ABSTRACT. The efficient resolution of logistics problems, particularly those aimed at minimizing costs and reducing environmental impact, represents a critical challenge in our globalized world. A prominent example of such problems is the Transshipment Problem, which seeks to determine the most cost-effective paths from sources (e.g., producers) through transshipment points to sinks (e.g., customers). Approaches to addressing this problem range from greedy algorithms, which may rapidly yield locally optimal solutions, to constraint-based methods that, given sufficient resources and computation time, can identify globally optimal solutions. In this study, we propose a hybrid approach that integrates greedy strategies into the solution process of constraint modeling for the Transshipment Problem. This integration aims to expedite the discovery of high-quality initial solutions while preserving the global optimization capabilities inherent in constraint-based search methods. To validate the effectiveness of this new hybrid approach, we conducted an extensive series of experiments, which demonstrate its significant advantages in solving the Transshipment Problem compared to both a conventional constraint model and pure greedy methods.

ABSTRACT. Evidence in arguments may be stated in various forms, including quantitative statements (i.e, numerical relations between entities). This measurable information can be validated against reliable sources like Wikipedia to combat the spread of misinformation. In this paper, we propose a four-step pipeline that combines rule-based techniques with prompting strategies for generative language models in a hybrid fashion. We use regular expressions to identify candidates in claim-premise structures, extract statements using GPT-4o, augment the data with tables from Wikipedia, and validate statements through retrieval-augmented generation (RAG). The pipeline is evaluated on two existing argumentation corpora and the generated dataset is manually annotated to assess the quality of our predictions, showing promising results for extraction and mixed results for validation. Our code and data are publicly available to foster further research in this area.

ABSTRACT. In order for AI systems to provide explanations of their decision-making that are concise and understandable, they need to have the ability of getting rid of irrelevant details and presenting a higher-level view. Answer Set Programming (ASP) is one of the core formalisms of Symbolic AI, based on logic programming with stable model semantics, widely used in various applications. Explaining the solutions (i.e., answer sets) of an answer set program continues to be a widely studied topic, with various systems available. This technical communication reports on a preliminary study for observing the effect of abstraction on the understandability of ASP explanations, by considering the recent abstraction notions which preserve the dependencies as much as possible while abstracting over answer set programs. We describe our experiment designwhich will be our base for further extensions of the study. Our preliminary results show the challenge of capturing the effect of abstraction on understandability, requiring further investigations in this direction.

ABSTRACT. Collaborative environments, particularly in data-sharing and cooperative settings, require robust mechanisms to fairly allocate rewards among participants. In many scenarios, participants contribute valuable data that improve the collective outcome, yet ensuring that each party is fairly compensated for their contribution remains a complex challenge. Classical cooperative game theory develops mechanisms under the assumption of noreplicable resources, which does not fit the scenario of sharing models or data. Hence, it does not address the potential imbalance in reciprocal benefits among pairs of participants, which can lead to strategic exploitation and, consequently, unfair reward allocations. We explicate fairness w.r.t. balanced reciprocity with an axiom and prove the existence of a reward allocation mechanism that fulfills this axiom, as well as other well-known axioms intended to capture incentivization and fairness. By emphasizing mutual fairness, our approach mitigates the risk of exploitation, fosters stable cooperation, and aligns with the principle of fair exchange. We propose a solution that guarantees proportional benefit allocation while maximizing group welfare under these constraints.

ABSTRACT. We summarise our recent research on statistical relational formalisms that allow for incorporating relative frequencies within a possible-worlds semantics (known as Type III probability logic). In a setting of lifted Bayesian networks, we evaluate conditional probability logic as a recent proposal using discrete relative frequency cut-offs and introduce our own formalism, functional lifted Bayesian networks, for integrating continuous relative frequency boundaries. Our main result is a strong asymptotic convergence theorem, which yields uniform convergence in parametric families. This is particularly important in what we consider the most important application of such asymptotic results, estimating the parameters of a statistical relational model from randomly sampled subsets.