[论文解读] The Graph Exploration Problem with Advice

Hans-Joachim Böckenhauer, Janosch Fuchs+1 more|arXiv (Cornell University)|Apr 18, 2018

Optimization and Search Problems参考文献 5被引用 5

一句话总结

本文提出了一种图探索问题的在线算法，结合建议复杂度，通过一种新颖的度数有界图转换方法和自适应的边分类技术，在有向图和无向图上实现了最优性能。关键贡献在于，对于稀疏图，建议复杂度达到 O(m + n) 位，优于经典的 O(n log n) 边界，通过每条边使用对数级建议编码结构知识，并通过定制化的边分类方案最小化冗余信息。

ABSTRACT

Moving an autonomous agent through an unknown environment is one of the crucial problems for robotics and network analysis. Therefore, it received a lot of attention in the last decades and was analyzed in many different settings. The graph exploration problem is a theoretical and abstract model, where an algorithm has to decide how the agent, also called explorer, moves through a network such that every point of interest is visited at least once. For its decisions, the knowledge of the algorithm is limited by the perception of the explorer. There are different models regarding the perception of the explorer. We look at the fixed graph scenario proposed by Kalyanasundaram and Pruhs (Proc. of ICALP, 1993), where the explorer starts at a vertex of the network and sees all reachable vertices, their unique names and their distance from the current position. Therefore, the algorithm recognizes already seen vertices and can adapt its strategy during exploring, because it does not forget anything. Because the algorithm only learns the structure of the graph during computation, it cannot deterministically compute an optimal tour that visits every vertex at least once without prior knowledge. Therefore, we are interested in the amount of crucial a-priori information needed to solve the problem optimally, which we measure in terms of the well-studied model of advice complexity. [..] We look at different variations of the graph exploration problem and distinguish between directed or undirected edges, cyclic or non-cyclic solutions, unit costs or individual costs for the edges and different amounts of a-priori structural knowledge of the explorer. [..] In this work, we present algorithms with an advice complexity of $\mathcal{O}(m+n)$, thus improving the classical bound for sparse graphs.

研究动机与目标

• 降低在未知环境中最优图探索的建议复杂度。
• 分析实现最优探索路径所需的最少先验信息（建议）量。
• 开发高效的算法，利用建议在有向图和无向图中实现最小建议开销的探索。
• 将现有建议复杂度模型扩展至处理循环与非循环探索序列。
• 在固定图场景下，为一般图、稀疏图和无向图提供建议复杂度的紧致界。

提出的方法

• 将任意有向图转换为度数有界的图（每个顶点的入边和出边最多为两条），以简化建议编码。
• 对每条无向边使用“六选一”问题模型，对最优路径中的边使用情况进行分类（0次、1次或双向多次遍历）。
• 利用建议位编码边分类（如 EMulti、E1、E0）和遍历模式，减少冗余。
• 采用两阶段建议策略：第一阶段，使用每条无向边 log(6) 位对边类型进行分类；第二阶段，编码变换后图中的遍历序列。
• 利用现有定理（如定理 5.11、6.3）对变换后图中的建议使用量进行界定。
• 通过结构不变性及顶点/边映射，将变换后图的解回传至原始图。

实验结果

研究问题

• RQ1 在有向图上解决循环图探索问题所需的最小建议复杂度是多少？
• RQ2 通过利用图的结构性质，能否将稀疏图的建议复杂度降低至 O(n log n) 以下？
• RQ3 当移除边的方向性时，无向图的建议复杂度如何变化？
• RQ4 将图转换为度数有界形式对建议需求有何影响？
• RQ5 能否通过利用循环探索解，以最小建议开销求解非循环探索序列？

主要发现

• 本文在有向图的循环图探索中实现了 2n + 23m 位的建议复杂度，优于稀疏图的古典 O(n log n) 边界。
• 对于无向图，建议复杂度降低至 log(6)(n + m) + 42m 位，通过每条边的“六选一”分类实现。
• 度数有界图转换使顶点和边的数量最多增加两倍，但建议成本随 m 和 n 线性增长。
• 该建议模型允许以最少信息实现最优探索，表明 O(m + n) 位足以在稀疏图中实现最优解。
• 该方法可通过在循环解基础上增加 ⌈log(n−1)⌉ 位，推广至非循环探索，实现高效适应。
• 结果表明，边分类与结构变换显著降低了建议开销，相比通用建议方案具有明显优势。