Periodic Reporting for period 1 - LANTERN (Low-latency and private edge computing in random-access networks)
Okres sprawozdawczy: 2021-10-01 do 2023-09-30
1. The problem/issue addressed
The problem addressed in the project is to develop a communication and computation infrastructure that is able to support the processing of a vast amount of data generated/collected by Internet-of-Things (IoT) devices/users. For communication, we addressed the problem of massive random access where a large number of devices send information to a common server in a sporadic and time-varying manner. For computation, we considered edge computing where the processing tasks, such as machine learning, is performed in edge servers deployed close to the users.
2. The importance for society
This problem is important because to the realization of the IoT requires such a communication and computation infrastructure. The IoT is a key enabler for a variety of applications such as immersive communications, autonomous driving, smart cities, and smart factories.
3. The overall objectives
The main goal of this project is to investigate how low-latency and privacy-preserving edge computing protocols can be developed in wireless random-access networks. To achieve this main goal, we specified two specific goals (SGs):
• SG1: Establish a foundation for privacy and reliability in latency-critical edge computing in random-access networks.
• SG2: Devise resilient coding schemes to achieve low latency and preserve privacy in distributed edge computing.
SG1 aims at understanding the fundamental limits, while SG2 aims at designing practical schemes for low-latency and private edge computing in random-access networks.
The problem addressed in the project is to develop a communication and computation infrastructure that is able to support the processing of a vast amount of data generated/collected by Internet-of-Things (IoT) devices/users. For communication, we addressed the problem of massive random access where a large number of devices send information to a common server in a sporadic and time-varying manner. For computation, we considered edge computing where the processing tasks, such as machine learning, is performed in edge servers deployed close to the users.
2. The importance for society
This problem is important because to the realization of the IoT requires such a communication and computation infrastructure. The IoT is a key enabler for a variety of applications such as immersive communications, autonomous driving, smart cities, and smart factories.
3. The overall objectives
The main goal of this project is to investigate how low-latency and privacy-preserving edge computing protocols can be developed in wireless random-access networks. To achieve this main goal, we specified two specific goals (SGs):
• SG1: Establish a foundation for privacy and reliability in latency-critical edge computing in random-access networks.
• SG2: Devise resilient coding schemes to achieve low latency and preserve privacy in distributed edge computing.
SG1 aims at understanding the fundamental limits, while SG2 aims at designing practical schemes for low-latency and private edge computing in random-access networks.
After an intensive research period of two years, we were able to obtain both: 1) fundamental understanding of the reliability, latency, energy efficiency, and privacy in massive random-access networks and edge learning systems, and 2) practical design of communication protocols and computation offloading schemes that guarantee the freshness of the delivered content as well as privacy against a malicious eavesdropper. The two SGs were accomplished by various publications during the project.
Specifically, for SG1, we have achieved the following results:
• A bound on the achievable reliability and energy efficiency of an unsourced random-access system with random and unknown number of active users.
• A bound on the achievable reliability and energy efficiency of an unsourced random-access system with coexistence of both massive and critical IoT traffic types.
• An analysis of privacy in federated learning with secure aggregation.
For SG2, we have achieved the following results:
• Design and analysis of a random-access protocol based on irregular repetition slotted ALOHA (IRSA) over the binary adder channel.
• Design and analysis of an IRSA-based random-access protocol that guarantees information freshness for multiple classes of users.
• Design and analysis of a slotted ALOHA random-access protocol that guarantees information freshness for energy-harvesting users.
• Joint optimization of communication and computation offloading in edge computing, assisted by reconfigurable intelligent surface.
We have also conducted a survey on the integration of edge computing and information-centric networking for future networks.
Specifically, for SG1, we have achieved the following results:
• A bound on the achievable reliability and energy efficiency of an unsourced random-access system with random and unknown number of active users.
• A bound on the achievable reliability and energy efficiency of an unsourced random-access system with coexistence of both massive and critical IoT traffic types.
• An analysis of privacy in federated learning with secure aggregation.
For SG2, we have achieved the following results:
• Design and analysis of a random-access protocol based on irregular repetition slotted ALOHA (IRSA) over the binary adder channel.
• Design and analysis of an IRSA-based random-access protocol that guarantees information freshness for multiple classes of users.
• Design and analysis of a slotted ALOHA random-access protocol that guarantees information freshness for energy-harvesting users.
• Joint optimization of communication and computation offloading in edge computing, assisted by reconfigurable intelligent surface.
We have also conducted a survey on the integration of edge computing and information-centric networking for future networks.
Our results for SG1 extended the state-of-the-art understanding of the limits for reliability and energy efficiency of random-access networks. Our bounds serve as performance benchmark for practical coding schemes. It tells how far the performance of these schemes is from the best achievable performance, i.e. how much improvement can be expected. These bounds have been used in some other researches. Furthermore, our analysis of privacy in federated learning breaks a common belief that secure aggregation preserves privacy: we show that it only provides weak privacy, and thus additional privacy enhancing techniques should be used.
Our results for SG2 provided design guidelines for practical communication protocols and computation offloading schemes from the users to the server. We explored a novel metric for information freshness, i.e. the age of information (AoI) metric. Guaranteeing information freshness ensures that the information sent from the users to the server remains relevant.
Our results for SG2 provided design guidelines for practical communication protocols and computation offloading schemes from the users to the server. We explored a novel metric for information freshness, i.e. the age of information (AoI) metric. Guaranteeing information freshness ensures that the information sent from the users to the server remains relevant.