Algorithms for planning under uncertainty and in the presence of selfish users

点击翻译按钮获取中文摘要

Technological advances over the past few years have led to the development of systems of rapidly increasing size and complexity. The design, maintenance and management of such systems, especially high-speed telecommunication networks and decentralized networks like the Internet, which are the important computing and communication platforms of the future, present several challenges. In this proposal, we shall focus on two important issues that arise in the design of such systems: (a) Uncertainty in the data or specifications. Uncertainty is a facet of many practical decision environments, and to function effectively one needs to anticipate uncertainty and take measures to be robust in the presence of uncertainty. We will investigate the key underlying algorithmic problems in models that  abstract these settings, and develop techniques and efficient algorithms for effectively tackling these problems. (b) The presence of uncoordinated components in the system with ``selfish'' interests. In an environment with self-interested users, like the Internet with its autonomous sub-networks, one can no longer assume that the users will act as stipulated by the system designer to optimize performance. We shall study, when, and to what extent, does selfish behavior affect performance, ways to mitigate the resulting inefficiencies, and ways to devise algorithms in such selfish environments that resist manipulation. In many cases, the algorithmic problems encountered are computationally intractable (e.g., NP-hard) optimization problems, so it is unlikely that exact solutions can be computed in polynomial time, and our approach will be to devise an approximation algorithm for the problem, that is, a polynomial-time algorithm that always delivers a provably near-optimal feasible solution. Approximation algorithms constitute a unifying theme of this research, and our research will also have impact on this rich and exciting field by furthering the development of general principles for the design and analysis of these algorithms.