Improving Generalization with Active Learning

@article{cohnEtAl1992:activeLearning, abstract = {Active learning differs from “learning from examples” in that the learning algorithm assumes at least some control over what part of the input domain it receives information about. In some situations, active learning is provably more powerful than learning from examples alone, giving better generalization for a fixed number of training examples. In this article, we consider the problem of learning a binary concept in the absence of noise. We describe a formalism for active concept learning called selective sampling and show how it may be approximately implemented by a neural network. In selective sampling, a learner receives distribution information from the environment and queries an oracle on parts of the domain it considers “useful.” We test our implementation, called an SG-network, on three domains and observe significant improvement in generalization.}, address = {Hingham, MA, USA}, author = {Cohn, David and Atlas, Les and Ladner, Richard }, citeulike-article-id = {1727805}, doi = {http://dx.doi.org/10.1023/A:1022673506211}, issn = {0885-6125}, journal = {Mach. Learn.}, keywords = {active\_learning, ai, pbd}, month = {May}, number = {2}, pages = {201--221}, posted-at = {2008-05-23 00:20:43}, priority = {2}, publisher = {Kluwer Academic Publishers}, title = {Improving Generalization with Active Learning}, url = {http://dx.doi.org/10.1023/A:1022673506211}, volume = {15}, year = {1994} }

TY - JOUR ID - cohnEtAl1992:activeLearning L3 - citeulike-article-id:1727805 TI - Improving Generalization with Active Learning JF - Mach. Learn. VL - 15 IS - 2 SP - 201 EP - 221 AD - Hingham, MA, USA PB - Kluwer Academic Publishers SN - 0885-6125 N2 - Active learning differs from “learning from examples” in that the learning algorithm assumes at least some control over what part of the input domain it receives information about. In some situations, active learning is provably more powerful than learning from examples alone, giving better generalization for a fixed number of training examples. In this article, we consider the problem of learning a binary concept in the absence of noise. We describe a formalism for active concept learning called selective sampling and show how it may be approximately implemented by a neural network. In selective sampling, a learner receives distribution information from the environment and queries an oracle on parts of the domain it considers “useful.” We test our implementation, called an SG-network, on three domains and observe significant improvement in generalization. KW - active_learning KW - ai KW - pbd AU - Cohn, David AU - Atlas, Les AU - Ladner, Richard PY - 1994/05// UR - http://dx.doi.org/10.1023/A:1022673506211 ER -