Every year, thousands of developers, researchers, engineers and entrepreneurs make a pilgrimage to San Jose Convention Center – the focal point of computing revolution powered by Nvidia’s most powerful GPUs. This year, Tooploox joined this crowd and this blogpost gives you a brief overview of our visit at GTC 2017.
If you wanna learn what GTC is about in less than three minutes – here’s the easiest way:
It’s an opening scene of this year’s keynote presentation by Jensen Huang, CEO of Nvidia. Although it covers a lot of topics, discussed during the GTC conference, I believe that the common thread that connects all of them is obvious: Artificial Intelligence (AI).
So if you are keen on learning how our AI-powered tomorrow will look like, fasten your seatbelts, sit down, relax and let us take you for a quick journey around GTC 2017.
We flew to California just before the conference started. The land of hi-tech startups and delicious avocado, greeted us with a beautiful cloudless sun and an amazing view over the Bay area:
The next day, we woke up pretty early (thank you, jetlag!) and headed out to the conference venue, as quickly as we could. We thought we were gonna be the first ones there, but…
… we were wrong. The line to the registration desk was pretty long, but it moved fairly quickly, so we were all set in no time.
The first thing that caught our attention after getting in were a bunch of cars, that looked almost like a regular ones, except for the fact that… they were not. Equipped with a bunch of sensors, LIDAR and cameras, Udacity’s Lincoln and Audi Q7 were greeting the visitors in the entrance lobby.
This was just the beginning of a set of events and talks around the topic of autonomous driving that took place at GTC. The organizers dedicated an entire auditorium and a huge space in the exhibition hall for everything that touches the topic of autonomous driving. There was even an invitation-only autonomous driving reception party (which we attended, dah!).
For full overview, you can visit Nvidia’s blogpost.
At GTC, Nvidia proves that deep learning and GPUs are inseparable. By inviting the most respected researchers in the field – Ian Goodfellow, Russ Salakhutdinov or Raquel Urtasun – to give a talk, Nvidia sends a clear signal: deep learning is the main focus of their products.
Among numerous talks on applying deep learning to solve industrial, fintech or agricultural problems, let’s highlight a couple that we found the most interesting.
Recent advancements in deep learning
This was a very interesting talk given by Russ Salakhutdinov from the famous deep learning group at University of Toronto. He started by introducing a broad class of deep learning models and showed that they can learn useful hierarchical representations from large volumes of high-dimensional data with applications in information retrieval, object recognition, and speech perception. He next talked about deep models that are capable of extracting a unified representation that fuses together multiple data modalities. In particular, he talked about models that can generate natural language descriptions (captions) of images, as well as generate images from captions using attention mechanism. Finally, he discussed an approach for unsupervised learning of a generic, distributed sentence encoder and introduced multiplicative and fine-grained gating mechanisms with application to question/answering systems and reading comprehension. After leaving the talk, we were pretty overwhelmed with the amount of effort Russ and his group puts into unsupervised learning and AI-reasoning powered by unlabeled data.
Sentiment analysis with unsupervised learning
Another talk on unsupervised learning we attended was given by Stephen McGough from Durham University. According to Stephen, est. 85% of worldwide data is held in unstructured/unlabelled formats – increasing at a rate of roughly 7 million digital pages per day. Exploiting these large datasets can open the door for providing policy makers, corporations, and end-users with unprecedented knowledge for better planning, decision making, and new services. Deep learning and probabilistic topic modeling have shown great potential for analysing such datasets. This analysis helps in: discovering anomalies within these datasets, unravelling underlying patterns/trends, or finding similar texts within a dataset. The talk illustrated how we can use a combined unsupervised deep learning and topic modeling approach for sentiment analysis requiring minimal feature engineering or prior assumptions, and outperforming the state of the art approaches to sentiment analysis.
Deep incremental scene understanding
The last talk we would like to highlight in this post was given by Federico Tombari from Bologne University, visiting Technical University of Munich. During the talk he demonstrated recent advances in the field of deep learning and computer vision aimed at scene understanding from images. He presented two research works on this subject. The first one relates to the use of deep learning for monocular simultaneous localization and mapping (SLAM) and semantic segmentation. The outcome is a technique able to carry out accurate real-time semantic mapping and 3D reconstruction from a single RGB camera. The second research work uses deep learning for solving ambiguous prediction problems. Federico demonstrated how the two approaches can be merged together to enable robust extraction of 3D semantic information such as pixel-wise labeling and object detection in real time by means of a simple webcam.
On top of the talks on deep learning, Nvidia organized a few other attractions for machine learning geeks. First of them was an opportunity to get an autograph of Ian Goodfellow on the Deep Learning book he co-authored. The line was preeeety long, so we didn’t actually wait, but it was interesting to see how famous deep learning researchers can get nowadays…
Another interesting event (or rather a series of events) was a set of meetings entitled “Connect with the experts” where beginner and intermediate-level programmers could find out more about deep learning and ask all the various questions came to their mind. It was a really exciting opportunity, judging by the size of the crowd around those stands, and we enjoyed listening to the experts too!
Medicine of the future
I am a healer – said the voice in the opening video at the Keynote presentation. And it was not only about the keynote. GTC was full of vivid examples on how GPU-powered deep learning can help doctors and patients all over the world. One notable example is Triage – a mobile app that helps to detect early stages of skin cancer using convolutional neural networks.
Several talks were also focused on applying deep learning to x-ray images and detecting cancerogenous tissues in them. Although the topic itself is quite interesting and was researched for a long time, only very recently have there been enough data available publicly that deep learning methods could be trained for x-ray analysis applications. Organization such as Lung Image Database Consortium or Image Database Resource Initiative made it possible for big datasets, such as LIDC-IDRI dataset, to be collected. Another dataset, Medical ImageNet, was also presented at the GTC conference. All the efforts focused on collecting big datasets of annotated medical data have made the development of AI-powered autodiagnostic tools possible and this is why Nvidia’s I am a healer claim is absolutely true.
Veni, vidi, vici. GTC was an amazing experience for us, we will definitely be back next year, even if just to 3D-scan ourselves again or take a look at our GAN-generated portraits!