AI as a game changer for intelligent Edge and transportation systems, including intelligent network control
AI is a game changer for various thematic areas within networks and distributed systems - below we draw attention to three sub-themes that we believe will have a tremendous impact on our modern society.
NB! Applicants are asked to apply for one of these sub-themes. Please indicate clearly which of the sub-themes you have chosen for your proposal by using one of the codes ND1, ND2, ND3.
Mentoring and internship will be offered by a relevant external partner.
Edge Intelligence (EI) is revolutionizing various sectors by bringing AI capabilities closer to the data source, enabling faster and more efficient processing. EI offers real-time processing, reduced latency, and improved performance, particularly in applications with stringent time-sensitive requirements. Herein we aim to harness EI to support a wide range of future services, including eHealth, Industry 5.0, smart cities, autonomous vehicles, and education. We will focus on real-time analytics and decision-making at the edge to optimize resource allocation, enhance user experiences, and enable new services requiring low latency and high reliability.
AI-driven digital twin technology maximizes the application of urban big data in the digital information space, optimizing smart city management and service business processes. The speed development of communication technologies, such as 6G and vehicular networks, fills the gap between the real world and the digital twin exchanging vast data at fast speed and flexible topologies. By deploying monitoring devices and roadside units, collecting real-time traffic data, and analyzing and applying it extensively, real-time conditions can be highly restored, offering more intelligent and adaptive solutions for optimized resource utilization, smart traffic management, and environmental monitoring (i.e., Cognitive Smart City Services).
AI is also making inroads to network control - when evaluating the performance of a network algorithm (e.g., a congestion control scheme), how to generate realistic background load is an important and hard question. Traditionally, such evaluations have been done under quite simplified conditions. Now, the bar for the necessary realism of competing network traffic in performance evaluation has been raised. Large corporations have become dominant in the standardization of new network technology, which is primarily because they can present realistic evaluation results from their real customer base – a similar type of evaluation is impossible for academics. If academics want to play a role for the future of networking, they need a means to generate convincingly realistic reactive network traffic, representing an application mix that one might encounter on a real home or office network. Such a traffic generator could also allow customers and service providers to better understand what, if any, benefits a capacity upgrade would give them (compared to traditional “speed test” methods).
Theme ND1. Edge Intelligence for Transformative Applications in eHealth and Industry
- Contact person: ?zgü Alay
- Keywords: Edge Intelligence, Mobile Networks, Future Applications, Cognitive Services
- Research group: Network and Distributed Systems (ND)
Edge Intelligence (EI) is revolutionizing various sectors by bringing AI capabilities closer to the data source, enabling faster and more efficient processing. EI offers real-time processing, reduced latency, and improved performance, particularly in applications with stringent time-sensitive requirements.
This project aims to harness EI to support a wide range of future services, including eHealth, Industry 5.0, smart cities, autonomous vehicles, and education. We will focus on real-time analytics and decision-making at the edge to optimize resource allocation, enhance user experiences, and enable new services requiring low latency and high reliability.
Our research will address the joint study of network and compute resources, including the complexities of distributed AI training and inference. We aim to balance these challenges with the stringent latency and reliability requirements inherent in these applications. Through collaboration with external partners, this project will pioneer the development and application of Edge Intelligence for transformative use in eHealth and industrial sectors. By leveraging state-of-the-art AI at the network edge, we aim to create real-time solutions that offer both technological advancements and
significant societal benefits.
Methodological Research Topics:
- AI algorithms for the network edge
- Predictive modeling and transfer learning
- Explainability and interpretability of AI models
- Real-time data processing and analytics
Application Domains:
- eHealth: EI can revolutionize patient monitoring and interventions through robotic systems, allowing for real-time health data analysis and personalized healthcare services. This enables improved patient outcomes, timely interventions, and a reduced burden on healthcare facilities.
- Industry: EI can enhance smart manufacturing processes and predictive maintenance in manufacturing and assembly operations. Real-time data processing at the edge improves operational efficiency , reduces downtime, and allows for more flexible and responsive production systems.
External Partners:
- In collaboration with our main partner, Telenor AS (confirmed), we will work with other leading companies and organizations such as ABB, Kongsberg, Cognata, Rikshospitalet, and Sunn?s Sykehus to advance this research.
Theme ND2. Intelligent Transportation with AI-driven Digital Twins
- Contact person: Tor Skeie
- Keywords: Intelligent Transportation, Vehicular Networks, Environmental Monitoring, Digital Twins, Cognitive Services
- Research group: Network and Distributed Systems (ND)
AI-driven digital twin technology maximizes the application of urban big data in the digital information space, optimizing smart city management and service business processes. The speed development of communication technologies, such as 6G and vehicular networks, fills the gap between the real world and the digital twin exchanging vast data at fast speed and flexible topologies. By deploying monitoring devices and roadside units, collecting real-time traffic data, and analyzing and applying it extensively, real-time conditions can be highly restored, offering more intelligent and adaptive solutions for optimized resource utilization, smart traffic management, and environmental monitoring (i.e., Cognitive Smart City Services). Simultaneously, leveraging historical and real-time monitoring data from vehicular networks, there is a great possibility to achieve real-time vehicle tracking and traffic management through AI-driven digital twin, enabling swift fault response, monitoring urban traffic operation, intelligent traffic flow dispatching, and predicting future traffic demands, thereby enhancing.
Relevance of methodological research:
- Self-adaptation in digital twins
- digital twin life-cycle
- Decision-making with digital twins
Relevance of natural sciences or technology:
- Engineering
- Application Domains:
- Public transportation
- Commercial transportation
- Private transportation
Potential external partners:
- The Avinor Group
- Statkraft AS
- Institute for Energy Technology
Theme ND3. Computer Networks, Future Internet Standards
- Contact person: Michael Welzl
- Keywords: Computer Networks, Internet Protocols, Network Traffic, Future Internet Standards
- Research group: Network and Distributed Systems (ND)
When evaluating the performance of a network algorithm (e.g., a congestion control scheme), how to generate realistic background load is an important and hard question. Traditionally, such evaluations have been done under quite simplified conditions. Now, the bar for the necessary realism of competing network traffic in performance evaluation has been raised. Large corporations have become dominant in the standardization of new network technology, which is primarily because they can present realistic evaluation results from their real customer base – a similar type of evaluation is impossible for academics. If academics want to play a role for the future of networking, they need a means to generate convincingly realistic reactive network traffic, representing an application mix that one might encounter on a real home or office network. Such a traffic generator could also allow customers and service providers to better understand what, if any, benefits a capacity upgrade would give them (compared to traditional “speed test” methods). This could help to avoid unnecessarily early hardware upgrades, which would reduce embodied CO2 emissions and make networks more sustainable.
The plan is to:
- Devise a method to test the reaction of diverse Internet applications to network disturbances, to obtain training data;
- Train ML models for reactive traffic generation from these data, to represent these applications;
- Fine-tune these ML models to be able to represent traffic aggregates consisting of a heterogeneous mix of traffic sources, which would react to network disturbances in different ways and at different time scales.
Methodological Research Topics:
- Supervised Machine Learning (and likely: Reinforcement Learning)
- Network Simulation
- Network Emulation
- Internet congestion control
Application Domains:
Internet standardization: Internet standards become increasingly dominated by huge corporations. The Internet started from academia, and the role of academia as a neutral player to bring forward new standards must be preserved if the Internet should remain fairly usable by anyone.
External Partners:
This project will benefit from collaboration with a network of colleagues from both academia (primarily) and industry within the IETF community.