Considered the largest and most important event of the year for developers that build technologies with GPUs and the entire ecosystem, GPU Technology Conference (GTC) Asia South 2015 was held in Singapore just yesterday.
This year, the GTC Asia South highlights the growing revolution in deep learning (quite the buzzword at the moment!), and how technology enables the growth of Smart Cities around the world.
We attended the conference with high hopes that it would give us important takeaways, and it didn’t disappoint. Here are some nuggets of wisdom that were imparted during the event:
Currently, GPUs are being used across a whole breadth of sectors, not just for gaming. Use cases such as PC visualization (for 3D and even 4D renderings), deep learning, medical advances, cloud gaming and more all highlight the need for GPUs to support the system.
GPUs are also powering deep learning for cloud services such as Google, Baidu, Facebook and Twitter. In fact, Google alone is using GPUs in 47 applications, where they are at the pace to emulate the human brain in the next 2 years.
The usage of GPUs is going to completely change search infrastructure. As told by Andrew Ng, Chief Scientist at Baidu Research in Silicon Valley, “In 5 years, 50% of our search queries will be on speech or images.”
The trends driving high-performance computing and big data advances for these past few years include:
a) The invention of new algorithms
b) The availability of large datasets (which only start growing in the last 5-10 years)
c) The availability of powerful accelerators and computer power
Thus causing a massive change in the technology landscape.
Recent investments in deep learning has reached half a billion USD (that is congregated mainly in the US, India, China and Australia)
To encourage a focus on deep learning software, ecosystem development, and training in Singapore, NVIDIA is joining forces with Singapore Economic Development Board to establish the NVIDIA Technology Centre Asia Pacific, which will be the first of its kind in the region to focus on deep learning research and development.
WeChat, a mobile text and voice messaging communication platform developed by Tencent in China with 1.1 billion registered accounts, uses GPUs to build a great ecosystem that connects people, services and things, via collective participation and big data.
It applies machine intelligence to perform image understanding, user modeling for ads, cross domain transfer learning, audio fingerprinting, mobile visual search, facial recognition and many more.
The area of deep learning often builds on the strength of the community to achieve breakthroughs. For instance, the Kaggle contest on the training of deep neural networks to identify signs of diabetic retinopathy in images has yielded amazing results. The accuracy score of the winning team reached 0.84958, which is higher than the 0.83 accuracy score performed by humans diagnosing the condition!
The above points are largely from the conference’s deep learning track sessions, that focused on state-of-the-art research powered by GPU technology for image classification, video analytics, speech recognition, and natural language processing, and their applications.
It is obvious that we have learnt a lot from attending the conference. The trends that recent developments in GPUs are driving are very exciting, and we at ViSenze are proud to be a part of this deep learning movement that is going to move and shake the world.
(PS: For now, it seems that deep learning is still a baby that is learning how to walk - er I mean, play ping pong. And we're really excited to see it reach maturity.)