Active Learning. Burr Settles

Чтение книги онлайн.

СКАЧАТЬ based on available training data (black dots). The bottom row plots the network’s output variance across the input range, which is used to select the query for the next iteration.

2.5 DISCUSSION

      Uncertainty sampling is possibly the most popular active learning strategy in practice. Perhaps this is because of its intuitive appeal combined with the ease of implementation. Most of the common uncertainty-based utility measures do not require significant engineering overhead to use. In fact, as long as the learner can provide a confidence or probability score along with its predictions, any of the measures in Section 2.3 can be employed with the learner as a “black box.” Standard classification or inference procedures can be used, leaving the choice of learning algorithm fairly modular. This is not necessarily the case for all active learning approaches. Furthermore, if inference is fast and tractable, then querying should also be fast and tractable.

      Our discussion of uncertainty sampling has, thus far, been limited to the pool-based setting where the learner selects the “best” query from the set of unlabeled data U. Uncertainty sampling can also be employed in the stream-based selective sampling setting, where unlabeled instances are drawn x ∽ D_X one at a time from an input distribution, and the learner decides on the spot whether to query or discard it. The simplest way to implement uncertainty-based selective sampling is to set a threshold on the uncertainty measure and use this to define a region of uncertainty. An instance is queried if it falls within the region of uncertainty, and discarded otherwise. The learner is re-trained after each new query instance (or batch of instances) is labeled and added to L.

      Конец ознакомительного фрагмента.

      Текст предоставлен ООО «ЛитРес».

      Прочитайте эту книгу целиком, купив полную легальную версию на ЛитРес.

      Безопасно оплатить книгу можно банковской картой Visa, MasterCard, Maestro, со счета мобильного телефона, с платежного терминала, в салоне МТС или Связной, через PayPal, WebMoney, Яндекс.Деньги, QIWI Кошелек, бонусными картами или другим удобным Вам способом.

/9j/4AAQSkZJRgABAQEASABIAAD/2wBDAAUDBAQEAwUEBAQFBQUGBwwIBwcHBw8LCwkMEQ8SEhEP ERETFhwXExQaFRERGCEYGh0dHx8fExciJCIeJBweHx7/2wBDAQUFBQcGBw4ICA4eFBEUHh4eHh4e Hh4eHh4eHh4eHh4eHh4eHh4eHh4eHh4eHh4eHh4eHh4eHh4eHh4eHh4eHh7/wAARCAKIAgsDASIA AhEBAxEB/8QAHwAAAQUBAQEBAQEAAAAAAAAAAAECAwQFBgcICQoL/8QAtRAAAgEDAwIEAwUFBAQA AAF9AQIDAAQRBRIhMUEGE1FhByJxFDKBkaEII0KxwRVS0fAkM2JyggkKFhcYGRolJicoKSo0NTY3 ODk6Q0RFRkdISUpTVFVWV1hZWmNkZWZnaGlqc3R1dnd4eXqDhIWGh4iJipKTlJWWl5iZmqKjpKWm p6ipqrKztLW2t7i5usLDxMXGx8jJytLT1NXW19jZ2uHi4+Tl5ufo6erx8vP09fb3+Pn6/8QAHwEA AwEBAQEBAQEBAQAAAAAA

СКАЧАТЬ