Objective
The goal of the project is to contribute to design efficient algorithms with proven worst-case performance guarantees, under strict limitations either on the running time or on the accessibility of the data. For NP-hard optimisation problems, constraints on the running time can often be satisfied by approximation algorithms; the second limitation leads to the notion of on-line algorithms. The project will address its objectives focusing on both traditional paradigmatic problems (graph theoretical, scheduling and packing problems) and algorithmic problems arising in new information technologies (resource management, communication and data management, telecommunication and other areas).
OBJECTIVES
Our theoretical objective is the development of a unified and well defined theory, with respect to computation practice, for approximation and on-line algorithms. The practical objective is to exploit this theory in real world applications and obtain practical algorithmic tools. The theoretically proven algorithms will be used as core algorithmic ideas that reflect the combinatorial structure of problems but need to be fine-tuned to the types of instances arising in specific applications to lead to practical algorithmic tools. Thus, there is a need for implementing and experimenting with the algorithms developed. The feedback from empirical experimentation is also at the basis of the critical revision of theoretical models and design techniques of approximation and on-line algorithms.
DESCRIPTION OF WORK
The design and management of computer systems and communication networks give rise to many combinatorial optimisation problems, that are usually NP-hard. We are interested in efficient algorithms that allow obtaining, on every input instance of a specific problem, a provably good solution, e.g. which is within a guaranteed factor (performance ratio) of the optimum. To understand the combinatorial structures that yield efficient approximation algorithms, we will focus on cornerstone problems.
Another source of difficulties that prevents one from obtaining optimal solutions arises when the input instance of a problem is not immediately available, in areas such as resource allocation in operating systems, distributed computing, scheduling, communication network. An on-line algorithm must deal with events as they arrive and take decisions without knowledge of future events.
We will focus on complementing competitive analysis with other tools or more relaxed assumptions that will make the results significant in those cases in which an omni powerful off-line adversary makes competitive analysis too pessimistic and conceals the real performances of an algorithm. New advances in the area should also be evaluated from a practical point of view by experimenting and by providing guidelines for improving heuristics. Thus, we will implement and experimentally evaluate the algorithms developed.
Fields of science
Topic(s)
Call for proposal
Data not availableFunding Scheme
THN - Thematic network contractsCoordinator
24098 KIEL
Germany