Abstract
We study the label complexity of pool-based active learning in the PAC model with noise. Taking inspiration from extant literature on Exact learning with membership queries, we derive upper and lower bounds on the label complexity in terms of generalizations of extended teaching dimension. Among the contributions of this work is the first nontrivial general upper bound on label complexity in the presence of persistent classification noise. © Springer-Verlag Berlin Heidelberg 2007.
Cite
CITATION STYLE
Hanneke, S. (2007). Teaching dimension and the complexity of active learning. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 4539 LNAI, pp. 66–81). Springer Verlag. https://doi.org/10.1007/978-3-540-72927-3_7
Register to see more suggestions
Mendeley helps you to discover research relevant for your work.
