Online Optimization with Lookahead
Fabian Dunke  (Karlsruher Institut fuer Technologie)

We consider online optimization problems where algorithms can acces a 
quantifiable amount of information on future input data. We call
this optimization paradigm online optimization with lookahead and present a 
generic framework for it. First off, we look at several lookahead types, their 
applications in practice, and what can be gained by them in terms of classical 
worst performance guarantees. To overcome some of the weaknesses in this type 
of analysis, alternative performance measures particularly suitable for 
simulation purposes are proposed. In the center of this talk stands the generic 
framework for online optimization with lookahead which is introduced to 
establish the problem setting as an intermediate form between pure offline and 
online optimization.

In addition, the framework intends to harmonize concepts and notation 
throughout application domains. Finally, we examine in two
applications how lookahead affects overall optimization results. We find out 
that an exact (local) solution of subproblems does not
necessarily contribute to an improved overall (global) solution, and that the 
value of lookahead is strongly problem-specific.