Real-valued Networks and Transformation to Logical Networks

Illustration of neural network transformation from real-valued to binary-valued representation.

Project image

We take trained recurrent neural networks of small/moderate size and analyze their transformation to binary-valued networks. In this student project a known algorithm is applied and variants for it may be explored. The goal is to compare the binary-valued network with the original and assess accuracy, efficiency, stability etc. of the transformation. We expect the efficiency to be highly increased, while yielding accurate or only moderately degraded results.

Requirements (Bachelor):

Basics in Computational Neuroscience and Machine Learning

Advisor:

Florian Röhrbein, florian.roehrbein@informatik.tu-chemnitz.de