Earlier this year, me and three colleagues of my convinced our PhD advisor to go to NIPS, the Neural Information Processing Systems conference in Montréal, Canada. As we all do research in the field of artificial intelligence - and in particular deep learning and neural networks - NIPS is really the high mass. It would be the first time for us, so we had no idea what to expect. It turned out to be very interesting and also a bit surreal at times.
As I had to make a report for my photography class, I took my camera with me all the time. I was too distracted by the presentations, posters, receptions… so I did not take that many pictures. I had the following gear in my backpack:
- Canon EOS 1100D camera body (very basic model, crop sensor, only usable up to ISO 1600),
- Canon EF-S 18-55mm \(f\)/3.5-5.6 (no IS),
- Canon EF-S 10-18mm \(f\)/4.5-5.6 (with IS),
- Canon EF 50mm \(f\)/1.4 (no IS),
- Canon EF 70-300mm \(f\)/4-5.6 (no IS).
As with my trip to New York City, I did not bring flashes or a tripod, since I did not want to disturb anyone or devote too much time setting up the camera. The pictures should be more like snapshots instead of well-considered ones. The images below are only slightly edited in Adobe Lightroom; click on them to enlarge.
We arrived in Montréal two days before the start of the conference, so we had some time to explore the city. I immediately noticed that the French language is alive and kicking over there, as is the love for France, shown by this underground sign which you would normally encounter in the centre of Paris.
The following picture also shows that the main language on almost all signs in the city and its restaurants is still French. ‘Le très hot chicken’ served with poutine: “a Canadian dish, originating in the province of Quebec, made with french fries and cheese curds topped with a light brown gravy-like sauce” [Wikipedia].
Next we went to the McGill University site and climbed up to the top of the Mount Royal (‘Mont Réal’).
This time of year the temperatures are normally much below zero degrees celcius, but we were lucky, as we had 5 to 10 degrees throughout the entire week. Still, it is obvious that winter usually poses severe conditions on the local vegetation.
The conference itself kicked off with a packed auditorium for the deep learning tutorial by Yoshua Bengio and Yann LeCun. It was very, very dark inside, so I had to zoom in to about 250mm to get the two in the picture. Still my shutter speed was too low to get a sharp image… The tutorial was not that interesting: if you already knew things about certain topics, you did not learn anything new; if you were unfamiliar with other topics, you were lost in the explanation.
The conference was held in the Palais de Congrès on the edge of Old Montréal, a huge building on top of the highway tunnel with a colourful facade.
The first conference day concluded with a poster session with more than 100 posters and a walking dinner. It was so crowded people got nearly crushed by each other. Later we heard that this NIPS edition broke records: more than 3,800 attendees this year!
The following days there were mostly talks during the day, which were often pretty boring. Bad speakers, thoughtless slide designs, lacking enthusiasm… I admit, the auditorium was extremely large, which makes it almost impossible to interact with the audience. The screens were also very far from the speaker, so that he or she could not interact with the slides in an effective way.
On the evening of the second day there were again posters and demonstrations. Two of my colleagues gave a very succesful demo of our platform DIANNE (for which I designed the logo). It is a JAVA- and OSGi-based internet-of-things library to distribute neural networks at runtime across devices.
On the two last days there were workshops. Some were overly crowded, such as the RAM workshop and the Deep Reinforcement Learning workshop. The father of reinforcement learning, Richard Sutton, was even denied access to the room1!
NIPS was quite an experience, a lot of people - including big names - a lot of posters, a lot of workshops, a lot of pushing and shoving, a lot of networking, a lot of receptions… Maybe next year again!comments powered by Disqus