Attraction and Repulsion for Visualizing High-Dimensional Data (Manifold Learning)

PLEASE SUBSCRIBE IF YOU LIKE THIS VIDEO

This talk was delivered to the Quantitative Methods Network (QMNET) at the University of Melbourne on September 25th, 2020 by Jan Niklas Böhm, University of Tübingen.

TITLE:
Attraction and Repulsion for Visualizing High-Dimensional Data

ABSTRACT:
Due to the increased importance of visualization, a number of new techniques are actively being developed and adopted for applied research. Despite the popularity of the algorithms like t-SNE and UMAP, a formal analysis regarding the optimization procedure has not been carried out. Interpreting the optimization routine as a balance between attractive and repulsive forces gives rise to a spectrum that unifies various visualization techniques.

BIO:
Niklas studied Computer Science in Wiesbaden, writing the BSc-thesis at the German Research Center for Artificial Intelligence (DFKI) in Kaiserslautern. The Master's degree with a focus on machine learning was recently completed in Tübingen. During the studies he spent time in Amsterdam at the Universiteit van Amsterdam (UvA) and completed an internship at the European Council for Nuclear Research (CERN). The work presented was done as a part of the Berenslab. The lab is part of the Institute for Ophthalmic Research, focusing on Data Science for Vision Research as a part of Computational Neuroscience.