[Adapt] [Seminar]Introduction to Conditional Random Field

[Luo Kangqi]

Hi Adapters,


The seminar will take place at 4:30 pm tomorrow, in SEIEE 3-404.


This time, I'm going to introduce Conditional Random Field (CRF). I thought many of you guys heard this model before, its a well known discriminative model for many tasks, especially for the sequential labeling task. CRF is proposed in 2001, the technique itself is a bit out of dated. But recently I encountered the problem of structured prediction in our project, and happened to read some papers on image object classification where CRF is applied, and even combined with neural networks, listed as below:
  Dense CRF:  http://graphics.stanford.edu/projects/densecrf/densecrf.pdf
  CRF as RNN: https://arxiv.org/pdf/1502.03240.pdf


I intended to introduce these two papers, but found it's very hard to make sure that everyone of you learns something within only 30 minutes. So I changed my plan, and just try to focus on the following questions:
- What does the name "Conditional Random Field" come from ?
- How do we get the formula modeling p(y|x) in CRF ?
- Is there any connections between other models (Naive Bayesian, Logistic Regression, Hidden Markov Model) ?


This talk is very basic, enjoy!


Best,
Kangqi
------------------
Kangqi Luo, PhD Candidate
ADAPT Lab, SEIEE 3-341
Shanghai Jiao Tong University
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cs.sjtu.edu.cn/pipermail/adapt/attachments/20170426/58fd8338/attachment.html>