X. Zhang, J.P. Mesirov, and D.L. Waltz. Hybrid system for protein secondary structure prediction. Journal of Molecular Biology, 225:1049--1063, 1992. 30  Home/Search   Document Not in Database   Summary   Related Articles   Check

....dividing the training set into two and using one half for training each level. When the training set is small, one can use leave k out (jackknife) for not to lose precious data [23] Significant improvement in success has been achieved using this approach on protein secondary structure prediction [35]. 6 IMPLEMENTATION RESULTS 6.1 Database We have collected two distinct samples by pairing respectively 28 and 23 cities of Turkey for training and test sets. The first is used for estimating the parameters and the second one to assess their performances. The training and test data sets contain ....

Zhang, X., Mesirov, J.P., Waltz, D.L., "Hybrid System for Protein Secondary Structure Prediction," Journal of Molecular' Biology, 225 (1992) 1049 1063. 16

....simple feature detectors [13] to decision trees [11] In addition to bagging and boosting, there has also been research on combining a more heterogeneous set of weak learners. Di erent training algorithms may be combined see the learning scheme by Goldman and Zhou [6] or work by Zhang et al. [15] as well as Larkey and Croft [7] Herein, we propose to combine classi ers trained by the same algorithm maximum entropy but using a di erent feature set for each classi er. The formula for Voting combining the results of classi ers usually takes the form of a weighted sum (or weighted ....

....Domingos [3] proposes Bayesian averaging, where the weights for the weighted sum is computed based on the probability of a weak learner given an example with its class assignment. Pennock et al. 10] provide a theoretical treatment of this issue, arguing for weighted sum voting. Zhang et al. [15] use a neural network to combine a statistical model, a memory based learner and a neural network for protein secondary structure prediction. An overview to the issues in combining classi ers is presented by Alpayd n [1] We chose neural network back propagation as the learning algorithm for ....

X. Zhang, J. P. Mesirov, and D. L. Waltz. Hybrid system for protein secondary structure prediction. Journal of Molecular Biology, 225:1049-1063, 1992.

....to linear pattern recognition problems (i.e. where one signal is mapped to another, and supervised learning is applied) Examples include prediction of protein secondary fold, a very difficult problem to attack without extensive domain knowledge or pure random sampling for specific subproblems. [5, 21] In research on protein folding that compares pure probabilistic methods (simple HMM learning by EM, followed by Viterbi based matching) to traditional (feedforward with backprop) and knowledge based (KBANN, EBLANN) ANN methods, EM has been shown to outperform extant connectionist systems. 5] ....

....that the pure statistical approach is based only on fixed width windows. The best predictor at the time was a hybrid system combining pure statistical, memory based, and connectionist learning (backprop) with a connectionist front end for data fusion (also a feedforward net trained with backprop) [21] Even compared to this system, EM achieved comparable cross validated prediction accuracy. 5] 2.2. Unsupervised HMM Learning Supervised learning of parameters for HMMs is a popular method for training a pattern recognition system. In spatiotemporal sequence modeling, however, it is often the ....

X. Zhang, J. P. Mesirov, and D. L. Waltz. A hybrid system for protein secondarystructure prediction. Journal of Molecular Biology, 1993.

....whole spectrum of regimes, ranging from the small samples to the asymptotic behaviour. Second, studies based on biological and statistical evidence have given birth to reliable estimates of the optimal level of accuracy associated with the di erent strategies implemented to tackle the problem (see [46] for an upper bound on the accuracy of predictions based on local information and single sequences, or [5] for an overview on performance obtained by using multiple alignments) Apart from the fact that the problem is of central importance in biology, it thus appears interesting due to the ....

X. Zhang, J.P. Mesirov, and D.L. Waltz. Hybrid System for Protein Secondary Structure Prediction. J. Mol. Biol., 225:10491063, 1992.

....Our experimental results show that the proposed combined classifiers indeed outperform the individual classifiers made up solely by Bayesian neural networks. 2 Using an ensemble of classifiers to process biomolecular data has been studied by Brunak et al. 3] Wang et al. 30] and Zhang et al. [36]. In [3] Brunak et al. exploited the complementary relation between exon and splice sites to build a joint recognition system by allowing the exon signal to control the threshold used to assign splice sites. Specifically, a higher threshold was required to avoid false positives for the regions ....

....are only small changes in the exon activity. A lower threshold was used to detect the donor site for the regions where the exon activity decreases significantly. Similarly, a lower threshold was used to detect the acceptor site for the regions where the exon activity increases significantly. In [36], Zhang et al. developed a hybrid system, which included a neural network, a statistical classifier and a memory based reasoning classifier, to predict the secondary structures of proteins. Initially, the three classifiers were trained separately. Then a neural network used as a combiner was ....

Zhang, X., Mesirov, J. P., and Waltz, D. L. Hybrid system for protein secondary structure prediction. Journal of Molecular Biology, 225(4):1049--1063, 1992. 26

....More recent methods often make use of inductive learning techniques, whereby a system is trained with a set of sample proteins of known conformation and then uses what it has learned to predict the structure of previously unseen proteins. Both neural networks [Qian Sejnowski, 88] Kneller, 90] Zhang et al. 92] and symbolic induction have been applied [King Sternberg, 90] Muggleton et al. 92] in the secondary structure prediction context. Despite the apparent practical importance of the secondary structure concept, the quarter of century long research efforts have shown the existence of a ....

X. Zhang, J.P. Mesirov & D.L. Waltz. Hybrid System for Protein Secondary Structure Prediction. J. Mol. Biol. 225, 1049-1063, 1992.

.... to create component classifiers include using different training parameters with a single learning method (Alpaydin 1993; Maclin Shavlik 1995) using different subsets of the training data with a single learning method (Breiman 1996a; Freund Schapire 1996) using different learning methods (Zhang, Mesirov, Waltz 1992), and explicitly searching for a set of classifiers that is both accurate and diverse (Opitz Shavlik 1996) Boosting is a method that creates component classifiers by using different subsets of the training data. Numerous methods have been suggested for combining the predictions of classifiers. ....

Zhang, X.; Mesirov, J.; and Waltz, D. 1992. Hybrid system for protein secondary structure prediction. Journal of Molecular Biology 225:1049--1063.

.... forecasting literature (Clemen, 1989; Granger, 1989) indicate that a simple averaging of the predictors generates a very good composite model; however, many later researchers (Alpaydin, 1993; Asker Maclin, 1997a, 1997b; Breiman, 1996c; Hashem, 1997; Maclin, 1998; Perrone, 1992; Wolpert, 1992; Zhang, Mesirov, Waltz, 1992) have further improved generalization with voting schemes that are complex combinations of each predictor s output. One must be careful in this case, since optimizing the combining weights can easily lead to the problem of overfitting which simple averaging seems to avoid (Sollich Krogh, 1996) ....

Zhang, X., Mesirov, J., & Waltz, D. (1992). Hybrid system for protein secondary structure prediction. Journal of Molecular Biology, 225, 1049--1063.

....factor. This result underlines why methods aimed at reducing only the variance or only the bias generally do not lead to significant improvements in overall classification performance. 5 Discussion Combining classifiers in output space has led to improved performance in many applications [12, 13, 20]. This paper concentrates on explaining the reasons for expecting such improvements and to quantify the gains achieved. Under the assumption that the a posteriori probability distributions for each class are locally monotonic functions about the decision boundaries, we showed that combining ....

X. Zhang, J.P. Mesirov, and D.L. Waltz. Hybrid system for protein secondary structure prediction. J. Molecular Biology, 225:1049--63, 1992.

....limited training data, high dimensional patterns or patterns that involve a large amount of noise, the performance of a single classifier is often unsatisfactory and or unreliable. In such cases, combining the outputs of multiple classifiers has been shown to improve the classification performance [18, 22, 35, 37, 40, 41, 43]. In this section we present two methods that use the results obtained from multiple classifiers to obtain an estimate for the Bayes error. 7 3.1 Bayes Error Estimation Based on Decision Boundaries If the outputs of the individual classifiers approximate the corresponding class posteriors, ....

X. Zhang, J.P. Mesirov, and D.L. Waltz. Hybrid system for protein secondary structure prediction. J. Molecular Biology, 225:1049--63, 1992. 29

....lines leading to f ind represent the decision of a specific classifier, while the dashed lines lead to f comb , the output of the combiner. beliefs in the Dempster Shafer sense are also available [37, 39, 50, 51] Combiners have also been successfully applied a multitude of real world problems [5, 7, 17, 25, 41, 52]. A survey of leading combining techniques, along with experimental results is given in [15, 17] Combining techniques such as majority voting can generally be applied to any type of classifier, while others rely on specific outputs, or specific interpretations of the output. For example, the ....

....with a classifier combiner. There does not seem to be a single type of network or combiner that can be labeled best under all circumstances. 7 Discussion Combining the outputs of several classifiers before making the classification decision, has led to improved performance in many applications [17, 50, 52]. This paper introduces a mathematical framework that explains the reasons for expecting such improvements and quantifies the gains achieved. We show that combining networks in output space reduces the variance in boundary locations about the optimum (Bayes) boundary decision. Moreover, the added ....

X. Zhang, J.P. Mesirov, and D.L. Waltz. Hybrid system for protein secondary structure prediction. J. Molecular Biology, 225:1049--63, 1992. 35

....0.27 3.55 .13 2 7.62 0.22 12.76 0.73 5.10 .21 RBF 4 7.62 0.22 12.36 0.50 4.95 .23 8 7.38 0.05 12.76 0.12 4.89 .20 2 6.80 0.27 9.78 0.36 BOTH 4 6.80 0.19 9.69 0.26 8 6.65 0.25 9.42 0. 30 6 DISCUSSION Combining classifiers in output space has led to improved performance in many applications [9, 14, 18]. This paper introduces a mathematical framework that explains the reasons for expecting such improvements and quantifies the gains achieved. We show that linearly combining the outputs of individual networks reduces the variance in boundary locations about the optimum boundary. Furthermore, the ....

X. Zhang, J.P. Mesirov, and D.L. Waltz. Hybrid system for protein secondary structure prediction. J. Molecular Biology, 225:1049--63, 1992.

....learns what the correct output is when level 0 generalizers give a certain output combination. Thus level 1 needs be trained on data unused in training the level 0 generalizers. Wolpert proposes to use leave one out though this is too costly and n fold cross validation seems to be better. Zhang et al. 1992) use stacking for protein secondary structure prediction with significant improvement in accuracy. In their study, aireda04.tex; 28 05 1997; 12:30; v.5; p.12 VOTING OVER MULTIPLE CONDENSED NEAREST NEIGHBORS 127 the level 0 generalizers are a statistical model, a lazy learner and a one hidden layer ....

Zhang, X., Mesirov, J. P. &Waltz, D. L. (1992). Hybrid System for Protein Secondary Structure Prediction. Journal of Molecular Biology 225: 1049--1063.

....of the behavior of stacked generalization. Jacobs (1995) reviews linear combination methods like that used in MLR. Previous work on stacked generalization, especially as applied to classification tasks, has been limited in several ways. Some only applies to a particular dataset (e.g. Zhang, Mesirov Waltz, 1992). Others report results that are less than convincing (Merz, 1995) Still others have a different focus and evaluate the results on just a few datasets (LeBlanc Tibshirani, 1993; Chan Stolfo, 1995; Kim Bartlett, 1995; Fan et al. 1996) One might consider a degenerate form of stacked ....

Zhang, X., J.P. Mesirov & D.L. Waltz (1992). Hybrid System for Protein Secondary Structure Prediction. Journal of Molecular Biology, 225, pp. 1049-1063.

....the standard ensemble approach of simply averaging the predictions of the component classifiers. 2 Background A number of researchers have demonstrated that ensembles are generally more accurate than any of their component classifiers [ Breiman, 1996; Clemen, 1989; Quinlan, 1996; Wolpert, 1992; Zhang et al. 1992 ] Figure 1 shows a basic framework for combining classifiers. Using an ensemble, the class of an example is predicted by first classifying the example with each of the component classifiers and then combining the resulting predictions into a single classification. To create an ensemble, a user ....

.... Seung, 1995 ] Examples of combination functions include voting schemes [ Hansen and Salamon, 1990 ] simple averages [ Lincoln and Skrzypek, 1989 ] weighted average schemes [ Perrone and Cooper, 1994; Rogova, 1994 ] and schemes for training combiners [ Rost and Sander, 1993; Wolpert, 1992; Zhang et al. 1992 ] Clemen (1989) demonstrated that in the absence of knowledge concerning a specific problem, almost any reasonable method, including the simple ones such as voting or using a weighted average, will result in an effective ensemble. Sequel applies a sequence of classifiers starting with its ....

X. Zhang, J. Mesirov, and D. Waltz. Hybrid system for protein secondary structure prediction. Journal of Molecular Biology, 225:1049--1063, 1992.

No context found.

X. Zhang, J.P. Mesirov, and D.L. Waltz. Hybrid system for protein secondary structure prediction. Journal of Molecular Biology, 225:1049--1063, 1992. 30

No context found.

J. Zhang et al., "A Hybrid System for Protein Secondary Structure Prediction", Journal of Molecular Biology, 225, 1992, pp. 1049-1063.

No context found.

X. Zhang, J. P. Mersirov, and D. L. Waltz, "A hybrid system for protein secondary structure prediction," J. Molecular Biol., vol. 225, pp. 1049--1063, 1992.

No context found.

X. Zhang, J.P. Mesirov, and D.L. Waltz. Hybrid System for Protein Secondary Structure Prediction. J. Mol. Biol., 225:10491063, 1992.

No context found.

X. Zhang, J.P. Mesirov, and D.L. Waltz. Hybrid system for protein secondary structure prediction. J. Molecular Biology, 225:1049--63, 1992. 30

No context found.

X. Zhang, J.P. Mesirov, and D.L. Waltz. Hybrid System for Protein Secondary Structure Prediction. J. Mol. Biol., 225:10491063, 1992.

No context found.

Zhang, X., Mesirov, J. P. & Waltz, D. L. (1992). Hybrid system for protein secondary structure prediction. J. Mol. Biol. ###, 10490-1063.

No context found.

Zhang, X., Mesirov, J.P., Waltz, D.L. Hybrid system for protein secondary structure prediction. J. Mol. Biol. 225: 1049--1063, 1992.

No context found.

, 1117--1129. Zhang, X., Mesirov, J.P., and Waltz, D.L. 1992. Hybrid system for protein secondary structure prediction. J. Mol. Biol.

No context found.

Zhang, X., Mesirov, J., and Waltz, D., Hybrid system for protein Secondary structure prediction, Journal of Molecular Biology, 225:1049--1063, 1992.

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