Timezone: »

Topological Autoencoders
Michael Moor · Max Horn · Bastian Rieck · Karsten Borgwardt

Tue Jul 14 10:00 AM -- 10:45 AM & Tue Jul 14 10:00 PM -- 10:45 PM (PDT) @ None #None

We propose a novel approach for preserving topological structures of the input space in latent representations of autoencoders. Using persistent homology, a technique from topological data analysis, we calculate topological signatures of both the input and latent space to derive a topological loss term. Under weak theoretical assumptions, we construct this loss in a differentiable manner, such that the encoding learns to retain multi-scale connectivity information. We show that our approach is theoretically well-founded and that it exhibits favourable latent representations on a synthetic manifold as well as on real-world image data sets, while preserving low reconstruction errors.

Author Information

Michael Moor (ETH Zurich)
Max Horn (ETH Zurich)
Bastian Rieck (ETH Zurich)
Karsten Borgwardt (ETH Zurich)

Karsten Borgwardt is Professor of Data Mining at ETH Z├╝rich, at the Department of Biosystems located in Basel. His work has won several awards, including the NIPS 2009 Outstanding Paper Award, the Krupp Award for Young Professors 2013 and a Starting Grant 2014 from the ERC-backup scheme of the Swiss National Science Foundation. Since 2013, he is heading the Marie Curie Initial Training Network for "Machine Learning for Personalized Medicine" with 12 partner labs in 8 countries (http://www.mlpm.eu). The business magazine "Capital" listed him as one of the "Top 40 under 40" in Science in/from Germany in 2014, 2015 and 2016. For more information, visit: https://www.bsse.ethz.ch/mlcb

More from the Same Authors