Multi-agent decision making in a stochastic environment, with application to 6G Edge computing // Multi-agent decision making in a stochastic environment, with application to 6G Edge computing

The introduction of new infrastructure, such as Mobile Edge Computing (MEC), introduces massive energy consumption, both in equipment manufacturing and operation, which should be minimised. This infrastructure can be owned by multiple agents, for instance Infrastructure Providers (IPs), which can cooperate in order to reach this goal.

The aim of the thesis is to propose a theoretical framework to describe the short term and long term decision making strategies that each agent would pursue in line with their own interest. These strategies include the amount of investment in computation capacity and in energy resources, the willingness to share infrastructure with other agents, etc. Our research question is to understand under which conditions such strategies converge to a sustainable outcome.

The challenge in this setting is that agents decide their strategies under imperfect information. Indeed, future demand as well as the availability and prices of energy and material, more or less green, that are used to operate the infrastructure are uncertain. Agents decisions are based on a belief of future conditions. This belief which can be modelled via probabilistic forecasts which are continuously updated as new information becomes available. This results in dynamic and stochastic agent behaviour, with continuously revised decision strategies. The setting is further complicated by the heterogeneity of risk aversion profile of agents: decisions may be more or less conservative face to uncertainty.

To tackle this work, we will make use of tools from decision theory, game theory, stochastic or distributionaly robust optimisation, and probabilistic modeling.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

The introduction of new infrastructure, such as Mobile Edge Computing (MEC), introduces massive energy consumption, both in equipment manufacturing and operation, which should be minimised. This infrastructure can be owned by multiple agents, for instance Infrastructure Providers (IPs), which can cooperate in order to reach this goal.

The aim of the thesis is to propose a theoretical framework to describe the short term and long term decision making strategies that each agent would pursue in line with their own interest. These strategies include the amount of investment in computation capacity and in energy resources, the willingness to share infrastructure with other agents, etc. Our research question is to understand under which conditions such strategies converge to a sustainable outcome.

The challenge in this setting is that agents decide their strategies under imperfect information. Indeed, future demand as well as the availability and prices of energy and material, more or less green, that are used to operate the infrastructure are uncertain. Agents decisions are based on a belief of future conditions. This belief which can be modelled via probabilistic forecasts which are continuously updated as new information becomes available. This results in dynamic and stochastic agent behaviour, with continuously revised decision strategies. The setting is further complicated by the heterogeneity of risk aversion profile of agents: decisions may be more or less conservative face to uncertainty.

To tackle this work, we will make use of tools from decision theory, game theory, stochastic or distributionaly robust optimisation, and probabilistic modeling.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Début de la thèse : 01/10/2025

Plan Investissement d'Avenir (Idex, Labex)*

Télécom SudParis

626 Ecole Doctorale de l'Institut Polytechnique de Paris

The prospective candidate must have excellent modelling and analytical skills, with a Masters or equivalent degree in Applied Mathematics, Engineering or Computer Science. Please send a copy of your CV, transcripts of all diplomas, half a page statement of purpose, and recommendation letters. Sending your ranking is not mandatory (but it is a big plus).
The prospective candidate must have excellent modelling and analytical skills, with a Masters or equivalent degree in Applied Mathematics, Engineering or Computer Science. Please send a copy of your CV, transcripts of all diplomas, half a page statement of purpose, and recommendation letters. Sending your ranking is not mandatory (but it is a big plus).