5G is about more than speed and access. Multiplying the capacity of wireless networks exponentially will bring connected cars, virtual reality and the Internet of Things into daily life. Innovation in the Intel-powered 5G ecosystem also will allow these new technologies to operate seamlessly on networks that have never been more challenging to run.
Nokia estimates the complexity of operating wireless networks will increase by a factor of 50 in the transition from 4G to 5G. The stakes for users go beyond cell reception. Smart cities, remote health monitoring, and automated manufacturing lines won’t tolerate downtime. Consumers’ lives also will become more intertwined with the web through wearable tech and smart media. The full potential of 5G will require something beyond new equipment: artificial intelligence (AI), machine learning (ML), and automation.
Machine Learning: Innovating Behind the Scenes
AI/ML processes a massive collection of data points looking for useful patterns. It then evaluates what it finds and modifies itself to improve the results of the next analysis. The patterns ML reveals can uncover new efficiencies and predict the likelihood of problems in ways that humans using traditional software can’t. Taking advantage of those predictions, automation will further enable optimization.
As AI/ML systems become widely deployed, they will analyze not only parameters for specific functions and elements but also how they interact—a catalyst for end-to-end network optimization. Then, optimization will move beyond maximizing individual performance to maximizing a target that depends on multiple network functions. For example, optimizing QoE requires operators to account for edge computing, backhaul, fronthaul, RAN conditions, and more. AI/ML will be able to provide recommendations to optimize the complex system of considerations.
Other areas where ML will be instrumental to 5G include traffic management, carrier aggregation, network slicing, beamforming, real-time troubleshooting, integrating earlier mobile technologies, and improving service quality.
Human Judgment Meets Machine Learning
For a deeper look, Intel recently sought the insight of industry leaders exploring AI/ML in 5G. At Nokia Mobile Networks, Tero Rissa applies 20 years of engineering and research experience as chief architect of machine learning.
“There is a lot of hype. And because of the hype and the lack of understanding of machine learning or AI, people incorrectly assume that it is magic,” he noted. “Machine learning is the automation of decisions. Human power can’t be used to make millions of consistent decisions in real time. We can use machines to make those decisions in an automated fashion.”
Over the last 10 years, related academic research has increased dramatically, while frameworks like Keras, TensorFlow, and PyTorch make development dramatically faster. At Nokia, Rissa’s team is studying ways AI can increase the productivity of radio networks. This includes the algorithms that run them, as well as opportunities for efficiency at the edge. In the future, Rissa is most excited about the potential for self-organization networks that can save resources through automated configuration, security management, and predictive maintenance.
Network optimization startup Uhana is already building systems to put AI in the hands of carriers. Founder Sachin Katti uses the analogy of a control panel with knobs adjusting an array of interrelated factors that can improve service and prevent problems.
“Uhana is building a real-time engine that learns the control plane of the network. We are building an engine that can tune all of these knobs based on real-time data, independent of human intervention,” said Katti. For example, the bitrate of a video could adjust automatically to enhance the next four seconds of content based on live network conditions.
There is already substantial commercial interest in this system, which uses the same type of deep-reinforcement learning Google used to beat human players in the strategy game “go.”
Intelligent Networks: Helping 5G Flourish
Humans aren’t very good at solving multi-dimensional problems. This is already a challenge in 4G and multi-access edge computing (MEC). Soon, 5G will add even more urgency to the development of AI that can operate wireless networks at their full potential. Intel continues to enhance processing capability for AI and ML algorithms as part of a roadmap to speed adoption of the technology.
This is not the mind-like general intelligence of science fiction. Instead, artificial intelligence and Machine Learning represent powerful tools to augment how people analyze and operate complex systems. 5G will be among the biggest beneficiaries as AI becomes smarter, adding more power to the networks set to revolutionize media, transportation, and business in the years ahead.
Learn more in our report from Senza Fili, AI and Machine Learning: Why Now? Network Optimization in the Age of 5G. And follow the latest Intel AI news at Intel.ai.