Association for the Advancement of Artificial Intelligence: AAAI Publications
Not a member yet
26155 research outputs found
Sort by
Data Cleaning, Discard Studies, and Discretionary Power
No full textData cleaning is an overlooked yet impactful step in the Artificial Intelligence (AI) development pipeline, leading to negative downstream impacts when performed carelessly. Using Discard Studies as a framework, I propose an ethnographic study of how data practitioners exercise their discretionary power during the data cleaning process, particularly with respect to discarded data. The in-depth knowledge of the data cleaning process gained as a result of this study will allow us to improve guidelines and education on data cleaning for more ethical AI development
Pseudo-Boolean Proof Logging for Optimal Classical Planning
No full textWe introduce lower-bound certificates for classical planning tasks, which can be used to prove the unsolvability of a task or the optimality of a plan in a way that can be verified by an independent third party. We describe a general framework for generating lower-bound certificates based on pseudo-Boolean constraints, which is agnostic to the planning algorithm used.
As a case study, we show how to modify the A* algorithm to produce proofs of optimality with modest overhead, using pattern database heuristics and hmax as concrete examples. The same proof logging approach works for any heuristic whose inferences can be efficiently expressed as reasoning over pseudo-Boolean constraints
On Picking Good Policies: Leveraging Action-Policy Testing in Policy Training
No full textTesting is a natural approach to assess the quality of learned action
policies π. Prior work introduced policy testing in AI planning
as searching for bugs in π, that is, states where π is
sub-optimal with respect to a given testing objective. Beyond quality
assurance, an obvious application of these methods is policy
selection: given several π to choose from, we can use testing to
select the "least buggy" one. Here, we integrate testing-based
policy selection into the training process. This includes making more
informed decisions when selecting the final policy after training, as
well as choosing more promising intermediate policies during the
training process. Our experiments with ASNets action policies show
that integrating testing allows us to more reliably obtain
good-quality policies
Sorting Colored Balls in Colored Tubes
No full textWe consider a game that was played in a German television show that is similar to the sorting balls puzzle. In it, we are assumed to move one colored ball after another in a set of colored tubes so that in the end, each ball is in the tube of its color. We are allowed to use one additional (uncolored) tube. We show general properties for solvability and that the problem of minimizing the number of moves is NP-hard, which is done by a reduction from the Feedback Arc Set Problem. Furthermore, we give an implementation of an algorithm to compute such a minimal sequence of moves. The algorithm is based on breadth-first search and accelerated by a lower bound on the number of moves from the current configuration to the final one that is obtained by solving a small instance of the Feedback Arc Set Problem. Our experiments show that instances with 7 colored tubes of height 4 can be solved in a reasonable amount of time and that the number of tubes is much more critical for the running time than the heights of the tubes
Heuristics for Bounded-Suboptimal Search
No full textIn heuristic search, it is well-established that different types of heuristics are suited for optimal heuristic search (OHS) and unbounded suboptimal search (USS). In OHS, the heuristic should minimize the error in estimating the true cost of the shortest path, whereas in USS, it is more beneficial for the heuristic to exhibit a clear gradient toward the goal, regardless of the error. However, no study has specifically investigated which heuristic is most effective for bounded suboptimal search (BSS), and the current standard is to use heuristics designed for OHS. This paper introduces a novel method for creating heuristics tailored to BSS by linearly combining heuristics that were designed for OHS and USS. Through experimental evaluation, the proposed method is compared with those suited for OHS and USS. The results demonstrate that, within certain suboptimality bounds, our new heuristic approach outperforms OHS and USS heuristics for various BSS algorithms
Reevaluation of Large Neighborhood Search for MAPF: Findings and Opportunities
No full textMulti-Agent Path Finding (MAPF) aims to arrange collision-free goal-reaching paths for a group of agents. Anytime MAPF solvers based on large neighborhood search (LNS) have gained prominence recently due to their flexibility and scalability, leading to a surge of methods, especially those leveraging machine learning, to enhance neighborhood selection. However, several pitfalls exist and hinder a comprehensive evaluation of these new methods, which mainly include: 1) Lower than actual or incorrect baseline performance; 2) Lack of a unified evaluation setting and criterion; 3) Lack of a codebase or executable model for supervised learning methods. To address these challenges, we introduce a unified evaluation framework, implement prior methods, and conduct an extensive comparison of prominent methods. Our evaluation reveals that rule-based heuristics serve as strong baselines, while current learning-based methods show no clear advantage on time efficiency or improvement capacity. Our extensive analysis also opens up new research opportunities for improving MAPF-LNS, such as targeting high-delayed agents, applying contextual algorithms, optimizing replan order and neighborhood size, where machine learning can potentially be integrated
Multi-Agent Path Finding for Schedule Constrained Automation (Extended Abstract)
No full textIn modern automation settings, jobs are processed across machines with interdependencies and are subject to limited equipment availability. When transportation between machines is considered, the problem evolves into a complex multi-agent routing task with operational constraints. Existing Multi-Agent Path Finding (MAPF) algorithms address challenges such as robustness and uncertainty, but practical applications involving scheduling constraints often require considerable manual effort for adaptation and heuristic design. In this paper, we introduce MAPF-SC, an extension of MAPF that incorporates scheduling constraints for continuous task flows. We explore the challenges of applying existing techniques to this problem, emphasizing the engineering effort involved in addressing these constraints. Our evaluation investigates the impact of temporal and topological variations on performance, highlighting key factors that influence real-world automation scenarios
Suboptimal Search with Dynamic Distribution of Suboptimality (Extended Abstract)
No full textIn bounded-suboptimal heuristic search, the aim is to find a solution path within a given bound as quickly as possible, which is crucial when computational resources are limited. Recent research has demonstrated Weighted A* variants such as XDP that find bounded suboptimal solutions without needing to perform state re-expansions; they work by shifting where the suboptimality in the search is allowed. However, the suboptimality distribution is fixed before the search begins. This abstract describes Dynamic Suboptimality Weighted A* (DSWA*), an algorithm introduced at AAAI 2025 that allows suboptimality to be dynamically distributed at runtime based on the properties of the search
Leveraging Public Sentiment for Resource Coordination in Disaster Response: A Multiagent Framework
No full textCrises such as natural disasters, misinformation-driven social panic, and economic disruptions place communities under immense stress, demanding rapid and adaptive response strategies. Traditional disaster management has often focused on operational logistics—such as resource allocation and task prioritization—while overlooking how evolving public sentiment and misinformation dynamics can reshape crisis outcomes. In this work, we present MiSC, a multiagent framework that unifies real-time sentiment modeling with multiagent reinforcement learning to contain misinformation and coordinate resources more effectively. By continuously tracking the spread of false narratives and gauging shifts in public sentiment, MiSC adapts counter-messaging campaigns and optimizes deployment decisions in real time. Through simulation-based evaluation, we demonstrate that this synergy between opinion modeling and adaptive decision-making yields significant gains over baseline methods, including faster sentiment recovery, enhanced misinformation control, and improved resource efficiency. By advancing scalable, interoperable AI systems that integrate social signal interpretation with crisis logistics, MiSC underscores the potential of AI-driven resilience for safeguarding communities against multifaceted and unpredictable challenges
Hierarchical Adversarially-Resilient Multi-Agent Reinforcement Learning for Cyber-Physical Systems Security
No full textCyber-Physical Systems are integral to modern critical infrastructure, including manufacturing, energy grids, and autonomous systems, but their increasing interconnectivity exposes them to sophisticated cyber threats. Traditional security measures, such as rule-based intrusion detection and single-agent learning, often fail against adaptive and zero-day attacks. To address this challenge, we propose a Hierarchical Adversarially-Resilient Multi-Agent Reinforcement Learning (HAMARL) framework, integrating adversarial training into a multi-agent security system. HAMARL leverages a hierarchical control structure where local agents manage subsystem security, and a global coordinator optimizes system-wide defense strategies. Additionally, an adversarially-aware learning loop simulates evolving cyber threats, allowing defenders to preemptively adapt to sophisticated attacks. Evaluations on a simulated industrial IoT testbed demonstrate that HAMARL significantly enhances attack detection, reduces response time, and maintains operational continuity compared to traditional MARL approaches. Our findings suggest that hierarchical MARL, combined with adversarial training, presents a promising advancement for securing next-generation CPS