Abstract:This paper takes a game theoretic approach to the design and analysis of online algorithms and illustrates the approach on the finite-horizon ski-rental problem. This approach allows beyond worst-case analysis of online algorithms. First, we define "subgame optimality" which is stronger than worst case optimality in that it requires the algorithm to take advantage of an adversary not playing a worst case input. Algorithms only focusing on the worst case can be far from subgame optimal. Second, we consider prior-independent design and analysis of online algorithms, where rather than choosing a worst case input, the adversary chooses a worst case independent and identical distribution over inputs. Prior-independent online algorithms are generally analytically intractable; instead we give a fully polynomial time approximation scheme to compute them. Highlighting the potential improvement from these paradigms for the finite-horizon ski-rental problem, we empirically compare worst-case, subgame optimal, and prior-independent algorithms in the prior-independent framework.
What problem does this paper attempt to address?
The core problem that this paper attempts to solve is to improve the design and evaluation of online algorithms by introducing a new analysis framework, especially the performance when facing non - worst - case inputs. Specifically, the article proposes two new concepts beyond worst - case analysis: **Subgame Optimality** and **Prior - Independent Design**.
### 1. Subgame Optimality
Traditional online algorithm analysis usually focuses on the competitive ratio in the worst - case, that is, the ratio of the algorithm performance in the worst - case to the performance of the optimal offline algorithm. However, this analysis method ignores the potential advantages that the algorithm can utilize in non - worst - case situations. Subgame optimality requires that the algorithm not only performs well in the worst - case, but also can fully take advantage of the mistakes of the opponent (or the environment) in non - worst - case situations, so as to achieve better performance. In other words, subgame optimality emphasizes that the algorithm should perform optimally in each sub - game, even if these sub - games are not the worst - case.
### 2. Prior - Independent Design
Prior - independent design analyzes and designs online algorithms under another assumption. In this framework, the opponent selects an independently and identically distributed input sequence, rather than a single worst - case input. The goal of prior - independent algorithms is to minimize the ratio between their performance and that of the optimal online algorithm that knows the input distribution. This method provides a more realistic input model and can give stronger performance guarantees than worst - case analysis.
### Application Case: Finite - Horizon Ski - Rental Problem
To illustrate the effectiveness of these new concepts, the paper selects the finite - horizon ski - rental problem as an application case. In this problem, the algorithm needs to decide whether to continue renting (continue to pay) or buy skiing equipment (pay a high cost at one time) after observing the weather conditions every day. The challenge for the algorithm is that it does not know the future weather conditions, so it needs to make decisions in an uncertain environment.
By comparing the performance of the worst - case optimal algorithm, the subgame optimal algorithm and the prior - independent algorithm in this problem, the paper shows how these new concepts can provide better performance guarantees than traditional worst - case analysis. In particular, the subgame optimal algorithm performs well in non - worst - case situations, and the prior - independent algorithm has an advantage in dealing with inputs of unknown distributions.
### Summary
The main contribution of this paper is to propose a new framework beyond worst - case analysis, and to show the effectiveness and potential advantages of these new methods through specific examples. This provides a new perspective for the design and analysis of online algorithms, especially when dealing with non - worst - case inputs, and can better capture the actual performance of algorithms.