آی اس آی پیپر

دانلود رایکان مقاله انگلیسی ISI با موضوع کاهش تعداد طبقه بندی کننده های فرعی برای ماشین های بردار پشتیبانی چند مقوله ای

عنوان فارسی مقاله:

کاهش تعداد طبقه بندی کننده های فرعی برای ماشین های بردار پشتیبانی چند مقوله ای دو به دو

عنوان انگلیسی مقاله:

Reducing the number of sub-classifiers for pairwise multi-category support vector machines

دانلود رایگان مقاله ISI با فرمت PDF:

مشاهده توضیحات کامل و خرید ترجمه فارسی با فرمت ورد تایپ شده:

بخشی از مقاله انگلیسی :

3. Uncertainty sampling

Before introducing our new method, we will review the uncertainty sampling strategy (Lewis and Gale, 1994) firstly. The uncertainty sampling strategy is an important sampling selecting strategy used in active learning. Active learning (Simon and Lea, 1974; Winston, 1975) is an effi- cient supervised learning algorithm that actively selects ‘‘helpful’’ samples to learn, instead of learning from the original training set passively. The uncertainty sampling strategy is used to select the ‘‘helpful’’ samples by measuring their uncertainty to the current classifier. A typical active learning framework is described in (Tong, 2001). In active learning, the whole data are divided into labeled samples X and unlabeled samples U. There is also a learner l and a deciding module q. The learner l is trained on the labeled samples X and the module q is used to decide which samples of U should be selected and labeled, and should be added into X. The updated X will be used to train l in the next step. According to the difference mechanism of deciding modules, active learning methods can be divided into two groups: uncertainty sampling and query by committee (QBC) Seung et al. (1992). The main idea of uncertainty sampling is that a classifier will benefit more from being trained on samples, which it is more uncertain to current classifier. Uncertainty sampling requires a probabilistic classifier that assigns to unlabeled samples each possible label with a certain probability. The unlabeled samples with most uncertainty are selected and labeled, and then are added into X. Various methods for measuring uncertainty have been proposed Lewis and Gale (1994), Iyengar et al. (2000). Query by committee is another group of active learning methods. It is based on the disagreement among a committee of classifiers.