|
Activity Number:
|
20
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, July 29, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #308910 |
|
Title:
|
Bayesian Classification of Sleep States in Mice (Student Paper Competition)
|
|
Author(s):
|
Chaitra Nagaraja*+ and Shane T. Jensen and Abraham Wyner
|
|
Companies:
|
University of Pennsylvania and University of Pennsylvania and University of Pennsylvania
|
|
Address:
|
2400 Chestnut St, Wharton School, Philadelphia, PA, 19103,
|
|
Keywords:
|
Bayesian ; classification ; hidden Markov model ; Viterbi
|
|
Abstract:
|
We examine the problem of inferring sleep states (awake, non-REM sleep, and REM sleep) in mice based on video data. Observed measures of velocity, size, and aspect ratio of a mouse were taken from video images over extended time periods. Data from one mouse was used to fit a hidden Markov model with the sleep states for each observation as latent variables. Mixture modeling in the emission densities was needed to account for multi-modality in the observed variables. We compare the accuracy of predictions based on the optimal path of sleep states generated from a Viterbi algorithm with predictions generated directly with the posterior state probabilities for each observation. The proposed Bayesian procedure provides more accurate classifications.
|
