68W40 Analysis of algorithms [See also 68Q25]
Refine
Document Type
- ZIB-Report (2)
- Doctoral Thesis (1)
Language
- English (3)
Has Fulltext
- yes (3)
Is part of the Bibliography
- no (3)
Keywords
- online optimization (2)
- Markov decision problem (1)
- Online Optimization (1)
- Online-Optimierung (1)
- Randomisierte Algorithmen (1)
- Routing (1)
- Telecommunications (1)
- Traveling-Salesman-Problem (1)
- Zweistufiges Problem (1)
- column generation (1)
Institute
- Mathematical Optimization (3) (remove)
Providing realistic performance indicators of online algorithms for a given online optimization problem is a difficult task in general. Due to significant drawbacks of other concepts like competitive analysis, Markov decision problems (MDPs) may yield an attractive alternative whenever reasonable stochastic information about future requests is available. However, the number of states in MDPs emerging from real applications is usually exponential in the original input parameters. Therefore, the standard methods for analyzing policies, i.e., online algorithms in our context, are infeasible.
In this thesis we propose a new computational tool to evaluate the behavior of policies for discounted MDPs locally, i.e., depending on a
particular initial state. The method is based on a column generation algorithm for approximating the total expected discounted cost of an
unknown optimal policy, a concrete policy, or a single action (which assumes actions at other states to be made according to an optimal policy). The algorithm determines an $\varepsilon$-approximation by
inspecting only relatively small local parts of the total state space. We prove that the number of states required for providing the approximation is independent of the total number of states, which underlines the practicability of the algorithm. The approximations
obtained by our algorithm are typically much better than the
theoretical bounds obtained by other approaches.
We investigate the pricing problem and the structure of the linear programs encountered in the column generation. Moreover, we propose and analyze different extensions of the basic algorithm in order to achieve good approximations fast.
The potential of our analysis tool is exemplified for discounted MDPs emerging from different online optimization problems, namely online bin coloring, online target date assignment, and online elevator control. The results of the experiments are quite encouraging: our method is mostly capable to provide performance indicators for online algorithms that much better reflect observations made in simulations than competitive analysis does. Moreover, the analysis allows to reveal weaknesses of the considered online algorithms. This way, we developed a new online algorithm for the online bin coloring problem that outperforms existing ones in our analyses and simulations.
The Dynamic Multi-Period Routing Problem DMPRP introduced by Angelelli et al. gives a model for a two-stage online-offline routing problem. At the beginning of each time period a set of customers becomes known. The customers need to be served either in the current time period or in the following. Postponed customers have to be served in the next time period. The decision whether to postpone a customer has to be done online. At the end of each time period, an optimal tour for the customers assigned to this period has to be computed and this computation can be done offline. The objective of the problem is to minimize the distance traveled over all planning periods assuming optimal routes for the customers selected in each period. We provide the first randomized online algorithms for the DMPRP which beat the known lower bounds for deterministic algorithms. For the special case of two planning periods we provide lower bounds on the competitive ratio of any randomized online algorithm against the oblivious adversary. We identify a randomized algorithm that achieves the optimal competitive ratio of $\frac{1+\sqrt{2}}{2}$ for two time periods on the real line. For three time periods, we give a randomized algorithm that is strictly better than any deterministic algorithm.
We consider a multicommodity routing problem, where demands are released \emph{online} and have to be routed in a network during specified time windows. The objective is to minimize a time and load dependent convex cost function of the aggregate arc flow. First, we study the fractional routing variant. We present two online algorithms, called Seq and Seq$^2$. Our first main result states that, for cost functions defined by polynomial price functions with nonnegative coefficients and maximum degree~$d$, the competitive ratio of Seq and Seq$^2$ is at most $(d+1)^{d+1}$, which is tight. We also present lower bounds of $(0.265\,(d+1))^{d+1}$ for any online algorithm. In the case of a network with two nodes and parallel arcs, we prove a lower bound of $(2-\frac{1}{2} \sqrt{3})$ on the competitive ratio for Seq and Seq$^2$, even for affine linear price functions. Furthermore, we study resource augmentation, where the online algorithm has to route less demand than the offline adversary. Second, we consider unsplittable routings. For this setting, we present two online algorithms, called U-Seq and U-Seq$^2$. We prove that for polynomial price functions with nonnegative coefficients and maximum degree~$d$, the competitive ratio of U-Seq and U-Seq$^2$ is bounded by $O{1.77^d\,d^{d+1}}$. We present lower bounds of $(0.5307\,(d+1))^{d+1}$ for any online algorithm and $(d+1)^{d+1}$ for our algorithms. Third, we consider a special case of our framework: online load balancing in the $\ell_p$-norm. For the fractional and unsplittable variant of this problem, we show that our online algorithms are $p$ and $O{p}$ competitive, respectively. Such results where previously known only for scheduling jobs on restricted (un)related parallel machines.