You are cordially invited to attend the PhD Defence of Arsham MOSTAANI which will take place in person on Friday, 12 May 2023 at 10:00 in meeting room CK D10/Paul Feidert in the Central Building.
Members of the Defence committee:
- Prof. Dr. Bhavani SHANKAR, University of Luxembourg, Chair
- Prof. Dr. Symeon CHATZINOTAS, University of Luxembourg, Vice-Chair
- Prof. Dr. Björn OTTERSTEN, University of Luxembourg, Supervisor
- Dr Jakob HOYDIS, Nvidia, Member
- Prof. Dr. Samson LASAULCE, University of Lorraine, Member
- Dr. Thang Xuan VU, Expert in an Advisory Capacity
Abstract :
As 5G is rolling out, we commence witnessing a surge of data-hungry applications in various domains such as IoT, industry 4.0, and autonomous vehicles. In contrast to the previous generations of cellular networks, 5G will serve many services in which a (subsystem of) ( an intelligent) machine is the receiving end of the communications. As soon as the receiving end of communications is no longer human, the ultimate goal of data transmission deviates from the traditional communication and data transmission systems. Under these circumstances, the communication is carried out only to address information deficiency at the receiving end. In particular, the receiving machine has a deficiency of information when the computations that it intends to carry out require further data than what is available to it. Communications are, thus, carried out to deliver the relevant/useful data required to perform the desired computations at the receiving end. This scenario demands a fresh approach for the design of the data transmission schemes since only the data that helps improve the computations of the receiving end are required to be transmitted.
Many of the services provided by cellular networks are traditionally designed to serve their primary users – humans. In contrast to humans, machines do not appreciate the extra descent quality of the received data/communications. The shift taking place in the number of non-human users, now, asks for revolutionary designs at all subsystems of a complete communication pipeline, where the relevance of data is taken into account when designing the specific subsystem. The relevance/usefulness of data can for instance change the way channel coding schemes behave by allowing the channel coder to know which data is more important to be protected. The relevance/usefulness of data can help the data compression schemes perform much more effectively, by discarding the part of data that will be useless in the computations of the receiving end. The relevance/usefulness can also help redesign the power/user scheduling schemes by giving priority to the users who are sharing more useful/relevant data. The importance of the massive research effort required to address these challenges becomes even more pronounced when we notice that by 2030, thirty billion machines will be served by communication networks. Task-oriented communication is an emerging field often overlapping with control theory, estimation theory, communication theory and machine learning, whose mission is to perform this fresh design at all layers of communication systems.
The focus of this thesis is on the task-oriented design of data compression/quantization methods. In particular, we limit ourselves to the design of data quantization algorithms for the control tasks and thus the task-oriented design of quantization for estimation tasks is out of the scope of this thesis. A very wide range of control tasks is classified under the control of multi-agent systems, where the current thesis finds its context. In multi-agent systems that operate under partial observability, inter-agent communications can prove as an essential tool to improve the overall performance of the system. We study different data compression schemes for communications between agents under different topologies of communication networks between agents. We also introduce schemes that have different capacities to scale with the number of agents in the system.
In particular, in chapter 3, we perform an indirect design of the communications in a multi-agent system (MAS) in which agents cooperate to maximize the averaged sum of discounted one-stage rewards of a collaborative task. Due to the bit-budgeted communications between the agents, each agent should efficiently represent its local observation and communicate an abstracted version of the observations to improve the collaborative task performance. We first show that this problem can be approximated as a form of data-quantization problem which we call task-oriented data compression (TODC). We then introduce the state-aggregation for information compression algorithm (SAIC) to solve the formulated TODC problem. It is shown that SAIC is able to achieve near-optimal performance in terms of the achieved sum of discounted rewards. The proposed algorithm is applied to a geometric consensus problem and its performance is compared with several benchmarks. Numerical experiments confirm the promise of this indirect design approach for task-oriented multi-agent communications.
Subsequently, in chapter 4, we consider a task-effective quantization problem that arises when multiple agents are controlled via a centralized controller (CC). While agents have to communicate their observations to the CC for decision-making, the bit-budgeted communications of agent-CC links may limit the overall performance of the system which is measured by the system’s average sum of stage costs/rewards. As a result, each agent should compress/quantize its observation such that the average sum of stage costs/rewards of the control task is minimally impacted. We address the problem of maximizing the average sum of stage rewards by proposing two different Action-Based State Aggregation (ABSA) algorithms that carry out the indirect and joint design of control and communication policies in the multi-agent system. While the applicability of ABSA-1 is limited to single-agent systems, it provides an analytical framework that acts as a stepping stone to the design of ABSA-2. ABSA-2 carries out the joint design of control and communication for a multi-agent system. We evaluate the algorithms – with average return as the performance metric – using numerical experiments performed to solve a multi-agent geometric consensus problem. The numerical results are concluded by introducing a new metric that measures the effectiveness of communications in a multi-agent system.
In our last technical chapter 5, we present a novel approach for designing scalable task-oriented quantization and communications in cooperative multi-agent systems (MAS). The proposed approach utilizes a task-oriented communication framework to enable efficient communication of observations between agents while optimizing the average return performance of the MAS, a parameter that quantifies the fulfilment of MAS’s task. Our approach uses the concept of the value of information to design quantization schemes that scale with the number of agents in the system. The designed quantization scheme enables agents to communicate task-relevant observations while minimizing the number of bits to be communicated. Computing the value of information, however, does not scale with the increasing number of agents in the MAS. We observe that one can reduce the computational cost of obtaining the value of information by exploiting insights gained from studying a similar two-agent system – instead of the original $N$-agent system. We then quantize agents’ observations such that their more valuable observations are communicated more precisely. We show analytically that under a wide range of problems, the proposed scheme is applicable. Our numerical results show that the proposed approach achieves significant improvements in reducing the computational complexity of the centralized training phase for the design of inter-agent communications in MAS problems while maintaining the average return performance of the system.