Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

A neural network implementation of Frank-Wolfe optimization

: Bauckhage, Christian


Lintas, Alessandra:
Artificial neural networks and machine learning - ICANN 2017. Pt.1 : 26th International Conference on Artificial Neural Networks, Alghero, Italy, September 11-14, 2017; Proceedings
Cham: Springer International Publishing, 2017 (Lecture Notes in Computer Science 10613)
ISBN: 978-3-319-68599-1 (Print)
ISBN: 978-3-319-68600-4 (Online)
International Conference on Artificial Neural Networks (ICANN) <26, 2017, Alghero>
Conference Paper
Fraunhofer IAIS ()

We revisit the Frank-Wolfe algorithm for constrained convex optimization and show that it can be implemented as a simple recurrent neural network with softmin activation functions. As an example for a practical application of this result, we discuss how to train such a network to act as an associative memory.