Simpler core vector machines with enclosing balls

Tsang, Ivor W.; Kocsor, András; Kwok, James T.

doi:10.1145/1273496.1273611

Cited by 156 publications

(95 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The average number of examples for submodel is denoted bȳ m K . The datasets Kdd-full, Ijcnn and extended Usps (abbreviated as Usps-ext) were used as in previous research to test the large-scale capabilities of CVMs [18] and are available at [17]. The other problems are available at [7] or [2].…”

Section: Experiments and Conclusionmentioning

confidence: 99%

“…For datasets Kdd-full, Usps-ext and Ijcnn we used the hyper-parameter values reported at [18,16]. For the small datasets (≤ 10 4 examples) hyper-parameters were determined using 10-fold cross-validation.…”

Section: Experiments and Conclusionmentioning

confidence: 99%

“…For the remaining datasets σ 2 was set to the average squared distance among training patterns and C was determined on a logarithmic grid [2 0 , 2 12 ] using a validation-set (30% of the training-set). In order to compute I 0 and α 0 for Algorithm 1 we adopted the initialization procedure of [17]. We also adopted the LRR caching strategy designed in [17] for CVMs to avoid the computation of recently used kernel values.…”

Section: Experiments and Conclusionmentioning

confidence: 99%

“…In order to compute I 0 and α 0 for Algorithm 1 we adopted the initialization procedure of [17]. We also adopted the LRR caching strategy designed in [17] for CVMs to avoid the computation of recently used kernel values. Tables 1 and 2 summarize the experimental results obtained after training with the parameters determined by the model-selection procedure.…”

Section: Experiments and Conclusionmentioning

confidence: 99%

See 3 more Smart Citations

A New Algorithm for Training SVMs Using Approximate Minimal Enclosing Balls

Frandi

Gasparo

Lodi

et al. 2010

Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications

View full text Add to dashboard Cite

Abstract.It has been shown that many kernel methods can be equivalently formulated as minimal enclosing ball (MEB) problems in a certain feature space. Exploiting this reduction, efficient algorithms to scale up Support Vector Machines (SVMs) and other kernel methods have been introduced under the name of Core Vector Machines (CVMs). In this paper, we study a new algorithm to train SVMs based on an instance of the Frank-Wolfe optimization method recently proposed to approximate the solution of the MEB problem. We show that, specialized to SVM training, this algorithm can scale better than CVMs at the price of a slightly lower accuracy.

show abstract

Section: Experiments and Conclusionmentioning

confidence: 99%

Section: Experiments and Conclusionmentioning

confidence: 99%

Section: Experiments and Conclusionmentioning

confidence: 99%

Section: Experiments and Conclusionmentioning

confidence: 99%

See 2 more Smart Citations

A New Algorithm for Training SVMs Using Approximate Minimal Enclosing Balls

Frandi

Gasparo

Lodi

et al. 2010

Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications

View full text Add to dashboard Cite

show abstract

“…Although conceptually simple, a sophisticated numerical solver is required for its implementation, which is computationally expensive when applying on large-scale problems with very large core-sets. After that, the Simpler Core Vector Machines (SCVM) [15] replace the numerical solver with an iterative algorithm, which results in a faster training than CVM with comparable accuracy on massive data sets. However, the Enclosing Ball (EB) problem solved in SCVM requires the ball's radius to be fixed, which limits the SCVM to be only feasible for certain kernels that satisfy…”

Section: Bv Data Reduction-based Algorithmmentioning

confidence: 99%

New Intelligent Classification Method Based On Improved Meb Algorithm

Wang

Liu

2014

International Journal on Smart Sensing and Intelligent Systems

View full text Add to dashboard Cite

Abstract-We present a simple approximate algorithm to compute the Minimum Enclosing Ball (MEB) of training samples in high dimensional Euclidean space. We prove theoretically that the proposed algorithm converges to the optimum within any precision quickly. Compared to popular MEB algorithms, it has the competitive performances on both training time and accuracy. Besides, the proposed algorithm does not need any extra requirement on kernels, it can be linked with extensive kernel methods, consequently. We also use the proposed algorithm to handle Binary Classification, Multi-class Classification, and Image Clustering problems. Experiments on both synthetic and realworld data sets demonstrate the validity of the algorithm we proposed.

show abstract

A new BVM based approach to transient security assessment

Mohammadi

Gharehpetian

Raoofat

2010

Int Trans Elec Energy Syst

View full text Add to dashboard Cite

SUMMARYMachine Learning techniques have been extensively used in Power Systems Analysis during the last years. This paper describes a ball vector machine based algorithm for on-line transient security assessment of large-scale power systems. The proposed ball vector machine based security assessment reduces the training time and space complexities in comparison with support vector machines, artificial neural networks, and other machine learning based algorithms. In addition, the proposed algorithm has less support vectors and therefore is faster than existing algorithms for on-line applications. A new decision tree based feature selection technique has been presented, too. The proposed ball vector machine based algorithm has been applied to New England 39-bus test system. The simulation results show the effectiveness of the proposed algorithm for on-line transient security assessment procedure.

show abstract

Simpler core vector machines with enclosing balls

Cited by 156 publications

References 13 publications

A New Algorithm for Training SVMs Using Approximate Minimal Enclosing Balls

A New Algorithm for Training SVMs Using Approximate Minimal Enclosing Balls

New Intelligent Classification Method Based On Improved Meb Algorithm

A new BVM based approach to transient security assessment

Contact Info

Product

Resources

About