著者

久保 良介 堀井 真吾 辻 和樹 江藤 力

出版者

一般社団法人 日本航空宇宙学会

雑誌

日本航空宇宙学会誌 (ISSN:00214663)

巻号頁・発行日

vol.58, no.672, pp.34-35, 2010-01-05 (Released:2019-04-19)

参考文献数

2

著者

藤巻 遼平 山口 勇太郎 江藤 力

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会論文誌 (ISSN:13460714)

巻号頁・発行日

vol.31, no.6, pp.AI30-I_1-9, 2016-11-01 (Released:2016-11-02)

参考文献数

34

被引用文献数

2

---

Piecewise sparse linear regression models using factorized asymptotic Bayesian inference (a.k.a. FAB/HME) have recently been employed in practical applications in many industries as a core algorithm of the Heterogeneous Mixture Learning technology. Such applications include sales forecasting in retail stores, energy demand prediction of buildings for smart city, parts demand prediction to optimize inventory, and so on. This paper extends FAB/HME for classification and conducts the following two essential improvements. First, we derive a refined version of factorized information criterion which offers a better approximation of Bayesian marginal log-likelihood. Second, we introduce an analytic quadratic lower bounding technique in an EM-like iterative optimization process of FAB/HME, which drastically reduces computational cost. Experimental results show that advantages of our piecewise sparse linear classification over state-of-the-art piecewise linear models.