Paper: Accelerated Training of Maximum Margin Markov Models for Sequence Labeling: A Case Study of NP Chunking

ACL ID C10-2161
Title Accelerated Training of Maximum Margin Markov Models for Sequence Labeling: A Case Study of NP Chunking
Venue International Conference on Computational Linguistics
Session Poster Session
Year 2010
Authors

We present the first known empirical results on sequence labeling based on maximum mar- gin Markov networks (M3N), which incorpo- rate both kernel methods to efficiently deal with high-dimensional feature spaces, and probabilistic graphical models to capture correlations in struc- tured data. We provide an efficient algorithm, the stochastic gradient descent (SGD), to speedup the training procedure of M3N. Using official dataset for noun phrase (NP) chunking as a case study, the resulting optimizer converges to the same qual- ity of solution over an order of magnitude faster thanthestructuredsequentialminimaloptimization (structured SMO). Our model compares favorably with current state-of-the-art sequence labeling ap- proaches. More importantly, our model can be eas- ily applied to other s...