Graph), a comprehensive benchmark of graph-based problem solving designed in natural language.

Here we cover the full specification of the experiments. Some details were omitted from the main text. If there were separate training and test sets, they were combined before creating the random 10-fold split. All attributes are normalized to mean 0 and standard deviation 1. Additional details for each model type follow.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. Thanks to your rebuttal, I think I now understand your algorithm, and I think it is correct. But why did you present in Figure 2 algorithm 2 with CB and not TCB? The algorithm with CB does not work, and it is misleading to put CB in Figure 2. I would recommend changing this and putting TCB in the presentation of your algorithm. Also, please comment on the necessity of knowing L(u_1,...,u_n) (or rather an upper bound on this, and rewrite the Thm with an upper bound since it is not realistic to have truly this quantity available).

Traditionally, these would be made according to hard-coded expert heuristics implemented in solvers.

A total of more than 3400 public shootings have occurred in the United States between 2016 and 2022. Among these, 25.1% of them took place in an educational institution, 29.4% at the workplace including office buildings, 19.6% in retail store locations, and 13.4% in restaurants and bars. During these critical scenarios, making the right decisions while evacuating can make the difference between life and death. However, emergency evacuation is intensely stressful, which along with the lack of verifiable real-time information may lead to fatal incorrect decisions. To tackle this problem, we developed a multi-route routing optimization algorithm that determines multiple optimal safe routes for each evacuee while accounting for available capacity along the route, thus reducing the threat of crowding and bottlenecking. Overall, our algorithm reduces the total casualties by 34.16% and 53.3%, compared to our previous routing algorithm without capacity constraints and an expert-advised routing strategy respectively. Further, our approach to reduce crowding resulted in an approximate 50% reduction in occupancy in key bottlenecking nodes compared to both of the other evacuation algorithms.

Game-theoretic resource allocation on graphs (GRAG) involves two players competing over multiple steps to control nodes of interest on a graph, a problem modeled as a multi-step Colonel Blotto Game (MCBG). Finding optimal strategies is challenging due to the dynamic action space and structural constraints imposed by the graph. To address this, we formulate the MCBG as a Markov Decision Process (MDP) and apply Reinforcement Learning (RL) methods, specifically Deep Q-Network (DQN) and Proximal Policy Optimization (PPO). To enforce graph constraints, we introduce an action-displacement adjacency matrix that dynamically generates valid action sets at each step. We evaluate RL performance across a variety of graph structures and initial resource distributions, comparing against random, greedy, and learned RL policies. Experimental results show that both DQN and PPO consistently outperform baseline strategies and converge to a balanced $50\%$ win rate when competing against the learned RL policy. Particularly, on asymmetric graphs, RL agents successfully exploit structural advantages and adapt their allocation strategies, even under disadvantageous initial resource distributions.

Graph computational tasks are inherently challenging and often demand the development of advanced algorithms for effective solutions. With the emergence of large language models (LLMs), researchers have begun investigating their potential to address these tasks. However, existing approaches are constrained by LLMs' limited capability to comprehend complex graph structures and their high inference costs, rendering them impractical for handling large-scale graphs. Inspired by human approaches to graph problems, we introduce a novel framework, PIE (Pseudocode-Injection-Enhanced LLM Reasoning for Graph Computational Tasks), which consists of three key steps: problem understanding, prompt design, and code generation. In this framework, LLMs are tasked with understanding the problem and extracting relevant information to generate correct code. The responsibility for analyzing the graph structure and executing the code is delegated to the interpreter. We inject task-related pseudocodes into the prompts to further assist the LLMs in generating efficient code. We also employ cost-effective trial-and-error techniques to ensure that the LLM-generated code executes correctly. Unlike other methods that require invoking LLMs for each individual test case, PIE only calls the LLM during the code generation phase, allowing the generated code to be reused and significantly reducing inference costs. Extensive experiments demonstrate that PIE outperforms existing baselines in terms of both accuracy and computational efficiency.