Training Log Linear Models using Smoothed Hamming Loss


In a paper that is to appear in NIPS this year [1], we proposed a new objective function for training Conditional Random Fields (CRFs). When using the traditional log-likelihood training, the training objective function is fundamentally different from the testing objective (the accuracy of the resulting parse). We wanted to develop a method of training CRFs… (More)


Figures and Tables

Sorry, we couldn't extract any figures or tables for this paper.

Slides referencing similar topics