The Infona portal uses cookies, i.e. strings of text saved by a browser on the user's device. The portal can access those files and use them to remember the user's data, such as their chosen settings (screen view, interface language, etc.), or their login data. By using the Infona portal the user accepts automatic saving and using this information for portal operation purposes. More information on the subject can be found in the Privacy Policy and Terms of Service. By closing this window the user confirms that they have read the information on cookie usage, and they accept the privacy policy and the way cookies are used by the portal. You can change the cookie settings in your browser.
We consider a general scheme of parallel classifier combinations in the framework of statistical pattern recognition. Each statistical classifier defines a set of output variables in terms of a posteriori probabilities, i.e. it is used as a feature extractor. Unlike usual combining schemes the output vectors of classifiers are combined in parallel. The statistical Shannon information is used as a...
The aim of this paper is to propose a simple procedure that a priori determines a minimum number of classifiers to combine in order to obtain a prediction accuracy level similar to the one obtained with the combination of larger ensembles. The procedure is based on the McNemar non-parametric test of significance. Knowing a priori the minimum size of the classifier ensemble giving the best prediction...
In this paper we propose a model of neural networks ensemble consisting of a number of MLPs, that deals with an imperfect learning supervisor that occasionally produces incorrect teacher signals. It is known that a conventional unitary neural network will not learn optimally from this kind of supervisor. We consider that the imperfect supervisor generates two kinds of input-output relations, the correct...
It has been shown by several researchers that multi-classifier systems can result in effective solutions to difficult tasks. In this work, we propose a multi-classifier system based on both supervised and unsupervised learning. According to the principle of “divide-and-conquer”, the input space is partitioned into overlapping subspaces and Support Vector Machines (SVMs) are subsequently used to solve...
Mammography is a not invasive diagnostic technique widely used for early detection of breast cancer. One of the main indicants of cancer is the presence of microcalcifications, i.e. small calcium accumulations, often grouped into clusters. Automatic detection and recognition of malignant clusters of microcalcifications are very difficult because of the small size of the microcalcifications and of...
In order to determine the output from an aggregated classifier a number of methods exists. A common approach is to apply the majority-voting scheme. If the performance of the classifiers can be ranked in some intelligent way, the voting process can be modified by assigning individual weights to each of the ensemble members. For some base classifiers, like decision trees, a given node or leaf is activated...
A classifier team is used in preference to a single classifier in the expectation it will be more accurate. Here we study the potential for improvement in classifier teams designed by the feature subspace method: the set of features is partitioned and each subset is used by one classifier in the team. All partitions of a set of 10 features into 3 subsets containing (4, 4, 2) features and (4, 3, 3)...
Using an ensemble of classifiers instead of a single classifier has been shown to improve generalization performance in many machine learning problems [4, 16]. However, the extent of such improvement depends greatly on the amount of correlation among the errors of the base classifiers [1,14]. As such, reducing those correlations while keeping the base classifiers’ performance levels high is a promising...
A mathematical analogy between the process of multiple expert fusion and the tomographic reconstruction of Radon integral data is outlined for the specific instance of the combination of classifiers containing discrete data sets. Within this metaphor all conventional methods of classifier combination come, to a greater or lesser degree, to resemble the unfiltered back-projection of the constituent...
Multiple classifier methods are effective solutions to difficult pattern recognition problems. However, empirical successes and failures have not been completely explained. Amid the excitement and confusion, uncertainty persists in the optimality of method choices for specific problems due to strong data dependences of classifier performance. In response to this, I propose that further exploration...
Genetic programming (GP) can automatically fuse given classifiers of diverse types to produce a combined classifier whose Receiver Operating Characteristics (ROC) are better than [Scott et al.1998b]’s “Maximum Realisable Receiver Operating Characteristics” (MRROC). I.e. better than their convex hull. This is demonstrated on a satellite image processing bench mark using Naive Bayes, Decision Trees...
In the field of pattern recognition, multiple classifier systems based on the combination of outputs of a set of different classifiers have been proposed as a method for the development of high performance classification systems. In this paper, the problem of design of multiple classifier system is discussed. Six design methods based on the so-called “overproduce and choose“ paradigm are described...
A scheme is proposed for classifier combination at decision level which stresses the importance of classifier selection during combination. The proposed scheme is optimal (in the Neyman-Pearson sense) when sufficient data are available to obtain reasonable estimates of the join densities of classifier outputs. Four different fingerprint matching algorithms are combined using the proposed scheme to...
We describe a multiple classifier system which incorporates an automatic self-configuration scheme based on genetic algorithms. Our main interest in this paper is focused on exploring the statistical properties of the resulting multi-expert configurations. To this end we initially test the proposed system on a series of tasks of increasing difficulty drawn from the domain of character recognition...
In this paper, we present a combined classification approach called the ‘virtual test sample method’. Contrary to classifier combination, where the outputs of a number of classifiers are used to come to a combined decision for a given observation, we use multiple instances generated from the original observation and a single classifier to compute a combined decision. In our experiments, the virtual...
We present a learning algorithm for two-class pattern recognition. It is based on combining a large number of weak classifiers. The weak classifiers are produced independently with diversity. And they are combined through a weighted average, weighted exponentially with respect to their apparent errors on the training data. Experimental results are also given.
One approach to deal with real complex systems is to use two or more techniques in order to combine their different strengths and overcome each other’s weakness to generate hybrid solutions. In this project we pointed out the needs of an improved system in toxicology prediction. An architecture able to satisfy these needs has been developed. The main tools we integrated are rules and ANN. We defined...
We propose a system for a regular updating of land-cover maps based on the use of temporal series of remote sensing images. Such a system is composed of an ensemble of partially unsupervised classifiers integrated in a multiple classifier architecture. The updating problem is formulated under the complex constraint that for some images of the considered multitemporal series no ground-truth information...
This article focuses on the use of multiple classifier systems (MCSs) based on dynamic classifier selection. Four implementation strategies of MCSs are compared: majority voting, belief networks, and two designs based on dynamic classifier selection. Experimental results indicate that the direction taken by Woods et al. [1] is the best alternative for remote sensing applications for which the classifier-dependent...
The need to optimize the classification accuracy of remotely sensed imagery has led to an increasing use of Earth observation data with different characteristics collected from a variety of sensors from different parts of the electromagnetic spectrum. Combining multisource data is believed to offer enhanced capabilities for the classification of target surfaces. In the paper several single and multiple...
Set the date range to filter the displayed results. You can set a starting date, ending date or both. You can enter the dates manually or choose them from the calendar.