Abstract:We introduce the study of search games between a mobile Searcher and an immobile Hider in a new setting in which the Searcher has some potentially erroneous information, i.e., a prediction on the Hider's position. The objective is to establish tight tradeoffs between the consistency of a search strategy (i.e., its worst case expected payoff assuming the prediction is correct) and its robustness (i.e., the worst case expected payoff with no assumptions on the quality of the prediction). Our study is the first to address the full power of mixed (randomized) strategies; previous work focused only on deterministic strategies, or relied on stochastic assumptions that do not guarantee worst-case robustness in adversarial situations. We give Pareto-optimal strategies for three fundamental problems, namely searching in discrete locations, searching with stochastic overlook, and searching in the infinite line. As part of our contribution, we provide a novel framework for proving optimal tradeoffs in search games which is applicable, more broadly, to any two-person zero-sum games in learning-augmented settings.
What problem does this paper attempt to address?
### Problems the paper attempts to solve
This paper aims to study the search game between the Searcher and the Hider in the presence of potentially misinformation (i.e., prediction). Specifically, the goal of the paper is to establish the optimal trade - off relationship between the consistency and the robustness of the search strategy.
- **Consistency**: refers to the worst - case expected payoff of the search strategy under the assumption that the prediction is accurate.
- **Robustness**: refers to the worst - case expected payoff of the search strategy without considering the accuracy of the prediction.
### Research background and problem description
Traditional search games usually assume that there are no assumptions about each other's strategies between the searcher and the hider. However, in practical applications, the searcher can often obtain some predictive information about the hider's location, although this information may not be completely accurate. For example, in search - and - rescue operations, the rescue team may speculate on the location of the missing person based on the testimony of witnesses, so as to find the target more quickly.
By introducing predictive information, the paper explores how to design the optimal search strategy in this situation and ensure that these strategies can perform well both when the prediction is accurate and inaccurate.
### Specific research contents
1. **Discrete - location search**: study the problem of finding the hider in a finite set of discrete locations, with a specified search cost for each location.
2. **Search with random omission**: study the search strategy in the situation where there is a certain probability of failing to find the hider during each search.
3. **Infinite linear search**: study the problem of finding the hider on an infinite line, and the searcher needs to alternately explore the left and right half - lines.
### Main contributions
The paper provides Pareto - optimal strategies for the above three basic problems, which can achieve the best trade - off between consistency and robustness. This is the first work to conduct such an analysis on randomized strategies (mixed strategies), not only in search games, but also more widely applicable to games with imperfect prediction.
### Methods and challenges
The introduction of randomized strategies brings new challenges, especially when applying the minimax theorem to obtain the lower bound of the performance of the search strategy. Through the method of multi - objective optimization, the paper finds a new game value, which enables the searcher to minimize the linear combination of consistency and robustness, thus providing a new methodology.
### Conclusion
By introducing predictive information, the paper expands the research scope of traditional search games and provides a theoretical basis for designing search strategies that perform well in uncertain environments.