Online One-Class SVMs with Active-Set Optimization for Data Streams

Gao, Katelyn

doi:10.1109/icmla.2015.101

Cited by 11 publications

(14 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This approach has several advantages. First, the hyperplane obtained by our approach is guaranteed to be optimal unlike the previous approaches for SVM (Drineas and Mahoney 2005;Gao 2015;Gómez-Verdejo et al 2011;Noumir, Honeine, and Richard 2012;Rahimi and Recht 2007). This is because our approach checks whether the obtained hyper-plane is theoretically identical to the optimal hyper-plane by exploiting the upper and lower bounds.…”

Section: Main Ideasmentioning

confidence: 97%

“…In the datasets, each column vector of data matrix X is standardized to have mean zero and variance one by following the previous papers (Ç eker and Upadhyaya 2016; Fisher, Camp, and Krzhizhanovskaya 2016). We compared our approach to the Nyström method (Drineas and Mahoney 2005), random Fourier features (Rahimi and Recht 2007), the coherence criterion-based approach (Noumir, Honeine, and Richard 2012), the activeset method-based approach (Gao 2015), and the shrinking approach (Joachims 1999). In the experiments, we set the number of landmark to 0.01•n for the Nyström method.…”

Section: Experimental Evaluationmentioning

confidence: 99%

“…Our approach can use SMO variants since it prunes data points before applying the solver as shown in Algorithm 2. Similarly, SMO variants can be used in the Nyström method (Drineas and Mahoney 2005), random Fourier features (Rahimi and Recht 2007), the active-set method-based approach (Gao 2015), and the shrinking approach (Joachims 1999). However, since the coherence criterion-based approach (Noumir, Honeine, and Richard 2012) uses the incremental approach, it cannot use SMO variants.…”

Section: Experimental Evaluationmentioning

confidence: 99%

“…In terms of efficiency, this approach outperforms the adaptive approach (Gómez-Verdejo et al 2011). Gao proposed the active-set method based online approach for one-class SVM to improve the efficiency (Gao 2015). It applies the solver to data points only if their distance to the center of the hyper-plane is not less than the radius of the hyper-plane.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Efficient Data Point Pruning for One-Class SVM

Fujiwara

Kanai

Arai

et al. 2019

AAAI

View full text Add to dashboard Cite

One-class SVM is a popular method for one-class classification but it needs high computation cost. This paper proposes Quix as an efficient training algorithm for one-class SVM. It prunes unnecessary data points before applying the SVM solver by computing upper and lower bounds of a parameter that determines the hyper-plane. Since we can efficiently check optimality of the hyper-plane by using the bounds, it guarantees the identical classification results to the original approach. Experiments show that it is up to 6800 times faster than existing approaches without degrading optimality.

show abstract

Section: Main Ideasmentioning

confidence: 97%

Section: Experimental Evaluationmentioning

confidence: 99%

Section: Experimental Evaluationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Efficient Data Point Pruning for One-Class SVM

Fujiwara

Kanai

Arai

et al. 2019

AAAI

View full text Add to dashboard Cite

show abstract

“…Online one‐class SVM with active set (Gao, ) is another approach to one‐class classification on data stream. In this method, active‐set method is used for updating classifier model dealing with quadratic programming.…”

Section: Related Workmentioning

confidence: 99%

Incremental one‐class classification on stationary data stream using two‐quarter sphere

Ghomanjani

Hamidzadeh

2018

Expert Systems

View full text Add to dashboard Cite

Data stream is a sequence of data that has unique features. In many data streams, data from one concept is available, and detection of other types of data in data stream is an essential task. One‐class classification is a famous approach to data classification when data from one class is accessible. The principal task of one‐class classification is separating input data in two different parts: target and outlier data. One of the main challenges concerning data classification on the stationary data streams is the response time, which causes negative effects on effective runtime. Most classifiers, that have been proposed on Rd feature space, solve a quadratic problem for classification that leads to extreme runtime increase. In this paper, an incremental one‐class classification on stationary data streams is proposed using two‐quarter sphere (IOCTQ) in order to achieve lower computation cost of classification time, and a linear optimization problem is solved. IOCTQ divides one data classification problem into two data classifiers, and each data point will be individually classified in one‐quarter sphere. The two‐quarter spheres can be run parallel. The results of the experiments have been compared with state‐of‐the‐art methods and show superiority of the IOCTQ method in classification accuracy and time complexity.

show abstract

A cyber‐physical management system for medium‐scale solar‐powered data centers

Dikaiakos

Chatzigeorgiou

Tryfonos

et al. 2023

Concurrency and Computation

View full text Add to dashboard Cite

Summary The effort to reduce the environmental impact and carbon footprint of data‐center operations has led to the emergence of “green” data centers, which are designed to reduce energy consumption and to increase their use of Renewable Energy Sources (RES). Despite the advances demonstrated by hyper‐scale facilities in energy efficiency and the use of green energy, small and medium‐scale data centers, which contribute to over 50% of the total electricity consumption and carbon emissions of the sector, face significant challenges in the adoption and exploitation of RES. In this article, we present the steps taken to transform a medium‐scale, academic data center into a “green” one that uses solar power. In particular, we describe the design and implementation of: (i) a collocated photovoltaic facility and (ii) a cyber‐physical system comprising IoT sensor devices, a microservices platform, and a visualization and analytics dashboard that supports the configuration and monitoring of the infrastructure. Using data collected from the platform and dashboard, we show the environmental and financial advantages derived from this transformation, and the potential that arises from the availability of integrated operational data.

show abstract

Online One-Class SVMs with Active-Set Optimization for Data Streams

Cited by 11 publications

References 8 publications

Efficient Data Point Pruning for One-Class SVM

Efficient Data Point Pruning for One-Class SVM

Incremental one‐class classification on stationary data stream using two‐quarter sphere

A cyber‐physical management system for medium‐scale solar‐powered data centers

Contact Info

Product

Resources

About