講座題目:Bayesian Nonparametric Methods for Text Modeling
主講人:John Paisley博士后
時間:2012年8月6日上午9點
地點:先進礦山裝備教育部工程中心研發(fā)樓學(xué)術(shù)報告廳(218室)
主辦單位:先進礦山裝備教育部工程研究中心
機械設(shè)備健康維護湖南省重點實驗室
Title: Bayesian Nonparametric Methods for Text Modeling
Abstract: Recent developments of Bayesian nonparametric methods within the machine learning community has had a major impact on text modeling applications. These methods rely heavily on the Dirichlet process, which breaks each document of text data into its underlying parts, called topics, without defining the number of parts in advance. In this talk, I will first review parametric Bayesian approaches to topic modeling, followed by its extension to the nonparametric setting. I will then present recent work on a Bayesian nonparametric approach to correlated topic modeling, which allows for modeling of correlations between topic probabilities in a collection of documents.
=======================================================================
講座題目:非參數(shù)層次貝葉斯圖像疏表示方法及應(yīng)用
主講人:丁興號教授
地點:先進礦山裝備教育部工程中心研發(fā)樓學(xué)術(shù)報告廳(218室)
主辦單位:先進礦山裝備教育部工程研究中心
機械設(shè)備健康維護湖南省重點實驗室
題目:非參數(shù)層次貝葉斯圖像疏表示方法及應(yīng)用
摘要:近年來,信號在冗余字典下的稀疏表示在圖像處理,、壓縮感知,、機器學(xué)習(xí)及計算機視覺等諸多領(lǐng)域得到廣泛應(yīng)用,。但基于優(yōu)化的冗余字典稀疏表示方法一般存在如下不足:1)需要預(yù)先設(shè)定信號的稀疏度;2)不同的信號采用相同的稀疏度設(shè)定方式,;3)需要預(yù)先設(shè)定噪聲方差或重構(gòu)殘差,;4)冗余字典中所含原子的數(shù)目需要預(yù)先設(shè)定等。針對上述不足在非參數(shù)層次貝葉斯框架下介紹一種新的圖像稀疏表示方法,,在此基礎(chǔ)上提出幾種改進方案并給出其在去噪,、背景與前景分離、壓縮感知等領(lǐng)域的成功應(yīng)用實例,。