Faculty of Computer Science

Research Group Theoretical Computer Science

Oberseminar: Heterogene formale Methoden

Abstract:

With the growing influence of deep learning approaches in modern applications, artificial neural networks have been brought back to the public attention. Whilst they have been designed to match the physical make-up natural neural networks, they struggle with basic abstraction tasks.
In this talk we will present the results of Weyde et al., that analyse the ability of artificial neural networks to learn abstract patterns, more specifically, their ability to recognise the pattern ABA in three letter words.