Optimal Algorithms for the Online Time Series Search Problem
COCOA '09 Proceedings of the 3rd International Conference on Combinatorial Optimization and Applications
Optimal algorithms for the online time series search problem
Theoretical Computer Science
Online algorithms for the multiple time series search problem
Computers and Operations Research
Optimal algorithms for online time series search and one-way trading with interrelated prices
Journal of Combinatorial Optimization
How much is it worth to know the future in online conversion problems?
Discrete Applied Mathematics
Hi-index | 0.00 |
Online search is a basic online problem. The fact that its optimal deterministic/randomized solutions are given by simple formulas (however with difficult analysis) makes the problem attractive as a target to which other practical online problems can be transformed to find optimal solutions. However, since the upper/lower bounds of prices in available models are constant, natural online problems in which these bounds vary with time do not fit in the available models. We present two new models where the bounds of prices are not constant but vary with time in certain ways. The first model, where the upper and lower bounds of (logarithmic) prices have decay speed, arises from a problem in concurrent data structures, namely to maximize the (appropriately defined) freshness of data in concurrent objects. For this model we present an optimal deterministic algorithm with competitive ratio $\sqrt{D}$, where D is the known duration of the game, and a nearly-optimal randomized algorithm with competitive ratio $\frac{\ln D}{1+\ln2-\frac{2}{D}}$. We also prove that the lower bound of competitive ratios of randomized algorithms is asymptotically $\frac{\ln D}{4}$. The second model is inspired by the fact that some applications do not utilize the decay speed of the lower bound of prices in the first model. In the second model, only the upper bound decreases arbitrarily with time and the lower bound is constant. Clearly, the lower bound of competitive ratios proved for the first model holds also against the stronger adversary in the second model. For the second model, we present an optimal randomized algorithm. Our numerical experiments on the freshness problem show that this new algorithm achieves much better/smaller competitive ratios than previous algorithms do, for instance 2.25 versus 3.77 for D=128.