Specialization abroad - DeepLearn 2017 – Deep Learning research training event

Posted by Marko Arsenovic on 02 August 2017

This summer I had a chance to participate in one of the biggest deep learning events held this year, DeepLearn 2017 in Bilbao, Spain.

If you don’t know what deep learning is, it is an element of Artificial Intelligence, and mostly used technique in machine learning today.

One of the things that I was definitely excite about was where this event took place. Bilbao is located in the northern Spain and it is the largest city in the Basque Country.  It is famous for its dual architecture. Bilbao's buildings display a variety of architectural styles, ranging from gothic, Art Deco, Art Nouveau, Neo-Gothic and contemporary architecture.

On the one side, there is the old town. The Casco Viejo of Bilbao is the heart of the city and it is known as “las 7 calles” in Spanish or “Zazpi Kaleak” in Basque, which means the seven streets. It is the most colorful neighborhood.  The beautiful narrow streets are mostly pedestrian-only and are packed to the brim with taverns offering tasty pintxos and delicious Rioja wine.

On the other side, there is an urban landscape dotted with elements of a country setting. In the center of the new, modern Bilbao is The Guggenheim museum. It is, without a doubt, the main attraction of Bilbao. Designed by star architect Frank Gehry, the museum opened to the public in 1997 and has since been hailed by both the public and critics as one of the most important structures of contemporary architecture. The building is wrapped in titanium panels that create what appears to be random organic curves. These panels were designed to capture light. For this reason, the building takes on a different appearance at night as the city lights reflect off it.

Bilbao is a culinary heaven. The region stands out for its cuisine, which is rich in local products and meticulously prepared. The city is full of tasty delights and boasts several Michelin-starred restaurants. The dishes prepared in Azurmendi, Mina, Boroa, Etxanobe, Etxebarri, Nerua, Zortziko, Andra Mari and Zarate, among other restaurant, are part of the best menu and represent the most refined Basque cuisine.

Bilbao, and the Basque region in general, are also appealing for film industry. The 1999. James Bond movie “The world is not enough”, starring Pierce Brosnan was filmed In Bilbao.

More recently, three sites in Spain’s northern Basque coast have been chosen as locations for filming scenes in the currently most popular TV series, Game of Thrones. The most impressive one is just half an hour from Bilbao. It is the ninth-century religious site of San Juan de Gaztelugatxe – a zigzag walk down a steep hill, passage across the Bay of Biscay via a picturesque stone footbridge, and a hike up 241 steps, to be exact. But the views from the top of the island make the effort worthwhile. It the “Game of Thrones” those are the stair which lead to Dragonstone. Daenerys Targaryen’s castle.

Given the fact that Bilbao is industrial center of Spain, there is no surprise they are following the trends in technology, as well. Due to that, this was a perfect place for high tech events such as DeepLearn 2017 .

In PanonIT, we are primarily focused on developing products in the highest quality reaching clients’ and market’s demands. Deep learning has been the most popular topic in Machine Learning community in the last years. Not only our R&D team follows current trends and breakthroughs in Deep Learning and AI in general, but tries to contribute by publishing scientific papers and practically implement the state-of-the-art methods and approaches in order to deliver the best products to our clients. Events like DeepLearn 2017 give us the chance to gather researchers, both from academia and industry, to talk about recent advances and challenges in this very popular AI branch.

DeepLearn 2017 was a research training 5-day event from 17th — 21th July 2017, that gathered nearly 1300 participant from more than 20 countries. Most deep learning subareas were displayed, and main challenges identified through 4 keynote lectures and 28 six-hour courses. There were a lot of interesting topics and challenges in computer vision, speech recognition, language processing, gaming, biomedical informatics, recommender systems and robotics introduced at these lectures.

Due to the fact that 4 courses ran in parallel during the whole event, sadly I could participate at only one of them at the time.  I mostly chose courses that are related to current researches and projects active in our R&D team back at PanonIT.

Entire event had a very tight schedule, having lectures during the whole day from 9 am to 9 pm, having only 1-hour launch break, and 15 minutes small breaks between lectures. Being there all the time was very hard, but nothing is difficult when you are in the area of top-level lecturers, which can really teach you a lot.

Regardless to the intensiveness of the event, topics and lectures were engaging with real-life hands-on examples, so the time passed quickly in learning. Lecturers were distinguished members of the deep learning community from industry and academia including Li Deng, current Chief AI Officer at Citadel and former Chief Scientist of AI at Microsoft, Ruslan Salakhutdinov, Director of AI Research at Apple and professor of Computer Science at Carnegie Mellon University, Jianfeng Gao Partner Research Manager at Microsoft AI & Research and many notable researchers from Facebook, NVIDIA, IBM, Univeristy of California, Georgia Tech and etc. At the open session event, I had a chance to present our recent release of the Cropsit, deep learning powered plant disease detection mobile system, along with our academic paper published at the same topic.

Most interesting courses were:
    • Applied Deep Learning for Visual Computing, Autonomous Vehicles, and Gaming
    • Deep Learning for Computer Vision
    • An Overview of Deep Learning for Speech, Image, Text, and Multi-modal Processing
    • Applications of Deep Learning Models in Human-Computer Interaction Research
    • Speech Recognition and Machine Translation: From Statistical Decision Theory to Machine Learning and Deep Neural Networks
    • Learning Representations for Vision, Speech and Text Processing Applications
    • Interpretable Deep Learning Models for Healthcare Applications

At the end of DeepLearn 2017, I had attended 10 courses consisted of 50 hours of lectures. From my point of view, these types of events in this rapid scientific area helps in fast sharing of cutting-edge knowledge and in building up the AI community in overall.

I was very eager to come back home from this inspiring event and to start making new things and tackle the new challenges along with my team at PanonIT in order to build beautiful and innovative new products.