Showing papers by "Zhi-Hua Zhou published in 2002"

Neural Networks for Multi-Instance Learning

Abstract: Multi-instance learning originates from the investigation on drug activity prediction, where the task is to predict whether an unseen molecule could be used to make some drug. Such a problem is difficult because a molecule may have many alternative shapes with low energy, yet only one of those shapes may be responsible for the qualification of the molecule to make the drug. Because of its unique characteristics and extensive existence, multi-instance learning is regarded as a new machine learning framework parallel to supervised learning, unsupervised learning, and reinforcement learning. In this paper, an open problem of this area is addressed. That is, a popular neural network algorithm is adapted for multi-instance learning through employing a specific error function. Experiments show that the adapted algorithm achieves good result on the drug activity prediction data.

Hybrid decision tree

Abstract: In this paper, a hybrid learning approach named hybrid decision tree (HDT) is proposed. HDT simulates human reasoning by using symbolic learning to do qualitative analysis and using neural learning to do subsequent quantitative analysis. It generates the trunk of a binary HDT according to the binary information gain ratio criterion in an instance space defined by only original unordered attributes. If unordered attributes cannot further distinguish training examples falling into a leaf node whose diversity is beyond the diversity-threshold, then the node is marked as a dummy node. After all those dummy nodes are marked, a specific feedforward neural network named Fannc that is trained in an instance space defined by only original ordered attributes is exploited to accomplish the learning task. Moreover, this paper distinguishes three kinds of incremental learning tasks. Two incremental learning procedures designed for example-incremental learning with different storage requirements are provided, which enables HDT to deal gracefully with data sets where new data are frequently appended. Also a hypothesis-driven constructive induction mechanism is provided, which enables HDT to generate compact concept descriptions.

Ensemble of GA based Selective Neural Network Ensembles

Abstract: Neural network ensemble is a learning paradigm where several neural networks are jointly used to solve a problem. In this paper, e-GASEN, a two-layer neural network ensemble architecture is proposed, in which the base learners of the final ensemble are also ensembles. Experimental results show that e-GASEN generalizes better than a popular ensemble method. The reason why e-GASEN works is also discussed. We believe that the different layers of e-GASEN attain good generalization ability for different reasons. The first layer ensembles profit from the selected individual neural networks that are moderately divergent but generalize well, while the second layer ensemble profits from the divergency among the first layer ensembles.

Selectively ensembling neural classifiers

Abstract: Ensembling neural classifiers can significantly improve the generalization ability of classification systems. In this paper, GASEN, a genetic algorithm based selective ensemble method, that has been shown to be excellent in ensembling neural regressors, is applied to neural classifiers. Experiments on four large data sets show that this method can generate ensembles of neural classifiers with stronger generalization ability than those generated by Bagging, Adaboost, or Arc-x4.

Rule learning based on neural network ensemble

Abstract: A neural network ensemble can significantly improve the generalization ability of neural network-based systems. In this paper, a novel rule-learning algorithm is proposed where the neural network ensemble acts as a front-end processor that generates data for the learning of rules. Experimental results show that the proposed algorithm can generate rules with strong generalization ability.

The Application of Visualization and Neural Network Techniques in a Power Transformer Condition Monitoring System

Abstract: In this paper, visualization and neural network techniques are applied together to a power transformer condition monitoring system. Through visualizing the data from the chromatogram of oil-dissolved gases by 2-D and/or 3-D graphs, the potential failures of the power transformers become easy to be identified. Through employing some specific neural network techniques, the data from the chromatogram of oildissolved gases as well as those from the electrical inspections can be effectively analyzed. Experiments show that the described system works quite well in condition monitoring of power transformers.

ALMA: a new approach to data initial association

Abstract: An adaptive limited memory approach (ALMA) is proposed to deal with the data initial association problem in tracking-while-scanning systems in high clutter environments. By using measurements in the window limited memory and exploiting the target detection probability and receiver noise sampling information, the initial track number is estimated. It in turn can be adopted to adjust the initial correlation region and the normalized residual square threshold adaptively; many false initial trajectories are therefore restrained. Computer simulation demonstrates that the real time data initial association problem can be settled by a combination of radar's dynamic characteristics and limited window measurements.