潛在狄利克雷分佈（Latent Dirichlet Allocation，LDA）在文件採礦（Text mining）中是被廣泛採用的模型，它擁有很好的判斷及訓練模式，然而參數估計卻相對複雜，於是採用了變分貝氏（Variational Bayes）估計的方法，此方法可使其得到一個不錯的估計結果。在變分貝氏估計中，需先找到一個概似函數的證據下界（Evidence Lower Bound，ELBO），本文中，我們利用Chernoff 信息量推導得一個信新的證據下界，並證明此下界較原先的為大。

Latent Dirichlet Allocation is a model which is widely used in text mining. It also has good judgment and training mode. Howere, the estimation of its parameter is relatively
complex. So we use the method, Variational Bayes, to get a good result of estimation. In Variational Bayes, we need to find an Evidence Lower Bound of likelihood function. In this paper we derive a new Evidence Lower Bound from Chernoff information and prove that it is bigger than the original one.

[1] Wikipedia, Data mining, https : //en:wikipedia:org/wiki/Data_ mining.
[2] Darroch, J.N. and Ratcliff, D. (1972) Generalized iterative scaling for log-linear models.The Annals of Mathematical Statistics, 43:1470–1480.
[3] Adam Berger. (1997) The Improved Iterative Scaling Algorithm: A Gentle Introduction,http://www.cs.cmu.edu/ aberger/pdf/scaling.pdf.
[4] Kilian Q. Weinberger, Matt J. Kusner, Nicholas I. Kolkin, Yu Sun. (2015) From Word Embeddings To Document Distances. The International Conference on Machine Learning(ICML), Washington University in St. Louis , 1 Brookings Dr., St . Louis, MO 63130.
[5] Andrew Y. Ng,David M. Blei ,Michael I. Jordan and John Lafferty. (2003) Latent Dirichlet Allocation. Journal of Machine Learning Research, 3:993–1022.
[6] David M. Blei, Francis Bach, Matthew D. Hoffman.(2010) Online Learning for Latent Dirichlet Allocation. Advances in Neural Information Processing Systems(NIPS),856-864.
[7] David M. Blei. (2011) Variational Inference. Lecture from Princeton, variational inference, https://www.cs.princeton.edu/courses/archive/fall11/cos597C/lectures/variationalinference-i.pdf
[8] Chernoff, Herman. (1952) A Measure of Asymptotic Efficiency for Tests of a Hypothesis Based on the sum of Observations. Annals of Mathematical Statistics, 25, 573-578.