돌아오는 수요일(11/30)에 다음과 같이 학과 세미나를 합니다.

 

대학원생들은 꼭 참석 바랍니다.

 

 

 

 1. 장소 및 일정

 

    - 자연과학관 111호

 

    - 11월 30일 수요일

 

    - 오후 4시30분~5시10분 : 최태련 교수님 (고려대학교 통계학과)

 

    - 오후 5시10분~5시20분 : coffee break

 

    - 오후 5시20분~6시        : 최호식 교수님 (호서대학교 정보통계학과)

 

 

 

2. 발표제목 및 초록

 

     1) 최태련 교수님 (고려대학교 통계학과)

        

         - 발표제목 : Bayesian Asymptotic Analysis of Regression Problems

        

         - 초록

 

               In recent years, the literature in the area of Bayesian asymptotics has been   

            rapidly growing, and it is increasingly important to understand the concept of

            posterior consistency and validate specific Bayesian methods in terms of

            consistency of posterior distributions. In this talk, we build up some conceptual

            issues in consistency of posterior distributions, and discuss consistency of

            Bayesian procedure for regression problems. Specifically, we investigate the

            asymptotic behaviors of posterior distribution and the Bayes factor for regression

            problems in which observations are not required to be independent and

            identically distributed.

 

 

     2) 최호식 교수님 (호서대학교 정보통계학과)

 

         - 발표제목 : The Binary support vector machines and some extensions

 

         - 초록

                 

               The SVM(support vector machines) has been established as a representative

            methodology for many learning problems including classification, regression,

            clustering and ranking etc. Especially, in this talk, some extended problems of

            binary classification are considered. As such cases, I will present that a

            classifier with a reject option and its extension to nonstandard situations

            (misclassification costs are different depending on the class label and that

            sampling biases exist) by utilizing a Fisher consistent surrogate loss function.

            Here, a reject option plays a role in reporting a warning in case of observations

            that are difficult to classify.

            will also show some recent works related to SVM approach with labeling errors

            on outputs and partial AUC(area under the ROC curve) maximizer for the ranking

            problem. All developed implementations in practice can be achieve efficiently by

            using an entire solution path following algorithm which is sufficiently fast and

            stable to analyze high throughput genomic data.

 

 

 

 

 

 

!StartFragment>