
Les nouveaux algorithmes de descente de gradient stochastique multi-classes parallèles visent à classer des millions d'images avec des signatures de très haute dimension en milliers de classes. Nous étendons la descente de gradient stochastique (SGD) pour les machines à vecteurs de support (SVM-SGD) de plusieurs façons pour développer le nouveau SVM-SGD multiclasse pour classer efficacement de grands ensembles de données d'image dans de nombreuses classes. Nous proposons (1) un algorithme de formation équilibré pour l'apprentissage des classificateurs binaires SVM-SGD, et (2) un processus de formation parallèle des classificateurs avec plusieurs ordinateurs/grilles multicœurs. L'évaluation sur 1000 classes d'ImageNet, ILSVRC 2010 montre que notre algorithme est 270 fois plus rapide que le classificateur linéaire de pointe LIBLINEAR.
Los nuevos algoritmos de descenso de gradiente estocástico multiclase paralelos tienen como objetivo clasificar millones de imágenes con firmas de muy alta dimensión en miles de clases. Ampliamos el descenso de gradiente estocástico (SGD) para máquinas de vectores de soporte (SVM-SGD) de varias maneras para desarrollar el nuevo SVM-SGD multiclase para clasificar de manera eficiente grandes conjuntos de datos de imágenes en muchas clases. Proponemos (1) un algoritmo de entrenamiento equilibrado para aprender clasificadores binarios SVM-SGD, y (2) un proceso de entrenamiento paralelo de clasificadores con varios ordenadores/cuadrícula multinúcleo. La evaluación de 1000 clases de ImageNet, ILSVRC 2010 muestra que nuestro algoritmo es 270 veces más rápido que el clasificador lineal de última generación LIBLINEAR.
The new parallel multiclass stochastic gradient descent algorithms aim at classifying million images with very-high-dimensional signatures into thousands of classes. We extend the stochastic gradient descent (SGD) for support vector machines (SVM-SGD) in several ways to develop the new multiclass SVM-SGD for efficiently classifying large image datasets into many classes. We propose (1) a balanced training algorithm for learning binary SVM-SGD classifiers, and (2) a parallel training process of classifiers with several multi-core computers/grid. The evaluation on 1000 classes of ImageNet, ILSVRC 2010 shows that our algorithm is 270 times faster than the state-of-the-art linear classifier LIBLINEAR.
تهدف خوارزميات الانحدار العشوائي متعددة الطبقات الموازية الجديدة إلى تصنيف ملايين الصور ذات التوقيعات عالية الأبعاد إلى آلاف الفئات. نوسع نطاق نزول التدرج العشوائي (SGD) لآلات ناقلات الدعم (SVM - SGD) بعدة طرق لتطوير SVM - SGD الجديد متعدد الطبقات لتصنيف مجموعات بيانات الصور الكبيرة بكفاءة إلى العديد من الفئات. نقترح (1) خوارزمية تدريب متوازنة لتعلم مصنفات SVM - SGD الثنائية، و (2) عملية تدريب موازية للمصنفات مع العديد من أجهزة الكمبيوتر/الشبكة متعددة النوى. يُظهر التقييم على 1000 فئة من ImageNet، ILSVRC 2010 أن خوارزميتنا أسرع 270 مرة من المصنف الخطي المتطور LIBLINEAR.
Artificial neural network, Radiology, Nuclear Medicine and Imaging, Artificial intelligence, Support vector machine, Semi-Supervised Learning, Visual Recognition, Representation Learning, Pattern recognition (psychology), Detection and Management of Retinal Diseases, Deep Learning, Stochastic gradient descent, Automated Analysis of Blood Cell Images, Artificial Intelligence, Health Sciences, Machine learning, Binary classification, Gradient descent, Transfer Learning, Computer science, Algorithm, Multiclass classification, Advances in Transfer Learning and Domain Adaptation, Computer Science, Physical Sciences, Medicine, Computer Vision and Pattern Recognition, Classifier (UML)
Artificial neural network, Radiology, Nuclear Medicine and Imaging, Artificial intelligence, Support vector machine, Semi-Supervised Learning, Visual Recognition, Representation Learning, Pattern recognition (psychology), Detection and Management of Retinal Diseases, Deep Learning, Stochastic gradient descent, Automated Analysis of Blood Cell Images, Artificial Intelligence, Health Sciences, Machine learning, Binary classification, Gradient descent, Transfer Learning, Computer science, Algorithm, Multiclass classification, Advances in Transfer Learning and Domain Adaptation, Computer Science, Physical Sciences, Medicine, Computer Vision and Pattern Recognition, Classifier (UML)
| selected citations These citations are derived from selected sources. This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | 18 | |
| popularity This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network. | Top 10% | |
| influence This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | Top 10% | |
| impulse This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network. | Top 10% |
