Techniques for the allocation of resources under uncertainty
|Abstract:||Resource allocation is an ubiquitous problem that arises whenever limited resources have to be distributed among multiple autonomous entities (e.g., people, companies, robots, etc). The standard approaches to determine the optimal resource allocation are computationally prohibitive. The goal of this thesis is to propose computationally efficient algorithms for allocating consumable and non-consumable resources among autonomous agents whose preferences for these resources are induced by a stochastic process. Towards this end, we have developed new models of planning problems, based on the framework of Markov Decision Processes (MDPs), where the action sets are explicitly parameterized by the available resources. Given these models, we have designed algorithms based on dynamic programming and real-time heuristic search to formulating thus allocations of resources for agents evolving in stochastic environments. In particular, we have used the acyclic property of task creation to decompose the problem of resource allocation. We have also proposed an approximative decomposition strategy, where the agents consider positive and negative interactions as well as simultaneous actions among the agents managing the resources. However, the main contribution of this thesis is the adoption of stochastic real-time heuristic search for a resource allocation. To this end, we have developed an approach based on distributed Q-values with tight bounds to diminish drastically the planning time to formulate the optimal policy. These tight bounds enable to prune the action space for the agents. We show analytically and empirically that our proposed approaches lead to drastic (in many cases, exponential) improvements in computational efficiency over standard planning methods. Finally, we have tested real-time heuristic search in the SADM simulator, a simulator for the resource allocation of a platform.|
|Document Type:||Thèse de doctorat|
|Open Access Date:||13 April 2018|
|Collection:||Thèses et mémoires|
All documents in CorpusUL are protected by Copyright Act of Canada.