Autonomous Agents and Multi-Agent Systems最新文献

{"title":"Team-wise effective communication in multi-agent reinforcement learning","authors":"Ming Yang,&nbsp;Kaiyan Zhao,&nbsp;Yiming Wang,&nbsp;Renzhi Dong,&nbsp;Yali Du,&nbsp;Furui Liu,&nbsp;Mingliang Zhou,&nbsp;Leong Hou U","doi":"10.1007/s10458-024-09665-6","DOIUrl":"10.1007/s10458-024-09665-6","url":null,"abstract":"<div><p>Effective communication is crucial for the success of multi-agent systems, as it promotes collaboration for attaining joint objectives and enhances competitive efforts towards individual goals. In the context of multi-agent reinforcement learning, determining “whom”, “how” and “what” to communicate are crucial factors for developing effective policies. Therefore, we propose TeamComm, a novel framework for multi-agent communication reinforcement learning. First, it introduces a dynamic team reasoning policy, allowing agents to dynamically form teams and adapt their communication partners based on task requirements and environment states in cooperative or competitive scenarios. Second, TeamComm utilizes heterogeneous communication channels consisting of intra- and inter-team to achieve diverse information flow. Lastly, TeamComm leverages the information bottleneck principle to optimize communication content, guiding agents to convey relevant and valuable information. Through experimental evaluations on three popular environments with seven different scenarios, we empirically demonstrate the superior performance of TeamComm compared to existing methods.</p></div>","PeriodicalId":55586,"journal":{"name":"Autonomous Agents and Multi-Agent Systems","volume":"38 2","pages":""},"PeriodicalIF":2.0,"publicationDate":"2024-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141741489","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"When is it acceptable to break the rules? Knowledge representation of moral judgements based on empirical data","authors":"Edmond Awad,&nbsp;Sydney Levine,&nbsp;Andrea Loreggia,&nbsp;Nicholas Mattei,&nbsp;Iyad Rahwan,&nbsp;Francesca Rossi,&nbsp;Kartik Talamadupula,&nbsp;Joshua Tenenbaum,&nbsp;Max Kleiman-Weiner","doi":"10.1007/s10458-024-09667-4","DOIUrl":"10.1007/s10458-024-09667-4","url":null,"abstract":"<div><p>Constraining the actions of AI systems is one promising way to ensure that these systems behave in a way that is morally acceptable to humans. But constraints alone come with drawbacks as in many AI systems, they are not flexible. If these constraints are too rigid, they can preclude actions that are actually acceptable in certain, contextual situations. Humans, on the other hand, can often decide when a simple and seemingly inflexible rule should actually be overridden based on the context. In this paper, we empirically investigate the way humans make these contextual moral judgements, with the goal of building AI systems that understand when to follow and when to override constraints. We propose a novel and general preference-based graphical model that captures a modification of standard <i>dual process</i> theories of moral judgment. We then detail the design, implementation, and results of a study of human participants who judge whether it is acceptable to break a well-established rule: <i>no cutting in line</i>. We then develop an instance of our model and compare its performance to that of standard machine learning approaches on the task of predicting the behavior of human participants in the study, showing that our preference-based approach more accurately captures the judgments of human decision-makers. It also provides a flexible method to model the relationship between variables for moral decision-making tasks that can be generalized to other settings.</p></div>","PeriodicalId":55586,"journal":{"name":"Autonomous Agents and Multi-Agent Systems","volume":"38 2","pages":""},"PeriodicalIF":2.0,"publicationDate":"2024-07-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10458-024-09667-4.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141612786","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Emergent cooperation from mutual acknowledgment exchange in multi-agent reinforcement learning","authors":"Thomy Phan,&nbsp;Felix Sommer,&nbsp;Fabian Ritz,&nbsp;Philipp Altmann,&nbsp;Jonas Nüßlein,&nbsp;Michael Kölle,&nbsp;Lenz Belzner,&nbsp;Claudia Linnhoff-Popien","doi":"10.1007/s10458-024-09666-5","DOIUrl":"10.1007/s10458-024-09666-5","url":null,"abstract":"<div><p><i>Peer incentivization (PI)</i> is a recent approach where all agents learn to reward or penalize each other in a distributed fashion, which often leads to emergent cooperation. Current PI mechanisms implicitly assume a flawless communication channel in order to exchange rewards. These rewards are directly incorporated into the learning process without any chance to respond with feedback. Furthermore, most PI approaches rely on global information, which limits scalability and applicability to real-world scenarios where only local information is accessible. In this paper, we propose <i>Mutual Acknowledgment Token Exchange (MATE)</i>, a PI approach defined by a two-phase communication protocol to exchange acknowledgment tokens as incentives to shape individual rewards mutually. All agents condition their token transmissions on the locally estimated quality of their own situations based on environmental rewards and received tokens. MATE is completely decentralized and only requires local communication and information. We evaluate MATE in three social dilemma domains. Our results show that MATE is able to achieve and maintain significantly higher levels of cooperation than previous PI approaches. In addition, we evaluate the robustness of MATE in more realistic scenarios, where agents can deviate from the protocol and communication failures can occur. We also evaluate the sensitivity of MATE w.r.t. the choice of token values.</p></div>","PeriodicalId":55586,"journal":{"name":"Autonomous Agents and Multi-Agent Systems","volume":"38 2","pages":""},"PeriodicalIF":2.0,"publicationDate":"2024-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10458-024-09666-5.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141588186","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An agent-based persuasion model using emotion-driven concession and multi-objective optimization","authors":"Zhenwu Wang,&nbsp;Jiayin Shen,&nbsp;Xiaosong Tang,&nbsp;Mengjie Han,&nbsp;Zhenhua Feng,&nbsp;Jinghua Wu","doi":"10.1007/s10458-024-09664-7","DOIUrl":"10.1007/s10458-024-09664-7","url":null,"abstract":"<div><p>Multi-attribute negotiation is essentially a multi-objective optimization (MOO) problem, where models of agent-based emotional persuasion (EP) can exhibit characteristics of anthropomorphism. This paper proposes a novel EP model by fusing the strategy of emotion-driven concession with the method of multi-objective optimization (EDC-MOO). Firstly, a comprehensive emotion model is designed to enhance the authenticity of the emotion. A novel concession strategy is then proposed to enable the concession to be dynamically tuned by the emotions of the agents. Finally, a new EP model is constructed by integrating emotion, historical transaction, persuasion behavior, and concession strategy under the framework of MOO. Comprehensive experiments on bilateral negotiation are conducted to illustrate and validate the effectiveness of EDC-MOO. These include an analysis of negotiations under five distinct persuasion styles, a comparison of EDC-MOO with a non-emotion-based MOO negotiation model and classic trade-off strategies, negotiations between emotion-driven and non-emotion-driven agents, and negotiations involving human participants. A detailed analysis of parameter sensitivity is also discussed. Experimental results show that the proposed EDC-MOO model can enhance the diversity of the negotiation process and the anthropomorphism of the bilateral agents, thereby improving the social welfare of both parties.</p></div>","PeriodicalId":55586,"journal":{"name":"Autonomous Agents and Multi-Agent Systems","volume":"38 2","pages":""},"PeriodicalIF":2.0,"publicationDate":"2024-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10458-024-09664-7.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141571666","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"From large language models to small logic programs: building global explanations from disagreeing local post-hoc explainers","authors":"Andrea Agiollo,&nbsp;Luciano Cavalcante Siebert,&nbsp;Pradeep K. Murukannaiah,&nbsp;Andrea Omicini","doi":"10.1007/s10458-024-09663-8","DOIUrl":"10.1007/s10458-024-09663-8","url":null,"abstract":"<div><p>The expressive power and effectiveness of <i>large language models</i> (LLMs) is going to increasingly push intelligent agents towards sub-symbolic models for natural language processing (NLP) tasks in human–agent interaction. However, LLMs are characterised by a performance vs. transparency trade-off that hinders their applicability to such sensitive scenarios. This is the main reason behind many approaches focusing on <i>local</i> post-hoc explanations, recently proposed by the XAI community in the NLP realm. However, to the best of our knowledge, a thorough comparison among available explainability techniques is currently missing, as well as approaches for constructing <i>global</i> post-hoc explanations leveraging the local information. This is why we propose a novel framework for comparing state-of-the-art local post-hoc explanation mechanisms and for extracting logic programs surrogating LLMs. Our experiments—over a wide variety of text classification tasks—show how most local post-hoc explainers are loosely correlated, highlighting substantial discrepancies in their results. By relying on the proposed novel framework, we also show how it is possible to extract faithful and efficient global explanations for the original LLM over multiple tasks, enabling explainable and resource-friendly AI techniques.</p></div>","PeriodicalId":55586,"journal":{"name":"Autonomous Agents and Multi-Agent Systems","volume":"38 2","pages":""},"PeriodicalIF":2.0,"publicationDate":"2024-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10458-024-09663-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141571665","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Navigating in a space of game views","authors":"Michael P. Wellman,&nbsp;Katherine Mayo","doi":"10.1007/s10458-024-09660-x","DOIUrl":"10.1007/s10458-024-09660-x","url":null,"abstract":"<div><p>Game-theoretic modeling entails selecting the particular elements of a complex strategic situation deemed most salient for strategic analysis. Recognizing that any game model is one of many possible views of the situation, we term this a <i>game view</i>, and propose that sophisticated game reasoning would naturally consider multiple views. We introduce a conceptual framework, <i>game view navigation</i>, for game-theoretic reasoning through a process of constructing and analyzing a series of game views. The approach is illustrated using a variety of existing methods, which can be cast in terms of navigation patterns within this framework. By formally defining these as well as recently introduced ideas as navigating in a space of game views, we recognize common themes and opportunities for generalization. Game view navigation thus provides a unifying perspective that sheds light on connections between disparate reasoning methods, and defines a design space for creation of new techniques. We further apply the framework by defining and exploring new techniques based on modulating player aggregation in equilibrium search.</p></div>","PeriodicalId":55586,"journal":{"name":"Autonomous Agents and Multi-Agent Systems","volume":"38 2","pages":""},"PeriodicalIF":2.0,"publicationDate":"2024-07-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141571668","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Beyond the echo chamber: modelling open-mindedness in citizens’ assemblies","authors":"Jake Barrett,&nbsp;Kobi Gal,&nbsp;Loizos Michael,&nbsp;Dan Vilenchik","doi":"10.1007/s10458-024-09655-8","DOIUrl":"10.1007/s10458-024-09655-8","url":null,"abstract":"<div><p>A Citizens’ assembly (CA) is a democratic innovation tool where a randomly selected group of citizens deliberate a topic over multiple rounds to generate, and then vote upon, policy recommendations. Despite growing popularity, little work exists on understanding how CA inputs, such as the expert selection process and the mixing method used for discussion groups, affect results. In this work, we model CA deliberation and opinion change as a multi-agent systems problem. We introduce and formalise a set of criteria for evaluating successful CAs using insight from previous CA trials and theoretical results. Although real-world trials meet these criteria, we show that finding a model that does so is non-trivial; through simulations and theoretical arguments, we show that established opinion change models fail at least one of these criteria. We therefore propose an augmented opinion change model with a latent ‘open-mindedness’ variable, which sufficiently captures people’s propensity to change opinion. We show that data from the CA of Scotland indicates a latent variable both exists and resembles the concept of open-mindedness in the literature. We calibrate parameters against real CA data, demonstrating our model’s ecological validity, before running simulations across a range of realistic global parameters, with each simulation satisfying our criteria. Specifically, simulations meet criteria regardless of expert selection, expert ordering, participant extremism, and sub-optimal participant grouping, which has ramifications for optimised algorithmic approaches in the computational CA space.</p></div>","PeriodicalId":55586,"journal":{"name":"Autonomous Agents and Multi-Agent Systems","volume":"38 2","pages":""},"PeriodicalIF":2.0,"publicationDate":"2024-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10458-024-09655-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141520821","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"On preferences and reward policies over rankings","authors":"Marco Faella,&nbsp;Luigi Sauro","doi":"10.1007/s10458-024-09656-7","DOIUrl":"10.1007/s10458-024-09656-7","url":null,"abstract":"<div><p>We study the rational preferences of agents participating in a mechanism whose outcome is a ranking (i.e., a weak order) among participants. We propose a set of self-interest axioms corresponding to different ways for participants to compare rankings. These axioms vary from minimal conditions that most participants can be expected to agree on, to more demanding requirements that apply to specific scenarios. Then, we analyze the theories that can be obtained by combining the previous axioms and characterize their mutual relationships, revealing a rich hierarchical structure. After this broad investigation on preferences over rankings, we consider the case where the mechanism can distribute a fixed monetary reward to the participants in a fair way (that is, depending only on the anonymized output ranking). We show that such mechanisms can induce specific classes of preferences by suitably choosing the assigned rewards, even in the absence of tie breaking.</p></div>","PeriodicalId":55586,"journal":{"name":"Autonomous Agents and Multi-Agent Systems","volume":"38 2","pages":""},"PeriodicalIF":2.0,"publicationDate":"2024-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10458-024-09656-7.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141520822","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Parameterized complexity of candidate nomination for elections based on positional scoring rules","authors":"Ildikó Schlotter,&nbsp;Katarína Cechlárová,&nbsp;Diana Trellová","doi":"10.1007/s10458-024-09658-5","DOIUrl":"10.1007/s10458-024-09658-5","url":null,"abstract":"<div><p>Consider elections where the set of candidates is partitioned into parties, and each party must nominate exactly one candidate. The P<span>ossible</span> P<span>resident</span> problem asks whether some candidate of a given party can become the unique winner of the election for some nominations from other parties. We perform a multivariate computational complexity analysis of P<span>ossible</span> P<span>resident</span> for several classes of elections based on positional scoring rules. We consider the following parameters: the size of the largest party, the number of parties, the number of voters and the number of voter types. We provide a complete computational map of P<span>ossible</span> P<span>resident</span> in the sense that for each choice of the four possible parameters as (i) constant, (ii) parameter, or (iii) unbounded, we classify the computational complexity of the resulting problem as either polynomial-time solvable or <span>NP</span>-complete, and for parameterized versions as either fixed-parameter tractable or <span>W</span>[1]-hard with respect to the parameters considered.</p></div>","PeriodicalId":55586,"journal":{"name":"Autonomous Agents and Multi-Agent Systems","volume":"38 2","pages":""},"PeriodicalIF":2.0,"publicationDate":"2024-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10458-024-09658-5.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141506306","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A comprehensive analysis of agent factorization and learning algorithms in multiagent systems","authors":"Andreas Kallinteris,&nbsp;Stavros Orfanoudakis,&nbsp;Georgios Chalkiadakis","doi":"10.1007/s10458-024-09662-9","DOIUrl":"10.1007/s10458-024-09662-9","url":null,"abstract":"<div><p>In multiagent systems, agent factorization denotes the process of segmenting the state-action space of the environment into distinct components, each corresponding to an individual agent, and subsequently determining the interactions among these agents. Effective agent factorization significantly influences the system performance of real-world industrial applications. In this work, we try to assess the performance impact of agent factorization when using different learning algorithms in multiagent coordination settings; and thus discover the source of performance quality of the multiagent solution derived by combining different factorizations with different learning algorithms. To this end, we evaluate twelve different agent factorization instances—or <i>agent definitions</i>—in the warehouse traffic management domain, comparing the training performance of (primarily) three learning algorithms suitable for learning coordinated multiagent policies: the Evolutionary Strategies (<i>ES</i>), the Canonical Evolutionary Strategies (<i>CES</i>), and a genetic algorithm (<i>CCEA</i>) previously used in a similar setting. Our results demonstrate that the performance of different learning algorithms is affected in different ways by alternative agent definitions. Given this, we can conclude that many important multiagent coordination problems can eventually be solved more efficiently by a suitable agent factorization combined with an appropriate choice of a learning algorithm. Moreover, our work shows that ES and CES are effective learning algorithms for the warehouse traffic management domain, while, interestingly, celebrated policy gradient methods do not fare well in this complex real-world problem setting. As such, our work offers insights into the intrinsic properties of the learning algorithms that make them well-suited for this problem domain. More broadly, our work demonstrates the need to identify appropriate agent definitions-multiagent learning algorithm pairings in order to solve specific complex problems effectively, and provides insights into the general characteristics that such pairings must possess to address broad classes of multiagent learning and coordination problems.</p></div>","PeriodicalId":55586,"journal":{"name":"Autonomous Agents and Multi-Agent Systems","volume":"38 2","pages":""},"PeriodicalIF":2.0,"publicationDate":"2024-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141506307","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}