Dance Dance Machine Learning Revolution
Years ago I had the idea of dancing elementary concepts of technology, programming and computation. I never started, because I always thought I need a professional choreograph, clearance for the songs I wanted to use and finally dancers to trust me that this concept will work.
Three weeks ago a colleague from a befriended institute called me to brainstorm on ideas for the 2nd Digital Science Night in Hamburg on January 23, 2020. As I had played the game Mensch, MASCHINE! with students from that Institute of Communication Networks some days before machine learning was still on my mind. I knew that there was a professional dancer among the students so I suggested to work on a dance performance on machine learning. To my surprise, the student agreed to give it a try!
We somehow convinced other students and colleagues to join and decided together on dancing supervised learning which meant classifying spam and normal mails. Our performance had a training phase in which the labeled mails showed up on stage with an individual move that characterized them as spam. The dancer in the center of the stage, a neural network, adapted their dancing style and this way refined the model. In the following validation phase unlabeled mails danced around him and got classified. The whole gig a narrator commented on the scence and explained what was going on. Me was playing a huge djembe which on the one hand solved any license problems and on the other hand made the movements on stage very lively and spontaneous.
We were told by the jury that we had broken the central rule of science slams - one person, one topic - but nevertheless won the first price. I’d like to thank everybody who joined this crazy plan and made it possible.
I had not expected that discussing what and how we’re going to dance would trigger so much technical discussion on the topic. We were over and over again explaining machine learning concepts to each other, refining the presentation and finding moves and actions that fit the correct computational rules for supervised machine learning.
Besides the fun I have to say that there is a huge potential to learn complicated and complex concepts by dancing them, if there is such a discussion. I’d like to do it again, perhaps with a series of standard algorithms. People can easily be sorted by dancing after certain constraints. And there are a lot of protocols on the web that could be danced. The term handshake e.g. is inspiring enough to let two humans interact on stage following a protocol. Let’s see…
