Machine Learning Summer School#
This post is long overdue, but jetlag after the summer school and a planned end-of-summer trip to visit friends got the best of me when I got home.
The Machine Learning Summer School in Tubigen was a great experience. I highly recommend anyone in machine learning to attend a summer school if possible(there’s at least one every year, 3 planned for 2014) and other graduate students to see if their field runs a similar program.
My favorites were the talks by Chris Bishop on graphical models and the Gaussian Processes series by Philipp Hennig. My machine learning course at school last year used Bishop’s book so I was generally familiar with the content and his notation, but the added insight and case studies clarified a lot. Also, sometimes just studying things in repetition makes them that much clearer. While presenting the technical details on how to use graphical models, he also presented a case for model based machine learning– a strategy for doing machine learning that I think is much more fluid and natural than a lot of what is done. What I liked best about the Gaussian processes talks was how he built up to the result. Again, the general material wasn’t really new to me, but the fresh insights and connections clarified things a lot. His slides may have been the best visual aids as well and I’m not alone in that assessment. Many attendees asked us about them and he released a [](http://mlss.tuebingen.mpg.de/Hennig_2013_Animating_Samples_from_Gaussian_Distributions.pdf”>tech report on how to generate the figures he used. Dr Hennig closed his talk with work on probabilistic numerics- taking the view that the numerical techniques used when an analytically solution is unavailable can be viewed as estimation and solved probabilistically. Again, the broader perspective and insight was what resonated most for me.