A Machine Learning Approach to Online Test Error Detection and Large Margin Principle in Hyperrectangle Learning (Statusvortrag)

Date

Feb 28, 2012

Time

10:30 AM - 11:30 AM

Speaker

Dipl.-Inf. Matthias Kirmse

Language

en

Main Topic

Informatik

Other Topics

Informatik

Description

As a consequence of high investments and short life-times related to semiconductor factories, manufactures are forced to optimize almost every step in their production processes to gain maximal yield. A crucial part therefore is the fast and accurate detection and classification of process failures. While a variety of approaches have recently been developed to monitor manufacturing equipment, the semiconductor test process has yet gained little attention. Therefore, we developed a new machine learning based approach to monitor the test process and efficiently detect and recover test errors. Besides this new approach, the talk provides an overview on current methods and relevant related work. The second main topic of our thesis and therefore of this talk is the “Large Margin Principle in Hyperrectangle Learning”. This principle has recently been identified as a unifying key property for a variety of machine learning approaches like Boosting, Neural Networks and Support Vector Machines. In our thesis we study the incorporation of this large margin principle into hyperrectangle learning to combine their beneficial features: accuracy and interpretability. In our presentation, we will provide some theoretical background to the large margin principle and present a novel meta learning approach called “Large Margin Rectangle Learning”. Betreuer: Doz. Dr.habil. Uwe Petersohn Fachreferent: Prof. Dr.-Ing.habil. R.G. Spallek

Last modified: Feb 28, 2012, 8:36:59 AM