In this paper we present a new learning system, the "Intelligent learningmachine" (ELM). We associate intelligence withthe power to learn, forget, grow, contract, interact, and co-operate incrementally, on...
详细信息
ISBN:
(纸本)1853128066
In this paper we present a new learning system, the "Intelligent learningmachine" (ELM). We associate intelligence withthe power to learn, forget, grow, contract, interact, and co-operate incrementally, on-line, and in real time. Intelligence in the ILM is based upon the use of a specially customized weight table. the ILM enables parallel data processing and it is well suited to a wide variety of applications and promises unprecedented performance gains in dynamic environments. Here we show how Linear and Non-linear Regression and Classification modeling methods are transformed into intelligent methods. this method has now been successfully software implemented and tested using a variety of databases. Hardware implementation of the ILM is feasible and we foresee an ILM chip for faster computations and mobile applications. Subsequent papers will show how the ELM can be applied to methods such as Bayesian Models, Markov Chain, Hidden Markov Models, Linear Discriminant Analysis, Association Rules, OneR, Principal Component Analysis and Linear Support Vector machines.
the problem of classifying rarely occurring cases is faced in many real life applications. the scarcity of the rare cases makes it difficult to classify them correctly using traditional classifiers. In this paper, we ...
详细信息
ISBN:
(纸本)0769521428
the problem of classifying rarely occurring cases is faced in many real life applications. the scarcity of the rare cases makes it difficult to classify them correctly using traditional classifiers. In this paper, we propose a new approach to use emerging patterns (EPs) [3] and decision trees (DTs) in rare-class classification (EPDT). EPs are those itemsets whose supports in one class are significantly higher than their supports in the other classes. EPDT employs the power of EPs to improve the quality of rare-case classification. To achieve this aim, we first introduce the idea of generating new non-existing rare-class instances, and then we over-sample the most important rare-class instances. Our experiments show that EPDT outperforms many classification methods.
Recent research in machinelearning, datamining and related areas has produced a wide variety of algorithms for cost-sensitive (CS) classification, where instead of maximizing the classification accuracy, minimizing ...
详细信息
ISBN:
(纸本)0769521428
Recent research in machinelearning, datamining and related areas has produced a wide variety of algorithms for cost-sensitive (CS) classification, where instead of maximizing the classification accuracy, minimizing the misclassfication cost becomes the objective. However, these methods assume that training sets do not contain significant noise, which is rarely the case in real-world environments. In this paper, we systematically study the impacts of class noise on CS learning, and propose a cost-guided class noise handling algorithm to identify noise for effective CS learning. We call it Cost-guided Iterative Classification Filter (CICF), because it seamlessly integrates costs and an existing Classification Filter [1] for noise identification. Instead of putting equal weights to handle noise in all classes in existing efforts, CICF puts more emphasis on expensive classes, which makes it especially successful in dealing withdatasets with a large cost-ratio. Experimental results and comparative studies from real-world datasets indicate that the existence of noise may seriously corrupt the performance of CS classifiers, and by adopting the proposed CICF algorithm, we can significantly reduce the misclassfication cost of a CS classifier in noisy environments.
In manufacturing as well as other application areas there is a need to learn standard operating conditions in order to detect future changes or deviations. this is related to the even more general problem of detecting...
详细信息
ISBN:
(纸本)1853128066
In manufacturing as well as other application areas there is a need to learn standard operating conditions in order to detect future changes or deviations. this is related to the even more general problem of detecting instances (cases, records) that are unusual compared to the bulk of the data (outliers). Examples of the problem are fault detection in chemical engineering and statistical process control. the outlier problem is ubiquitous. If specific deviations are not a priori specified, this is a type of unsupervised learning problem. the focus here is on the important, practical case for modem data environments. that is, training data with multiple (usual many) variables of mixed types (without the expedient assumptions common in statistics of multivariate normality that rarely holds in practice). An elegant technique is used to transform an unsupervised leaming problem to a supervised one. this methodology uses an artificial reference distribution. For the focus here such a specific reference distribution requires appropriate properties. then an effective, universal, and nonparametric supervised learner (a gradient boosting machine) is applied to the transformed problem. the results are then in a sense inverted to the original problem. Extensions are mentioned as well as additional insight that becomes available. An illustrative example is presented to justify the validity of this generic and general methodology.
Classification learning is dominated by systems which induce large numbers of small axis-orthogonal decision surfaces which biases such systems towards particular hypothesis types. However, there is reason to believe ...
详细信息
ISBN:
(纸本)1853128066
Classification learning is dominated by systems which induce large numbers of small axis-orthogonal decision surfaces which biases such systems towards particular hypothesis types. However, there is reason to believe that many domains have underlying concepts which do not involve axis orthogonal surfaces. Further, the multiplicity of small decision regions mitigates against any holistic appreciation of the theories produced by these systems, notwithstanding the fact that many of the small regions are individually comprehensible. We propose the use of less strongly biased hypothesis languages which might be expected to model concepts using a number of structures close to the number of actual structures in the domain. An instantiation of such a language, a convex hull based classifier, CH1, has been implemented to investigate modeling concepts as a small number of large geometric structures in n-dimensional space. A comparison of the number of regions induced is made against other well-known systems on a representative selection of largely or wholly continuous valued machinelearning tasks. the convex hull system is shown to produce a number of induced regions about an order of magnitude less than well-known systems and very close to the number of actual concepts. this representation, as convex hulls, allows the possibility of extraction of higher level mathematical descriptions of the induced concepts, using the techniques of computational geometry.
Feature selection is a key problem to patternrecognition and machinelearning, and it is difficult to get the optimal feature subset for its NP-hand. Currently, the dimensionality of feature set or instance set is ve...
详细信息
ISBN:
(纸本)0780384032
Feature selection is a key problem to patternrecognition and machinelearning, and it is difficult to get the optimal feature subset for its NP-hand. Currently, the dimensionality of feature set or instance set is very high in many applications, such as information retrieval, so the feature selection from high-dimensional data is also an urgent task for researchers. the paper presents a new approach, which is a two-level filter model system integrateing the ReliefF and a newly developed algorithm of feature cluster, to reduce the dimensionality of large-scale feature set via the feature correlation(relevance) including the feature-feature correlation and feature-class correlation. Our major contributions are: (1) to present a system to perform feature selection from high-dimensional data;(2) to analyze the change of system architecture according to the time cost of the parts in the system;(3) to summarize and comment on the calculations of feature correlation;(4)to perform experiments to show the effective of the proposed approach, which has shown that the system can efficiently get a better compromise between dimensionality reduction and accuracy rate of classification than just part of the system. In many cases, it can improve the accuracy rate and dimensionality reduction.
We describe here the three main platforms in the ERIM family of Web-based environments for human interpreting, two of them in more details, ERIM-Interp and ERIM-Collect, then ERIM-Aid. Each platform supports an aspect...
详细信息
ISBN:
(纸本)2951740816
We describe here the three main platforms in the ERIM family of Web-based environments for human interpreting, two of them in more details, ERIM-Interp and ERIM-Collect, then ERIM-Aid. Each platform supports an aspect of the collecting or study of spontaneous bilingual dialogues, translated by an interpreter. ERIM-Interp is the core environment, providing mediated communication between speakers and human interpreters over the network. Using ERIM-Collect, French-Chinese interpreting data have been collected within the 3-year "ChinFaDial" project supported by LIAMA, a French-Chinese laboratory in Beijing. these "raw" speech data will be made available in the spring of 2004 on an open-access basis, using the DistribDial server, on a CLIPSGETA website. Our goal is to extend such corpora, on a collaborative scheme, to allow other research groups to contribute to the site whatever annotations they may have created, and to share them under the same conditions (GPL). An ERIM-Aid variant is intended to provide focused machine AIDS to Web-based human interpreters, or to monolingual distant speakers conversing in different languages.
"learning with side-information" is attracting more and more attention in machinelearning problems. In this paper, we propose a general iterative framework for relevant linear feature extraction. It efficie...
详细信息
ISBN:
(纸本)0769521282
"learning with side-information" is attracting more and more attention in machinelearning problems. In this paper, we propose a general iterative framework for relevant linear feature extraction. It efficiently utilizes boththe side-information and unlabeled data to enhance gradually algorithms' performance and robustness. Both good relevant feature extraction and reasonable similarity matrix estimation can be realized. Specifically, we adopt relevant component analysis (RCA) under this framework and get the derived iterative self-enhanced relevant component analysis (ISERCA) algorithm. the experimental results on several data sets show that ISERCA outperforms RCA.
For the purpose of gene identification, we propose an approach to gene expression dataminingthat uses a combination of unsupervised and supervised learning techniques to search for useful patterns in the data. the a...
详细信息
In this paper, we are interested in the sender's name extraction in fax cover pages through a machinelearning scheme. For this purpose, two analysis methods are implemented to work in parallel. the first one is b...
详细信息
In this paper, we are interested in the sender's name extraction in fax cover pages through a machinelearning scheme. For this purpose, two analysis methods are implemented to work in parallel. the first one is based on image document analysis (OCR recognition, physical block selection), the other on text analysis (word feature extraction, local grammar rules). Our main contribution consisted in introducing a neural network to find an optimal combination of the two approaches. Tests carried on real fax images show that the neural network improves performance compared to an empirical combination function and to each method used separately.
暂无评论