What happens when AI experts from Silicon Valley and China meet
Why we're taking the AI Conference to Beijing.
Why we're taking the AI Conference to Beijing.
Having traveled to China several times over the last few years, I can attest to the strong interest in applications of AI among technologists, business leaders, and policy makers. China is adopting AI tools and technologies at a rapid pace, and since current AI systems rely on large data sets, startups are able to start using AI tools much earlier (a startup in China can quickly have many millions of users). People in the West are curious about the progress of AI research and business models in China. On the flip side, having organized a couple of large conferences in Beijing, I also know that people in China want to hear from AI experts outside their country.
For our upcoming Artificial Intelligence conference in Beijing, we are showcasing AI experts, technologies, and applications from inside and outside China. We will have sessions in English and Chinese (some translation will be provided), networking events, and other activities designed to foster conversations that will hopefully lead to interesting collaborations in the future.
In this short post, I wanted to highlight some of the areas that we’ll be focusing on at our inaugural AI conference in China. We’ve assembled a strong program comprised of some of the leading companies in China and the West.
AI solutions and technologies are beginning to be deployed across many industries. Opportunities for automation cut across many tasks. How does one find the right use cases for AI? An initial list of candidate tasks can be gathered by applying the following series of simple questions:
The best way to understand the potential of AI technologies is to see how machine intelligence is being used in real-world applications. We’ve assembled a series of sessions to highlight AI applications in several industries:
Renewed interest in AI can be traced back to the success of deep learning architectures in image classification and speech recognition. Since its emergence in 2012, computer vision researchers continue to push deep learning architectures, most recently toward content generation via semi-supervised learning. Computer vision is an active research area where ideas are being turned into real-world products (particularly in China). We will showcase:
Around 2011, a group of researchers from Microsoft and the University of Toronto created a speech recognition system based on deep learning that vastly outperformed existing systems. That marked the start of a shift toward systems based on neural networks. These days, speech researchers and speech products have coalesced around deep learning. Researchers in natural language processing and understanding are also beginning to use deep learning, but other machine learning methods remain important in many commercial products. Because their applications are so widespread, natural language and speech technologies are a strong focus in our conference in Beijing:
One of the areas we’re really interested in is the emerging applications of reinforcement learning (RL). We’ve all read about the key role RL played in systems that learned how to exceed human players in computer games and classic board games. But can RL be used in practical, real-world applications? As always, it’s good to start out with disclaimers: RL requires a lot of data and simulations, and research results tend to be difficult to reproduce.
However, two things seem point toward the direction of RL applications. First, tools for writing RL models and plugging them into simulators are starting to emerge, and many of them target developers who aren’t experts in machine learning. Second, companies are very interested in automation, particularly automating low-skilled tasks that occupy high-skilled workers. In this context, automation is sometimes referred to as robotics process automation or enterprise workflow automation. Many tasks that involve sequential decision-making are amenable to learning/training, making them ideal candidates for RL-based automation solutions. The democratization of tools coupled with the interest in automation (i.e., the use of learning rather than programming and rules), points toward interesting applications of RL in the near future.
We will feature keynotes, talks, and tutorials that will introduce the latest RL tools as well as applications to industrial automation and manufacturing, autonomous vehicles, and software development.
Given that deep learning has been behind the surge in interest in AI technologies, it’s no surprise that we have many sessions on this important machine learning technique.
How do leading companies architect and develop AI products? In a series of sessions, companies will share their internal platforms for machine learning and AI. These are battle-tested platforms used in production, some at extremely large scale. Here are a few such sessions from the conference:
One of the reasons we’re excited to hold a conference in China is that it’s one of the countries where innovation throughout the entire AI hardware and software stack is happening. Deep learning requires big data, big models, and big compute. Thus, the right combination of hardware and software infrastructure are essential. If you take a step back, data collection usually involves a host of sensors, many of which are equipped with compute resources. While stories in the popular press focus on machine learning, an end-to-end AI application actually involves many important hardware and software components that need to work seamlessly. With recent global investments in hardware startups, new hardware optimized for AI workloads will be coming in the near future. We will have sessions describing best practices and recent developments in the following key areas:
Hardware: Specialized hardware for AI systems is an area of focus in our Beijing conference:
Mobile and edge computing