Prof. Emo Welzl and Prof. Bernd Gärtner
|Mittagsseminar Talk Information|
Date and Time: Thursday, August 22, 2013, 12:15 pm
Duration: 30 minutes
Location: OAT S15/S16/S17
Speaker: Christoph Schwirzer
This talk is about the problem of learning object classification using a small training set. We examine the problem of recognizing handwritten characters. Using a simple neural network we achieve a classification rate of more than 87% on handwritten decimal digits after learning as few as 20 samples per digit only. Based on a similar model, we examine the classification of simple geometric figures. Starting with a long and unsupervised learning stage the similarity of objects is learned. A following short and supervised learning stage is shown to be enough to perfectly distinguish certain basic geometric figures like squares and diamonds.
Automatic MiSe System Software Version 1.4803M | admin login