For a third year in a row, three of the Data Science Campus’ data scientists attended this year’s DeepLearn International Summer School. The annual conference was held in July, in Warsaw, Poland. The international summer school aims to “update participants about the most recent advances in the critical and fast developing area of deep learning”. Comprising three keynote speakers, and 22 four-and-a-half hour courses, this year’s summer school was attended by over 1,200 people from around the globe.
Held at the newly-built, not quite finished Global Expo building, located in the old industrial area north of central Warsaw, the first hurdle was finding suitable accommodation. Fortunately, Warsaw has frequent trams linking the centre to the conference venue, and failing that you can hop onto one of the numerous e-scooters that are parked around the city.
The conference began on day one with an uplifting speech by the organisers on how 25% of this year’s attendance were female, and that two out of the three keynotes were also female. What wasn’t mentioned though was that 100% of course leaders were male, and a quick analysis of course material citations (combined with Google Scholar) showed that 97% of the first authors cited were male.
To kick off their learning, all three of our data scientists (and the majority of the attendees for that matter) opted to sit in on Tomas Mikolov’s Using Neural Networks for Modeling and Representing Natural Languages session. Mikolov is renowned in the field of deep learning and natural language processing due to his creation of the word2vec method of word embedding, as well as his involvement with Facebook’s fastText model. Mikolov provided a comprehensive journey through his and others work, which was a fairly gentle and interesting start to the week.
For the rest of the week, the data scientists split ways and attended various talks from Understanding the Brain with Machine Learning, to Deep Generative Models, to Explainable Artificial Intelligence. The days were long, at the earliest they began at 08:45, and at the latest, ended at 20:15. DeepLearn is a marathon, not a sprint.
The summer school’s website says it is suitable for master’s students, PhD students, postdocs, and industry practitioners. They state there is no formal pre-requisites for attendance, but specific knowledge background may be assumed for some of the courses. In reality, specific domain knowledge was required for all the courses, including for many, advanced mathematical and statistical knowledge.
DeepLearn 2019 provided a very interesting environment to learn the research at the forefront of deep learning and data science, but there was limited to no opportunities to implement the learned knowledge or time to understand how best it could be used back at the Data Science Campus.
Networking opportunities were limited to coffee breaks and lunch; there was some dinner with strangers events on some of the evenings, but these were scheduled after 20:15 and thus after almost 12-hours of learning. In hindsight the Campus should have made use of the open-sessions that proceeded the days course material, where we could have present our work in the field of deep learning. This was a missed trick on our behalf.
In summary, DeepLearn 2019 was a unique experience for our data scientists to learn from some of the world leaders in deep learning; however, there was limited opportunity to consolidate this learned knowledge.