CHI2016: Resolvable vs. Irresolvable Ambiguity
In May 2016, I presented a position paper from our research group at the workshop on Human-Centered Machine Learning (HCML) at the CHI2016 conference in San Jose. In this paper, we tackle the conventional assumption that, in supervised machine learning, ground truth data always needs to provide exactly one correct label per training example. We propose a new hybrid framework for dealing with uncertainty in ground truth that fully acknowledges the notion of irresolvable expert disagreement.
Hacking Brain-Computer Interfaces
In March 2015, I had the pleasure to give a talk on brain-computer interface (BCI) technologies at the “Singularity meets Self-Improvement” (SMSI) meetup in Berlin which is regularly hosted by AI researcher and start-up founder Dr. Trent McConaghy. In this talk, the focus is on different ways of presenting, processing and analyzing electroencephalographic (EEG) data, alleging some fun examples from real-world BCI applications. In addition to that, the open-source/open-hardware device OpenBCI is introduced and demonstrated.
Implicit Surface Modeling for 3D Printing
In July 2015, the award-winning computer graphics experts at xy:matic invited to the first-ever WebGL Meetup Berlin at the time for which I was a speaker on the topic of implicit surface modeling for 3D printing. Implicit surfaces are a traditional way of accurately representing the shape of complex 3D objects in space. This type of representation was neglected in the field of computer graphics for a long time due to insufficient computing capabilities in GPUs. Today, as GPUs have become increasingly powerful, implicit surfaces are flourishing again as representations of customizable objects for 3D printing.