from (Quinlan, 1993), more recently (Saunders et al. 1998) has shown comparable performance to support vector regression. However, here we are interested in region prediction, not point prediction. The first interesting property of this algorithm is that, as stated in remark 1 it returns almost precise confidence ....

Quinlan, J. R. (1993). Combining Instance-Based and Model-Based Learning. Proceedings ML'93. San Mateo, CA: Morgan Kaufmann.

....0.4 0.6 0.8 Figure 5: The left hand side shows the target function of the artificial regression problem. The right hand side shows the output of a soft WTA circuit for n = 500, k = 250 after training. was 2:656, which is in the ballpark of several other algorithms for which the reported error [Quinlan, 1993] is between 2:23 and 2:90. A cross validation error of our algorithm (n = 150, k = 75) on the Wisconsin breast cancer data was 6:00 which is again comparable with the results of other algorithms. Whereas it was already known that winner take all circuits are universal approximators, it had ....

Quinlan, J.R. (1993). Combining instance-based and model-based learning. Proc. of the 10th Int. Conf. on Machine Learning, Morgan Kaufmann (San Mateo), 236-243.

....tracts within the Boston metropolitan area (in 1970) the data gives 13 input variables, including per capita crime rate and nitric oxides concentration, and one output, the median housing price for that tract. A ten fold cross validation method was used to evaluate the performance, as detailed in [9]) The dataset was divided into ten blocks of near equal size and distribution of class values (I used the same partitions as in [9] For each block in turn the parameters of the Gaussian process were trained on the remaining blocks and then used to make predictions for the hold out block. For ....

....concentration, and one output, the median housing price for that tract. A ten fold cross validation method was used to evaluate the performance, as detailed in [9] The dataset was divided into ten blocks of near equal size and distribution of class values (I used the same partitions as in [9]) For each block in turn the parameters of the Gaussian process were trained on the remaining blocks and then used to make predictions for the hold out block. For each of the ten experiments the input variables and targets were linearly transformed to have zero mean and unit variance, and five ....

J. R. Quinlan. Combining Instance-Based and Model-Based Learning. In P. E. Utgoff, editor, Proc. ML'93. Morgan Kaufmann, San Mateo, CA, 1993.

.... Gamma1 1 1 C C C C C A : 3) The parallel plug in classifiers of the components are exclusively dedicated to the single two class problems C j and C Gamma j of the complete task. The plug in classification approach essentially differs from methods based on combining multiple classifiers [15] or ones that combine multiple runs of the same learning algorithm [16] The desired mapping of a single parallel unit is given by R n f Gamma1; 1g; x 7 ( 1 ; if C k 2 C j Gamma1 ; else; 4) where C k denotes the category to which the input vector x belongs. In the following, some ....

....voting. Whereas homogeneous voting is based on combining different hypotheses after multiple runs of the same learning algorithm [16] non homogeneous voting improves the performance of the decision rule by combining multiple classifiers that have been constructed by different learning algorithms [15]. However, in plug in classification systems, the same learning algorithm is applied many times to different classification problems (dichotomies) and it has been shown that error correcting output coding reduces both bias and variance of the constructed classifier [7] Apart from the 1 out of K ....

J.R. Quinlan. Combining instance-based and model-based learning. In Proceedings of the 10th International Conference on Machine Learning, Amherst, Massachusetts, pages 236--243. Morgan Kaufmann, 1995.

No context found.

J. R. Quinlan. Combining instance-based and model-based learning. In Proceedings of the Tenth International Conference on Machine Learning, pages 236--243, Amherst, Massachusetts, 1993.

No context found.

J. R. Quinlan. Combining Instance-Based and Model-Based Learning. In Machine Learning: Proceedings of the Tenth International Conference, 1993.

No context found.

Quinlan, J. R. (1993). Combining instance-based and model-based learning. In Proceedings tenth international machine learning conference.

No context found.

Quinlan, J. R. 1993. Combining instance-based and model-based learning. In Proceedings Tenth International Machine Learning Conference. Amherst. MA. Morgan Kaufmann.

No context found.

J. R. Quinlan. Combining instance-based and model-based learning. In Proceedings of ICML'93, pages 236--243, 1993.

No context found.

J. R. Quinlan and R. M. Cameron-Jones. Combining instance-based and model-based learning. In Proceedings of the Tenth International Conferenceon Machine Learning, pages 236-243, 1993.

No context found.

R. Quinlan. Combining instance-based and model-based learning. In Proceedings of the 10th International Conference of Machine Learning, pages 236--243. Morgan Kaufmann, 1993.

No context found.

J. Quinlan. Combining instance-based and modelbased learning. In Proc. 10th Int. Conf. on Machine Learning, 1993.

No context found.

Quinlan, J.R. (1993) Combining instance-based and model-based learning. In Proc. Tenth International Conference on Machine Learning, pages 236--243. Morgan Kaufmann, San Mateo, CA.

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