Differences

This shows you the differences between two versions of the page.

--- nlp:label_bias_problem [2021/04/01 09:44] – created jmflanig
+++ nlp:label_bias_problem [2023/06/15 07:36] (current) – external edit 127.0.0.1
@@ Line 2: / Line 2: @@
 ===== Papers =====
-  * [[Andor et al 2016 - Globally Normalized Transition-Based Neural Networks]] Proves that global models can be strictly more expressive than local models.
+  * [[https://repository.upenn.edu/cgi/viewcontent.cgi?article=1162&context=cis_papers|Lafferty et al 2001 - Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data]] Introduced the label bias problem, which is present in Maximum-Entropy Markov Models (MEMMs) but not Conditional Random Fields (CRFs).  This drawback of MEMMs was one of the main reasons for inventing CRFs
+  * [[https://arxiv.org/pdf/1808.10006.pdf|Murray & Chiang 2018 - Correcting Length Bias in Neural Machine Translation]] Argues that the beam search problem in NMT occurs because of the label bias problem (see section 2)
+  * [[https://arxiv.org/pdf/1603.06042.pdf|Andor et al 2016 - Globally Normalized Transition-Based Neural Networks]] Proves that global models can be strictly more expressive than local models.
   * [[https://www.aclweb.org/anthology/N19-1171.pdf|Goyal et al 2019 - An Empirical Investigation of Global and Local Normalization for Recurrent Neural Sequence Models Using a Continuous Relaxation to Beam Search]]