Distributed optimization of multi-class SVMs.

PLoS One

Department of Computer Science, Humboldt University of Berlin, Berlin, Germany.

Published: September 2017

Training of one-vs.-rest SVMs can be parallelized over the number of classes in a straight forward way. Given enough computational resources, one-vs.-rest SVMs can thus be trained on data involving a large number of classes. The same cannot be stated, however, for the so-called all-in-one SVMs, which require solving a quadratic program of size quadratically in the number of classes. We develop distributed algorithms for two all-in-one SVM formulations (Lee et al. and Weston and Watkins) that parallelize the computation evenly over the number of classes. This allows us to compare these models to one-vs.-rest SVMs on unprecedented scale. The results indicate superior accuracy on text classification data.

Download full-text PDF

Source
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC5453486PMC
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0178161PLOS

Publication Analysis

Top Keywords

number classes
16
one-vs-rest svms
12
svms
5
distributed optimization
4
optimization multi-class
4
multi-class svms
4
svms training
4
training one-vs-rest
4
svms parallelized
4
number
4

Similar Publications

Somatic DNA Variants in Epilepsy Surgery Brain Samples from Patients with Lesional Epilepsy.

Int J Mol Sci

January 2025

Department of Neuropediatrics, Charité-Universitätsmedizin Berlin, Corporate Member of Freie Universität Berlin, Humboldt-Universität Berlin, and Berlin Institute of Health, 13353 Berlin, Germany.

Epilepsy affects 50 million people worldwide and is drug-resistant in approximately one-third of cases. Even when a structural lesion is identified as the epileptogenic focus, understanding the underlying genetic causes is crucial to guide both counseling and treatment decisions. Both somatic and germline DNA variants may contribute to the lesion itself and/or influence the severity of symptoms.

View Article and Find Full Text PDF

Class IA PI3K p110δ and p110α subunits participate in TCR and costimulatory receptor signals involved in T cell-mediated immunity, but the role of p110α is not completely understood. Here, we analyzed a mouse model of the Cre-dependent functional inactivation of p110α (kinase dead) in T lymphocytes (p110αKD-T, KD). KD mice showed increased cellularity in thymus and spleen and altered T cell differentiation with increased number of CD4CD8 DP thymocytes, enhanced proportion of CD4 SP lymphocytes linked to altered apoptosis, lower Treg cells, and increased AKT and ERK phosphorylation in activated thymocytes.

View Article and Find Full Text PDF

Genome-Wide Identification, Phylogenetic Evolution, and Abiotic Stress Response Analyses of the Late Embryogenesis Abundant Gene Family in the Alpine Cold-Tolerant Medicinal Species.

Int J Mol Sci

January 2025

Key Laboratory of Resource Biology and Biotechnology in Western China, Ministry of Education, College of Life Sciences, Northwest University, Xi'an 710069, China.

Late embryogenesis abundant (LEA) proteins are a class of proteins associated with osmotic regulation and plant tolerance to abiotic stress. However, studies on the gene family in the alpine cold-tolerant herb are still limited, and the phylogenetic evolution and biological functions of its family members remain unclear. In this study, we conducted genome-wide identification, phylogenetic evolution, and abiotic stress response analyses of family genes in species, alpine cold-tolerant medicinal herbs in the Qinghai-Tibet Plateau and adjacent regions.

View Article and Find Full Text PDF

: Myoclonus is already associated with a wide variety of drugs and systemic conditions. As new components are discovered, more drugs are suspected of causing this disabling abnormal involuntary movement. This systematic review aims to assess the medications associated with drug-induced myoclonus (DIM).

View Article and Find Full Text PDF

A comprehensive pharmacovigilance surveillance on antibacterials is lacking. This study aims to investigate safety signals of antibacterial-related adverse drug events (ADEs) with seriousness and to identify predictors of serious ADEs. This study investigated 52,503 antibacterial-induced ADEs reported to the Korea Adverse Event Reporting System Database from January 2013 to December 2022.

View Article and Find Full Text PDF

Want AI Summaries of new PubMed Abstracts delivered to your In-box?

Enter search terms and have AI summaries delivered each week - change queries or unsubscribe any time!