Learning to Learn Deeply

Another year, another summer school on machine learning. From 25 to 29 of September, we organized the 2nd Int’l Summer School on Data Science in Split, this time focusing on deep learning. Was it a blast? Keep reading!

Something we learned the last time we organized Summer School of Data Science (SSDS) in Split, is that it’s hard to beat a machine learning summer school organized at the coast side. That is why it comes as no surprise that we decided to follow in the same footsteps this year as well, this time from 25 to 29 of September. 🙂

Together with the Center of Research Excellence for Data Science and Advanced Cooperative Systems, we decided to teach the participants about deep learning, one of the most prominent machine learning paradigms today. To help us in this endeavor, we invited a number of great speakers coming from KU Leuven, University of Amsterdam and Oxford, and even Google DeepMind. From our side, Jan and Martin gave their best to make everything run smoothly as possible, while Abbas, Bojana, Maria, and Mladen simply enjoyed the lectures.

In the first lecture of the summer school, Matthew Blaschko (KU Leuven) introduced some fundamental machine learning concepts. The next day, Thomas Mensink (University of Amsterdam) covered convolutional networks and zero-shot learning, while the third day was entirely dedicated to language modeling and recurrent networks, presented by Tim Rocktäschel (University of Oxford). The penultimate lecture, given by Jakub M. Tomczak (University of Amsterdam) focused on generative adversarial networks and variational autoencoders. Last but not the least, Nal Kalchbrenner (Google DeepMind) concluded with the final lecture on autoregressive generative networks.

All lectures were accompanied by the practical lab sessions in TensorFlow, in which the participants went into the nitty-gritty details of the mentioned deep learning models. Among the technical details, they have also familiarized themselves with all the voodoo tricks often needed to make some of these models actually work. 🙂

Of course, no summer school is complete without some touristy things. We all enjoyed a guided tour of the city, sampling some of the social hotspots and meeting ever so interesting people of Split. In retrospect, we learned that there is still a lot of things we don’t know, so we return humbled and motivated to learn deep learning even more deeply.