|
As Daniel Hsu and John Langford pointed out in a recent blog post, there are now practical algorithms for active learning which are competitive with supervised learning and dramatically reduce the number of examples needed to learn. In which papers are these algorithms described? The vowpal wabbit tutorial references a paper as BHLZ’10 and another as BDL’09. What are they? Which other papers should one read to familiarize oneself with this line of research? |
|
BHLZ’10 is Agnostic Active Learning Without Constraints and BDL’09 is Importance Weighted Active Learning. Have you seen Burr Settles' active learning survey? It's pretty up-to-date (was out in 2010), and has plenty of references. |
|
I think Leon Bottou's LaSVM is one of the bigger projects: http://leon.bottou.org/projects/lasvm (paper reference on the page) |
