ぎょーむ日誌 2003-09-14

2003 年 09 月 14 日 (日)

• 0840 起床． 朝飯． コーヒー． 1005 自宅発． 台風一過で晴． 1020 研究室着．
• ぱいぷ樹木 電気盆栽 …… こればっか， と思われるかもしれないけどこればっかしかやっていないのである．

• 2330 研究室発． 2340 帰宅． 体重 70.6kg．
• [今日の素読]
  • Salsburg, D. 2001. `` The Lady Tasting Tea -- How statistics revolutionized science in the twentieth century''. Owl Book.
  • Chapter 13. The Bayesian Heresy
    • The Baysian hierarchal model
  • By repeated use of Bayes's theorem, we can determine the distribution of the parameters, the of the hyperparameters. In princeple, we could extend this hierarchy further by finding the distribution of the hyper-hyperparameters, and so on. In this case, there is no obvious candidate for the generation of an additional level of uncertainty. Using the estimates of hyper- and hyper-hyperparameters, Mosteller and Wallece were able to measure the probability associated with the statement: Madson (or Hamilton) wrote this paper.
  • Hierachal Bayesian models have been applied very successfully since the early 1980s to many difficult problems in engineering and biology. One such problem arises when the data seem to come from two or more distributions. The analyst proposes the existence of an unobserved variable the defines which distribution a given observation comes from. This identifying marker is parameters, but it has a probability distribution (with hyperparameters) that can be incorporated into the likelihood function. Laird and Ware's EM algorithm is particularly suited to this type of problem.
• [今日の運動]
  • 北大構内走 1755-1845． ストレッチング．
• [今日の食卓]
  • 朝 (0930): 米麦 0.7 合． タマネギ・卵の炒飯． トマト．
  • 昼 (1430): 研究室お茶部屋． 「北欧」バゲット．
  • 晩 (2430): 米麦 0.7 合． ジャガイモ・タマネギ・エノキダケ・豆腐のカレー．