Skip to main content

2022 | Book

Multi-Agent-Based Simulation XXII

22nd International Workshop, MABS 2021, Virtual Event, May 3-7, 2021, Revised Selected Papers


About this book

This book constitutes the thoroughly refereed post-conference proceedings of the 21st International Workshop on Multi-Agent-Based Simulation, MABS 2021, held in May 2021 as part of AAMAS 2021. The conference was held virtually due to COVID 19 pandemic.
The 14 revised full papers included in this volume were carefully selected from 23 submissions. The workshop focused on finding efficient solutions to model complex social systems, in such areas as economics, management, organizational and social sciences in general. In all these areas, agent theories, metaphors, models, analysis, experimental designs, empirical studies, and methodological principles, all converge into simulation as a way of achieving explanations and predictions, exploration and testing of hypotheses, better designs and systems and providing decision-support in a wide range of applications.

Table of Contents

Social Simulation for Non-hackers
Computer simulation is a powerful tool for social scientists, but popular platforms require representing the semantics of the model being simulated in computer code, leading to models that are either expensive to construct, inefficient, or inaccurate. We introduce SCAMP (Social Causality using Agents with Multiple Perspectives), a social simulator that uses stigmergy to execute models that are written as concept maps and spreadsheets, without requiring any programming expertise on the part of the modeler. This Repast-based framework has been extensively exercised in the DARPA Ground Truth program to generate realistic social data for analysis by social scientists.
H. Van Dyke Parunak
Using Causal Discovery to Design Agent-Based Models
Designing agent-based models is a difficult task. Some guidelines exist to aid modelers in designing their models, but they generally do not include specific details on how the behavior of agents can be defined. This paper therefore proposes the AbCDe methodology, which uses causal discovery algorithms to specify agent behavior. The methodology combines important expert insights with causal graphs generated by causal discovery algorithms based on real-world data. This causal graph represents the causal structure among agent-related variables, which is then translated to behavioral properties in the agent-based model. To demonstrate the AbCDe methodology, it is applied to a case study in the airport security domain. In this case study, we explore a new concept of operations, using a service lane, to improve the efficiency of the security checkpoint. Results show that the models generated with the AbCDe methodology have a closer resemblance with the validation data than a model defined by experts alone.
Stef Janssen, Alexei Sharpanskykh, S. Sahand Mohammadi Ziabari
Multi-level Adaptation of Distributed Decision-Making Agents in Complex Task Environments
To solve complex tasks, individuals often autonomously organize in teams. Examples of complex tasks include disaster relief rescue operations or project development in consulting. The teams that work on such tasks are adaptive at multiple levels: First, by autonomously choosing the individuals that jointly perform a specific task, the team itself adapts to the complex task at hand, whereby the composition of teams might change over time. We refer to this process as self-organization. Second, the members of a team adapt to the complex task environment by learning. There is, however, a lack of extensive research on multi-level adaptation processes that consider self-organization and individual learning as simultaneous processes in the field of Managerial Science. We introduce an agent-based model based on the NK-framework to study the effects of simultaneous multi-level adaptation on a team’s performance. We implement the multi-level adaptation process by a second-price auction mechanism for self-organization at the team level. Adaptation at the individual level follows an autonomous learning mechanism. Our preliminary results suggest that, depending on the task’s complexity, different configurations of individual and collective adaptation can be associated with higher overall task performance. Low complex tasks favour high individual and collective adaptation, while moderate individual and collective adaptation is associated with better performance in case of moderately complex tasks. For highly complex tasks, the results suggest that collective adaptation is harmful to performance.
Darío Blanco-Fernández, Stephan Leitner, Alexandra Rausch
Fast Agent-Based Simulation Framework with Applications to Reinforcement Learning and the Study of Trading Latency Effects
We introduce a new software toolbox for agent-based simulation. Facilitating rapid prototyping by offering a user-friendly Python API, its core rests on an efficient C++ implementation to support simulation of large-scale multi-agent systems. Our software environment benefits from a versatile message-driven architecture. Originally developed to support research on financial markets, it offers the flexibility to simulate a wide-range of different (easily customisable) market rules and to study the effect of auxiliary factors, such as delays, on the market dynamics. As a simple illustration, we employ our toolbox to investigate the role of the order processing delay in normal trading and for the scenario of a significant price change.
Owing to its general architecture, our toolbox can also be employed as a generic multi-agent system simulator. We provide an example of such a non-financial application by simulating a mechanism for the coordination of no-regret learning agents in a multi-agent network routing scenario previously proposed in the literature.
Peter Belcak, Jan-Peter Calliess, Stefan Zohren
On the Same Wavelengths: Emergence of Multiple Synchronies Among Multiple Agents
People spontaneously synchronize their mental states and behavioral actions when they interact. This paper models general mechanisms that can lead to the emergence of interpersonal synchrony by multiple agents with internal cognitive and affective states. In our simulations, one agent was exposed to a repeated stimulus and the other agent started to synchronize consecutively its movements, affects, conscious emotions and verbal actions with the exposed agent. The behavior displayed by the agents was consistent with theory and empirical evidence from the psychological and neuroscience literature. These results shed new light on the emergence of interpersonal synchrony in a wide variety of settings, from close relationships to psychotherapy. Moreover, the present work could provide a basis for future development of socially responsive virtual agents.
Sophie C. F. Hendrikse, Jan Treur, Tom F. Wilderjans, Suzanne Dikker, Sander L. Koole
Multi-agent Simulation for AI Behaviour Discovery in Operations Research
We describe ACE0, a lightweight platform for evaluating the suitability and viability of AI methods for behaviour discovery in multi-agent simulations. Specifically, ACE0 was designed to explore AI methods for multi-agent simulations used in operations research studies related to new technologies such as autonomous aircraft. Simulation environments used in production are often high-fidelity, complex, require significant domain knowledge and as a result have high R&D costs. Minimal and lightweight simulation environments can help researchers and engineers evaluate the viability of new AI technologies for behaviour discovery in a more agile and potentially cost effective manner. In this paper we describe the motivation for the development of ACE0. We provide a technical overview of the system architecture, describe a case study of behaviour discovery in the aerospace domain, and provide a qualitative evaluation of the system. The evaluation includes a brief description of collaborative research projects with academic partners, exploring different AI behaviour discovery methods.
Michael Papasimeon, Lyndon Benke
Using Agent-Based Modelling to Understand Advantageous Behaviours Against COVID-19 Transmission in the Built Environment
The global Covid-19 pandemic has raised many questions about how we occupy and move in the built environment. Interior environments have been increasingly discussed in numerous studies highlighting how interior spaces play a key role in the spread of pandemics. One societal challenge is to find short-term strategies to reopen indoor venues. Most current approaches focus on an individual’s behavior (maintaining social distance, wearing face masks, and washing their hands) and government policies (confinement, curfew, quarantine, etc.). However, few studies have been conducted to understand a building’s interior where most transmission takes place. How will the utilization of existing interior spaces be improved above and beyond universally applied criteria, while minimizing the risk of disease transmission? This article presents an agent-based model that examines disease transmission risks in various “interior types” in combination with user behaviors and their mobility, as well as three types of transmission vectors (direct, airborne and via surfaces). The model also integrates numerous policy interventions, including wearing masks, hand washing, and the possibility of easily modifying the organization of spaces. Different studies at various scales were conducted both on the University of Guadalajara (UdeG) campus as well as at the MIT Media Lab to illustrate the application of this model.
Arnaud Grignard, Tri Nguyen-Huu, Patrick Taillandier, Luis Alonso, Nicolas Ayoub, Markus Elkatsha, Gamaliel Palomo, Monica Gomez, Mario Siller, Mayra Gamboa, Carlos Ivan Moreno, Kent Larson
Quantifying the Effects of Norms on COVID-19 Cases Using an Agent-Based Simulation
Modelling social phenomena in large-scale agent-based simulations has long been a challenge due to the computational cost of incorporating agents whose behaviors are determined by reasoning about their internal attitudes and external factors. However, COVID-19 has brought the urgency of doing this to the fore, as, in the absence of viable pharmaceutical interventions, the progression of the pandemic has primarily been driven by behaviors and behavioral interventions. In this paper, we address this problem by developing a large-scale data-driven agent-based simulation model where individual agents reason about their beliefs, objectives, trust in government, and the norms imposed by the government. These internal and external attitudes are based on actual data concerning daily activities of individuals, their political orientation, and norms being enforced in the US state of Virginia. Our model is calibrated using mobility and COVID-19 case data. We show the utility of our model by quantifying the benefits of the various behavioral interventions through counterfactual runs of our calibrated simulation.
Jan de Mooij, Davide Dell’Anna, Parantapa Bhattacharya, Mehdi Dastani, Brian Logan, Samarth Swarup
MAS Network: Surrogate Neural Network for Multi-agent Simulation
Multi-agent simulation (MAS) plays an important role in analyzing our societies because it can model complexity in societies and assimilate a variety of social data. However, the execution of MAS is computationally expensive. When running numerous executions to determine optimal policy, it is crucial to develop a more computationally efficient mathematical model that is able to sufficiently substitute for the original simulation. In this paper, we propose a machine learning framework for developing neural network models, called \( {MAS\ network}\), that can substitute for MAS. Furthermore, we propose an effective feature representation of agent parameters and a systematic dataset design for learning. We confirmed that the MAS network replicated the system dynamics of the simulation and that the MAS network accurately learned the sensitivity of output and input relation even at unknown parameter points.
Hiroaki Yamada, Masataka Shirahashi, Naoyuki Kamiyama, Yumeka Nakajima
Real-Time Inference of Urban Metrics Applying Machine Learning to an Agent-Based Model Coupling Mobility Mode and Housing Choice
This paper describes the latest advancements in the Housing and Mobility Mode Choice module of CityScope, a data-driven tangible platform developed by MIT City Science (CS) to facilitate more participatory decision-making processes. The ultimate objective of the Module is to easily predict people’s reactions to potential urban disruptions and policies by previously characterizing their behavioural patterns. The main phase of this work consisted of a generic Agent-Based Model coupling mobility mode and housing choice, which was calibrated and validated for the Metropolitan Boston Area and Kendall Square in Cambridge, US. However, the integration of such model onto the CityScope platform resulted challenging, due to the complexity of the represented dynamics. The present paper addresses this problem making use of machine learning to train a surrogate model that will enable the real-time visualization and analysis of the suggested actions. The real-time nature of the obtained urban metrics will allow to append this Module to the current easily-understandable CityScope feedback system, bringing different stakeholders together to consensually shape the most favourable urban scenario. This Module represents the first step towards the development of a dynamic incentive system where CS seeks to promote urban characteristics such as equality, diversity, walkability, and efficiency.
Mireia Yurrita, Arnaud Grignard, Luis Alonso, Kent Larson
Changing Perspectives: Adaptable Interpretations of Norms for Agents
For agent-based social simulations to be a powerful tool for policy makers and other decision makers in a given context (e.g. the current COVID-19 pandemic), they need to be socially realistic and thus, appropriately represent complex social concepts, such as social rules. In this paper, we focus on norms. Norms describe ‘normal’ behavior and aim at assuring the interests and values of groups or the society as a whole. People react differently to norms, and focus only on the parts that are relevant for them. Furthermore, norms are not only restrictions on behavior, but also trigger new behavior. Seeing a norm only as a restriction on certain behavior misses important aspects and leads to simulations that can be very misleading. Different perspectives need to be incorporated into the simulation to capture the variety of ways different stakeholders react to a norm and how this affects their interaction. We therefore present an approach to include these different perspectives on norms, and their consequences for different people and groups in decision support simulations. A perspective is specified by their goals, actions, effects of those actions, priorities in values, and social affordances. Through modeling perspectives we enable policy makers and other decision makers (the users) to be active in the modeling process and to tailor the simulation to their specific needs, by representing norms as modifiable objects, and providing textual and graphical representations of norms. This provides them with differentiated insights meaningful for the decisions they are faced with. We indicate the requirements for both the simulation platform as well as the agents that follow from our approach. Early explorations of our social simulation are showing the necessity of our approach.
Christian Kammler, Frank Dignum, Nanda Wijermans, Helena Lindgren
Exploration of Model Coupling Strategies in a Hybrid Agent-Based Traffic Simulation
Traffic simulation is a tool used by urban planners to assess the impact of new urban designs and public policies on mobility. Over the years, numerous traffic models have been proposed, each model offering different levels of details and performances. Multi-level model coupling is an interesting approach to combine the advantages of complementary representations while limiting their drawbacks. In this paper, we design and evaluate the performances of hybrid traffic models combining a microscopic model (IDM) with a mesoscopic model (event-driven and queue-based). The results show that microscopic models have more diversity in terms of behaviors but reduce the vehicle average speed and mesoscopic models are more efficient in terms of computational time but display a higher vehicle speed. Their hybridization then enables to find a balance between scalability and the variety of the observed behaviors.
Jean-François Erdelyi, Frédéric Amblard, Benoit Gaudou, Elsy Kaddoum, Nicolas Verstaevel
The Recruitment Game: An Agent-Based Simulation
While the studies on terrorism and radicalization are foregoing, the socialization aspect of recruitment for terrorist organizations has stayed under-explored. In this paper, we develop an agent-based model simulating the socialization process of recruitment for terrorist organizations. In conceptualizing the socialization process, we implement an asymmetric game-theoretical model, with the two players of recruiter and target. The players have predominant strategies in which they differ based on their kinds. Our results show that initial ratios of different kinds in the population such as denouncer and vulnerable, in the simulation environment, have significant effects on the population of recruiters.
Siavash Farahbakhsh, Mario Paolucci
Fishing Together?
Exploring the Murky Waters of Sociality
Collective action research of natural resource use aims to understand why and when collective overuse arises. Agent-based simulations and behavioural experiments are part of the toolkit for this quest. In most agent-based simulation models however, individual and collective decision-making are discerned, but the crucial transition between these two stances is understudied. In this paper we formalise computational agents able to think and act from an individual, social, or collective stance using a combination of empirical findings and theoretical models on togetherness. To this end, we use a conceptual agent framework to adapt and extend an existing agent-based model designed to advance the understanding of group processes for sustainable governance of dynamic common pool resource environments. The findings of the paper are mainly a conceptual model and future research will further develop the framework as well as the agent-based model for further understanding of the processes involved.
Nanda Wijermans, Harko Verhagen
Multi-Agent-Based Simulation XXII
Dr. Koen H. Van Dam
Nicolas Verstaevel
Copyright Year
Electronic ISBN
Print ISBN

Premium Partner