Feb 4, 2020
手机体育投注平台Qualcomm products mentioned within this post are offered by Qualcomm Technologies, Inc. and/or its subsidiaries.
5G and AI are two of the most disruptive technologies the world has seen in decades. While each is individually revolutionizing industries and enabling new experiences, the combination of 5G and AI is going to be truly transformative. In fact, this intersection is fundamental to our vision of the intelligent wireless edge, in which on-device processing, the edge cloud, and 5G go hand-in-hand to create a ubiquitous connectivity fabric of smart devices and services.
Significant amounts data will be processed closer to its source, whether that be through on-device AI processing or through additional processing by the edge cloud over low-latency 5G. Processing data closer to the source through on-device AI is important since it offers crucial benefits such as privacy, personalization, and reliability, in addition to helping scale intelligence.
The intelligent wireless edge will not only enhance existing use cases but also enable new use cases and verticals. In this blog post, we’ll share our thoughts on how AI can make 5G better, how 5G can make AI-powered experiences better, and how distributed learning can happen over wireless.
AI is making 5G better — in the network and on the device
Applying AI to both the 5G network and the device will lead to more efficient wireless communications, longer battery life, and enhanced user experiences. AI is a powerful tool, and the key to harnessing AI to improve wireless is to focus on important wireless challenges that are both difficult to solve with traditional methods and are also a good fit for machine learning. Deep wireless domain knowledge is required to know where to use AI’s capabilities. That domain knowledge fits right in Qualcomm Technologies’ strengths thanks to our longstanding research in both wireless and AI.
Much of the talk in the wireless industry has been around how AI will make the 5G network better. And it is very clear that AI will have a strong impact on several key areas of 5G network management -- such as enhanced service quality, simplified deployment, higher network efficiency, and improved network security. For example, AI could be used to detect anomalies in network traffic, such as flooding or impersonation, by detecting unusual spectrum usage.
Discussed less often is how on-device AI is going to improve the 5G end-to-end system. Radio awareness is at the heart of how AI will improve 5G since machine learning, rather than a hand-crafted algorithm, is the perfect tool to make sense out of the complex RF signals around the device. Increased radio awareness enables a variety of improvements, such as enhanced device experience, improved system performance, and better radio security.
5G is making AI-powered experiences better
The low latency and high capacity of 5G will also allow AI processing to be distributed among the device, edge cloud, and central cloud – enabling flexible system solutions for a variety of new and enhanced experiences. This wireless edge architecture is adaptable and allows appropriate tradeoffs to be made per use case. For example, performance and economic tradeoffs may help determine how to distribute workloads to reach the required latency or compute requirements for a particular application.
We see 5G making AI-enhanced experiences better in scenarios such as personalized retail through boundless XR, intuitive virtual assistants through vastly improved voice UI, and the reconfigurable factory of the future through adaptive optimization.
Let’s imagine how shopping and retail might look like in the future. With boundless XR, rendering and AI processing workloads can be split between the device and edge cloud over a low-latency 5G link. When window shopping (Figure 4), we envision a much more personalized shopping experience where everything you see is of interest to you – whether that means seeing clothes that match your taste, your interest in a particular sale, or a gift for an upcoming birthday – making your shopping experiences much more engaging, productive, and efficient.
A new computing paradigm: distributed learning over wireless
In order to scale and make sense of the digitized world, we need to move beyond the idea of cloud-centric AI. Today, we see partially distributed AI thanks to the proliferation of power-efficient on-device AI inference, which allows devices to refine the data before it is passed on to the cloud for aggregated analysis. The next step for on-device AI is to go beyond inference itself and do training on the device as well. In the future, we see fully distributed AI with lifelong on-device learning that allows for the next level of personalization with privacy. How do we get there?
Distributed learning over 5G is the way to scale training beyond the cloud. Let’s walk through how this could work at a high level. The first step is that a central or edge cloud sends a state-of-art global AI model to the devices. Next, each device collects personal data and performs on-device training (see Figure 6). Large scale training is very computationally intensive, which is why it has been done in the cloud until now. By doing small training runs on smaller datasets, the workload becomes more manageable. Plus, on-device AI capabilities have been increasing exponentially, along with improvements in algorithms and software.
On-device training has three very important benefits that will lead to mass adoption of AI:
- Scale: By spreading processing over many devices, such as millions of smartphones, we can harness a significant amount of computational power.
- Personalization: With your own data used for your device training, the AI model learning is inherently customized.
- Privacy: The raw data never leaves your device to go to the cloud. By training on device with the data, you are extracting the value of the data and preserving privacy.
The next steps are iteratively improving the state-of-the-art global AI model. Without sending raw data to the cloud, the question becomes how can we improve the global AI model while continuing to maintain privacy? We accomplish this by adding noise to the parameters of the AI model to obfuscate the data, compressing the parameters, encrypting the compressed model, and finally sending the update to the cloud. The cloud then updates the AI model based on input from all the devices and then sends the improved AI model to the devices (see Figure 7). This loop iterates so that the AI model keeps getting better.
Of course, it’s not quite as simple as this short summary, and so please tune in on February 6, 2020 at 9 a.m. PT for our where we will go into much more detail about the powerful intersection between 5G and AI.
At Qualcomm Technologies, we are not only working on cutting-edge research for 5G and AI, but we are also exploring their synergies to realize our vision of the intelligent wireless edge. If you’re excited about solving big problems with cutting-edge AI and 5G research — and improving the lives of billions of people — we’d like to hear from you. We’re recruiting for several and career openings. Join us to help create what’s next in AI and wireless.