log in  |  register  |  feedback?  |  help
Logo
Classification and clustering based on LDA-like models
Yoichi Tomiura - Kyushu University
Wednesday, November 1, 2017, 11:00 am-12:00 pm Calendar
  • You are subscribed to this talk through .
  • You are watching this talk through .
  • You are subscribed to this talk. (unsubscribe, watch)
  • You are watching this talk. (unwatch, subscribe)
  • You are not subscribed to this talk. (watch, subscribe)
Abstract

I introduce two studies used statistical model with latent variables similar to Latent Dirichllet Allocation (LDA) model.

The first is a study on detecting human values such as 'justice' or 'freedom' that a speaker/writer has sought to reflect in a sentence. We treated human values reflected in words or word sequences as latent variables, designed the statistical model that generates human values reflected in a sentence, and trained the model using a collection of sentences with human values annotated by a social scientist. This is joint work with Yasuhiro Takayama, Emi Ishita, Douglas W. Oard, An-Shou Cheng, and Ken Fleischmann. 
 
The second is a study on subdividing the glomerular layer of the olfactory bulb into a small region where the neurons show the same response to the odorants. We treated types of pixels (types of olfactory receptors connected to glomerulous in the pixel) as latent variables, designed the statistical model that generates the activation pattern images of the glomerular layer, and trained the model using the activation pattern images of the rats' glomerular layer put on the web cite (http://gara.bio.uci.edu/index.jsp). This is joint work with Kenshi Hayashi.
 
Through the two studies, I would show that we can relatively easily design the statistical model for classify or clustering by reflecting the properties of the target problem well in the prior distribution.
 
Bio

Yoichi Tomiura is a Professor of the Department of Informatics and the Department of Library Science at Kyushu University. He is also a Deputy Director General of University Libraries at Kyushu University. His interests lie at Natural Language Processing and Information Extraction using statistical models.

This talk is organized by Marine Carpuat