Statistical learning theory

About: Statistical learning theory is a research topic. Over the lifetime, 1618 publications have been published within this topic receiving 158033 citations.

Segmentation of Images Corrupted by Noise Based on Fuzzy Weighted Support Vector Machine Approach

Abstract: Image segmentation is critical to computer vision. Support vector machine approach is considered a good candidate because of its good generalization performance, especially when the number of training samples is very small and the dimension of feature space is very high. The presented paper proposes the fuzzy weighted support vector machine approach for segmentation of images corrupted by noise. Experimental results show that the fuzzy weighted support vector machine approach is more robust than classical support vector machine approach.

Design and Realization of SVM Topic Crawler Based on Incremental Learning

Abstract: Using information technology such as search engine for collecting and monitoring of network public opinion is a practical and effective method. This paper puts forward an improved algorithm of SVM classifier based on incremental learning online, and then implements a topic crawler system for network public opinion and to grab the public opinion. SVM (Support Vector Machine) is first proposed by Vapnik in 1995. It exhibits many unique advantages in addressing the small sample, nonlinear and high dimensional pattern recognition. SVM pattern recognition is based on statistical learning theory method has important applications in computer pattern recognition field. The SVM research is not perfect, cannot efficiently solve pattern recognition problems. But with the study of the application of statistical learning theory and neural networks than the new machine learning methods encounter some significant difficulties, such as how to determine the issue of network structure, through learning and learning problems due to local minima problems, leading to many researchers added to the study of SVM classification algorithm to improve. This effectively promotes the rapid development of SVM classification algorithm and continuous improvement, and SVM classifier was quickly extended to other machine learning problems fitting in function today SVM classification algorithm on text categorization has been successful applications. Since the 1990s, the rapid development of Internet technology, automatic text classification research has entered a new stage, based on machine learning text classification technology gradually replaced the method based on knowledge engineering automatic text classification has become the main form. Carried out in comparison with the classification Bayesian k-nearest neighbor and decision tree, the support vector machine method achieved the best classification accuracy since more and more researchers began to pay attention to them, and for the support of two standard corpus SVM and text classification were studied and put forward a number of new methods. In recent years, the introduction of SVM classifier topic crawlers, used to guide and supervise the theme crawler got the attention of many scholars. Johnson first SVM classification algorithm supervision focused crawler conducted theoretical research, proposed a SVM classifier model to guide the crawling reptile theme, and a lot of related experiments. More and more scholars began to use support vector machine guidance and supervision topic reptiles, Michelangelo and other support vector machine to guide the theme reptiles, and proposes a support vector machine classification algorithm page. Topic relevance determination method of obtaining a direct impact on the rate of theme crawler, traditional themes reptiles crawling in the website, acquisition rate has been low. Based on SVM classifier, this paper further studies SVM classification algorithm in the prediction of the page subject classification and proposes an incremental learning of SVM classification algorithm, and finally applies it to the network crawl on public opinion.

A performance bound of the multi-output extreme learning machine classifier

Abstract: This paper concerns estimation of generalization performance of the multi-output extreme learning machine classifier (M-ELM) in the framework of statistical learning theory. The performance bound is derived under the assumption that the expectation of the extreme learning machine kernel exists. We first show that minimizing the least square error is equal to minimizing an upper bound of the error concerning the margin of M-ELM in the training set, which implies that M-ELM ends up with high confidence after training. Afterwards, we derive the bound based on the margin of M-ELM and the empirical Rademacher complexity. The bound not only gives a theoretical explanation of good performance of M-ELM especially in the small-sample cases, but also shows that the performance of M-ELM is insensitive to the number of hidden nodes, which is consistent with previous experimental results. The bound also offers an insight that the performance of M-ELM is not significantly affected by the number of classes, which proves the effectiveness of the learning process of M-ELM.

State Trend Prediction of Spacecraft Using PSO-SVR

Abstract: Fault prediction is the core content and crucial technology for health monitoring of the in-orbit spacecraft, and predicting the future trend of telemetry data is the prerequisite and basis for fault prediction. This paper presents a state trend prediction method for spacecraft based on particle swarm optimization (PSO) and support vector regression (SVR). The method applies SVR to construct a regression prediction model of telemetry data. SVR is a learning procedure based on statistical learning theory, which employs the training data to build an excellent prediction model in the situations of small sample. The complexity and generalization performance of the SVR model is influenced by its training parameters. In this paper, PSO is applied to optimize the parameters of SVR model. The results show that the method is efficient and practical for telemetry data prediction of the in-orbit spacecraft.

Linear Programming Regressive Support Vector Machine

Abstract: Based on the analysis of the general norm in structure risk to control model complexity for regressive problem, two kinds of linear programming support vector machine corresponding to l1-norm and linfin-norm are presented including linear and nonlinear SVMs. A numerical experiment has been done for these two kinds of linear programming support vector machines and classic support vector machine by artificial data. Simulation results show that the generalization performance of this two kind linear programming SVM is similar to classic one, l1-SVM has less number of support vectors and faster learning speed, and learning result is not sensitive to learning parameters