Advances in Practical Applications of Survivable Agents and Multi-Agent Systems: The PAAMS Collection

Simulating and predicting planetary-scale techno-social systems poses heavy computational and modeling challenges. The DARPA SocialSim program set the challenge to model the evolution of GitHub, a large collaborative software-development ecosystem, using massive multi-agent simulations. We describe our best performing models and our agent-based simulation framework, which we are currently extending to allow simulating other planetary-scale techno-social systems. The challenge problem measured participant’s ability, given 30 months of meta-data on user activity on GitHub, to predict the next months’ activity as measured by a broad range of metrics applied to ground truth, using agent-based simulation. The challenge required scaling to a simulation of roughly 3 million agents producing a combined 30 million actions, acting on 6 million repositories with commodity hardware. It was also important to use the data optimally to predict the agent’s next moves. We describe the agent framework and the data analysis employed by one of the winning teams in the challenge. Six different agent models were tested based on a variety of machine learning and statistical methods. While no single method proved the most accurate on every metric, the broadly most successful sampled from a stationary probability distribution of actions and repositories for each agent. Two reasons for the success of these agents were their use of a distinct characterization of each agent, and that GitHub users change their behavior relatively slowly.

Health support programs play a vital role in public health and prevention strategies at local and national levels, for issues such as smoking cessation, physical rehabilitation, nutrition, or to regain mobility. A key success factor in these topics is related to the appropriate use of behavior change techniques, as well as tailored recommendations for users/patients, adapted to their goals and the continuous monitoring of their progress. Social networks interactions and the use of multi-agent technologies can further improve the effectiveness of these programs, especially through personalization and profiling of users and patients. In this paper we propose an agent-based model for supporting behavior change in eHealth programs. Moreover, we identify the main challenges in this area, especially regarding profile and domain modeling profiles for healthcare behavioral programs, where the definition of goals, expectations and argumentation play a key role in the success of a intervention.

Features of blockchain technology (BCT) such as decentralisation, trust, fault tolerance, and accountability, are of paramount importance for multi-agent systems (MAS). In this paper we argue that a principled approach to MAS-BCT integration cannot overlook the foundational character of agency—that is, autonomy. Accordingly, we present a custom BCT implementation where autonomy is placed in smart contracts (SC) interpreted as software agents. We show how agency can enhance SC expressiveness with autonomy, situatedness, sociality, and intelligence, and highlight the limitations of state-of-art BCT in supporting MAS design and implementation.

Interaction networks, either being implicitly or explicitly specified between the agents, play a crucial role in all multi-agent systems. These structures define and limit the ways the agents should interact with their peers, and hence help to manage the coordination problems in large-scale systems. It is widely accepted that the structure of an interaction network plays a significant role in the performance of the systems. In this article, we use the interaction network of the initial agent population to construct a holonic multi-agent system. Being based on agent interaction networks, the performance of the resulting holonic multi-agent system highly depends on the structure of the underlying agent network. Here, we study this dependency in more details. The study is carried out by applying the holonification algorithm on various network topologies and assessing the constructed holonic structure in a task environment.

In this work, we want to show that the introduction of categories can strongly improve the performance of recommendation, within the new digitally infrastructured societies. We state that, inside these highly dynamic contexts, in which more and more people are connected to each other but a substantial part of the communication happens between strangers, it is fundamental to restructure the concept of recommendation. We strongly believe that a good solution for many situations would be to combine inferential processes with recommendations, i.e. focusing on recommending categories of agents rather than specific individuals. Specifically, in this work we prove that category’s recommendations are more robust to untrustworthy recommenders than individual recommendation. We tested our idea by the mean of a multi-agent social simulation. The results we obtained are in agreement with our hypotheses and can be of important interest for the development of this sector.

Intelligent Tutoring Systems (ITS) are technological resources widely used in teaching-learning processes, and their studies are directed, mainly, at distance learning. In this sense, the purpose of this work is the redesign of a Student Model agent, in the context of an ITS applied to the teaching of Natural Deduction in Propositional Logic (NDPL) for computing. It is expected that the agent will be able to identify and represent the students’ knowledge states. In the modeling stage, we present the details of the knowledge representation, as well as about the inference mechanism, based on Bayesian networks. Regarding the results, students are satisfied with the Heráclito environment and that the agent achieves its main objective, evidencing the possibility of implementing personalized teaching strategies based on individual characteristics and knowledge from the students.

An Agent-Based Model (ABM) accounting for tax-morale and loss-aversion was implemented over different network systems with social interactions at the local level to study the phenomenon of tax evasion. This ABM is an innovative model which integrates endogenous characteristics of heterogeneous agents and proposes a more relaxed assumption on the information exchanged between agents as compared to previous social models. The current study gives an insight on the possibility that choosing specific network structures may yield to more realistic outcomes. Moreover, this ABM manages to replicate both individual and aggregate results from previous experimental and computational models of tax evasion. A clearcut novelty might be the non-linear channel through which the network centrality enhances a positive effect on the aggregated level of tax compliance. There is a large area of action for public policy makers to further research the presented results about how audit rates, fines and tax morale non-linearly increase income disclosure, whereas tax rates have a non-linear negative impact on tax compliance.

Human Activity Recognition (HAR) using wearable sensors is becoming more practical in the field of security and health care monitoring. Deep and Machine learning techniques have been widely used in this area. Since smartphones and their applications are the parts of daily life, they can be very helpful in data gathering and online learning in HAR problems. In this paper, a new hierarchical neural network structure with the hierarchical learning method is presented and applied to the HAR. The proposed model is a deep learner neural network without using heavy computations that CNN-based deep learners usually suffer from. This makes the suggested model suitable for being embedded in agent and multi-agent based solutions and online learning, especially when they are implemented in small devices such as smartphones. In addition the hidden layer of the first section of the proposed model benefits automatic nonlinear feature extraction. The extracted features are proper for classifications. Handling the dimension of data is one of the challenges in HAR problem. In our model, data dimension reduction is automatically performed in the hidden layers of the different network sections. According to the empirical results, our proposed model yields better performance on the Opportunity data sets, compared to the similar ML algorithms.

The design of Multi-Attribute Double-Sided Auctions (MADSA) is an important problem being examined in a variety of domains. Despite significant efforts, an ideal compromise between expressiveness of preference representation and the tractability of MADSA mechanisms is still subject to much debate. In this paper, we propose a MADSA mechanism whereby bids are placed in the form of Generalised Additively Independent-Decomposable (GAI-D) utility functions. We show that by applying a set of constraints on the composition of these functions a relaxation of the Kalai bargaining solution becomes tractable for large double-sided markets. Experimental results show that the proposed mechanism provides efficient results when compared to the well known k-priced greedy market mechanism.

We present the application of multiagent reinforcement learning to the problem of traffic light signal control to decrease travel time. We model roads as a collection of agents for each signalized junction. Agents learn to set phases that jointly maximize a reward function that encourages short vehicle queuing delays and queue lengths at all junctions. The first approach that we tested exploits the fact that the reward function can be splitted into contributions per agent. Junctions are modeled as vertices in a coordination graph and the joint action is found with the variable elimination algorithm. The second method exploits the principle of locality to compute the best action for an agent as its best response for a two player game with each member of its neighborhood. We apply the learning methods to a simulated network of 6 intersections, using data from the Transit Department of Bogotá, Colombia. These methods obtained significant reductions in queuing delay with respect to the fixed time control, and in general achieve shorter travel times across the network than some other reinforcement learning based methods found in the literature.

With the increasing adoption of Internet of Things (IoT) technologies, the number of agents offering equivalent capabilities is increasing more and more. The services of these capability equivalent agents may have different Quality of Service (QoS) levels. Therefore, the selection of the most appropriate services that best match some given requirements becomes a challenging issue in the HARMS (Human, Agent, Robot, Machine, Sensor) multi-agent systems. In this paper, a Social Group optimization-based QoS-aware agents services Composition Algorithm (SG-QCA) is proposed to enable HARMS interaction layer with the capability of composing agents services in large-scale IoT services environments. The simulation results show that for both randomly generated and real datasets, the proposed approach is scalable and achieves a near-to-optimal composition in a reduced composition time compared to other services composition approaches.

GPU-based parallelization of agent-based modeling (ABM) has been highlighted for the last decade to address its computational needs for scalable and long-run simulations in practical use. From the software productivity viewpoint, model designers would prefer general ABM frameworks for GPU parallelization. However, having transited from single-node or cluster-computing platforms to GPUs, most general ABM frameworks maintain their APIs at the script level, delegate only a limited number of agent functions to GPUs, and copy agent data between host and device memory for each function call, which cannot ease agent description nor maximize GPU parallelism. To respond to these problems, we have developed the MASS (Multi-Agent Spatial Simulation) CUDA library that allows users to describe all simulation models in CUDA C++, to automate entire model parallelization at GPU, and to minimize host-to-device memory transfer. However, our straightforward implementation did not improve the parallel performance. Focusing on the data-parallel computation with GPU, we examined MASS overheads in GPU memory usage and developed optimization techniques that reduce kernel context switches, optimize kernel configuration, use constant memory, and reduce overheads incurred by agent population, migration, and termination. These techniques improved Heat2D and SugarScape’s execution performance, respectively 3.9 times and 5.8 times faster than the corresponding C++ sequential programs. This paper gives details of our GPU parallelization techniques for multi-agent simulation and demonstrates the MASS CUDAs performance improvements.

In this paper we develop a method for planning and coordinating a multi-agent team deployment to periodically gather information on demand. A static operation center (OC) periodically requests information from changing goal locations. The objective is to gather data in the goals and to deliver it to the OC, balancing the refreshing time and the total number of information packages. The system automatically splits the team in two roles: workers to gather data, or collectors to retransmit the data to the OC. The proposed three step method: (1) finds out the best area partition for the workers; (2) obtains the best balance between workers and collectors, and with whom the workers must to communicate, a collector or the OC; (3) computes the best tour for the workers to visit the goals and deliver them to the OC or to a collector in movement. The method is tested in simulations in different scenarios, providing the best area partition algorithm and the best balance between collectors and workers.

The shepherding problem is interesting for multiagent systems research as it requires multiple actors (e.g., dogs, humans) to exert indirect control over autonomous agents (e.g., sheep, cattle) for containment or transportation. Accordingly, plenty of research has focused on designing algorithms for robotic agents to solve such tasks. Almost no research, however, has utilized this task to investigate human-human or human-machine interactions, even though the shepherding problem encapsulates desirable qualities for an experimental paradigm to investigate the dynamics of human group and mixed-group coordination in complex tasks. This paper summarizes our recent research that has employed the shepherding problem to study complex multiagent human-human and human-machine interaction. The paper concludes with a discussion of practical applications for using the shepherding problem for the design of assistive agents that can be incorporated into human groups or enhance training and human learning.

This paper, deals with the usually need of data to simulate behavior and efficiency of proposed solutions in several fields, and also knowing that personal data always bring privacy and security issues. This work wants to promote a balanced solution between the need of personal information and the user’s privacy expectations. We propose a solution to overcome these issues, and don’t compromise the balance between security and personal comfort based on generating real context data of users, that allow to test user dependent systems.

TWM -Traffic Weighted Multi-maps- is presented as a novel traffic route guidance model to reduce urban traffic congestion, focusing on individual trip and collective objectives considering citizens, individual multi-modal mobility, and heterogeneous traffic groups. They have different interests, goals and regulation, so new multi-objective cost functions and control systems are required. TWM is structured around a novel control paradigm, based on the generation and distribution of complementary cost maps for traffic collectives (fleets), oriented towards the application of differentiated traffic planning and control policies. Agents receive a customized view TWM of the network that is used to calculate individual route using standard means and tools. The research describes the TWM theoretical model and microscopic simulations over standard reference traffic network grids, different traffic congestion scenarios, and several driver’s adherences to the mechanism. Travel-time results show that TWM can have a high impact on the network performance, leading to enhancements from 20% to 50%. TWM is conceived to be compatible with existing traffic routing systems. The research has promising future evolution applying new algorithms, policies and network profiles.

Trading strategies are often assessed against historical financial data in an effort to predict the profits and losses a strategy would generate in future. However, using only data from the past ignores the evolution of market microstructure and does not account for market conditions outside historical bounds. Simulations provide an effective supplement. We present an agent-based model to simulate financial market prices both under steady-state conditions and stress situations. Our new class of agents utilize recent advances in deep learning to make trading decisions and employ different trading objectives to ensure diversity in outcomes. The model supports various what-if scenarios such as sudden price crash, bearish or bullish market sentiment and shock contagion. We conduct evaluations on multiple asset classes including portfolio of assets and illustrate that the proposed agent decision mechanism outperforms other techniques. Our simulation model also successfully replicates the empirical stylized facts of financial markets.

The paper is addressing new challenges in agriculture, which are becoming nowadays critical for many countries, including climate changes, exhausted soils, aged farmers, etc. One of the new trends is associated with a step from Agriculture–4.0 focused on automation of physical processes for precision farming – to Agriculture–5.0 based on Artificial Intelligence (AI) for digitalization of domain knowledge and automation of farmer decision-making processes. A brief overview of existing IT systems for precision farming is given, key limitations are discussed and business requirements for developing AI solutions are formulated. The concept of digital eco-system of smart services for precision farming is proposed based on AI-technologies. The paper presents functionality and architecture of multi-agent platform and eco-system and identifies vitally important smart services for everyday operations of farmers. The structure and content of ontology-driven knowledge base for precision agriculture is considered, aimed at formalizing specifications of modern types of crops, agro- and bio-technologies, etc. The virtual “round table” is proposed as a generic framework for forming well-balanced recommendations for farmers with the use of ontology-based model of agricultural enterprise, which forms a specification of situation for automatic decision-making. Finally, the first case studies of the industrial prototype of the solution development are discussed.

The SMACH platform is a multi-agent based simulator supporting the study of human activity at the scale of a household, and its impact on the electricity consumption. It generates both activity diagrams and load curves for every electrical appliance in the household. Three different user interfaces can be used to manipulate the simulator: a participatory simulation interface, an educational interface and a technical interface for energy experts. This demonstration for PAAMS presents all three interfaces and the features offered by the SMACH platform.

Interoperability is a key factor of Cyber-Physical System (CPS) concept. Based on studies that use Multi-Agent System (MAS) as the core of a CPS, we are proposing to model many resources of the factories following the well-known Agents and Artifacts model for integrating agents and their environment. To enhance the interoperability of this system, we use the Apache Camel framework, a middleware to define routes to integrate a wide range of endpoints using different protocols. In this paper, we are demonstrating our camel component for artifacts in the context of Industry 4.0.

In this paper we demonstrate a tool that can be employed to build agent-based models (ABMs) for use in social archaeology. Specifically, our tool is based on the NetLogo modeling environment, and enables the creation of agent-based models of ancient societies, based on archaeological input. The models created by our tool can be used to obtain a better understanding of ancient societies, and assist archaeologists in testing the validity of existing or novel hypotheses and theories. We note that apart from assisting archaeologists in their work, the demonstrated tool can serve educational or recreational purposes as well: the ABMs created with the tool can, for instance, constitute the “backbone” of interactive platforms for use in schools or museums; and can conceivably be employed in history-focused digital strategic game environments.

Autonomous systems are expected to have a major impact in future coalition operations to assist humans in achieving complex tasks. Policies are typically used by systems to define their behavior and constraints and often these policies are manually configured and managed by humans. This paper presents a demonstration of a recent Generative Policy-based Model (GPM) approach applied to generating coalition policies for asset serviceability. This demonstrates the flexibility of the approach for generating policies in a distributed coalition environment to facilitate effective collaboration between coalition partners.

This demo shows an empirical analysis of the Dynamic Vehicle Routing Problem in multiple configurations, within a agent-swarm optimization scenario. With the purpose of statistically evaluating how solutions to this problem evolve when varying gobal system parameters, an agent-swarm simulator has been implemented within the Netlogo framework, and used to extract experimental data.

The present article aims to present the Heraclito environment that has in the Logical Studies and Exercises Logic (LOGOS) an important tool. The Heraclito environment assists students in solving various types of Logic exercises and provides the LOGOS Electronic Notebook to create and edit formulas, truth tables and proofs of Propositional Logic. The LOGOS Electronic Notebook is compatible with tablets, smartphones and PCs. It is currently being used with 1st and 2nd year undergraduate students in curricula in the scientific and technological areas. In addition, the Heraclito environment has an Intelligent Tutor System based on Multiagent Systems that identifies the individual knowledge of each student in the context of Natural Deduction in Propositional Logic.

We present a demonstration of two coordination methods for the application of multiagent reinforcement learning to the problem of traffic light signal control to decrease travel time. The first approach that we tested exploits the fact that the reward function can be splitted into contributions per agent. The second method computes the best response for a two player game with each member of its neighborhood. We apply both learning methods through SUMO traffic simulator, using data from the Transit Department of Bogotá, Colombia.

This paper describes the implementation of a modular, flexible and self-organized cyber-physical conveyor system build up with different individual modular and intelligent transfer modules. For this purpose, multi-agent systems (MAS) is used to distribute intelligence among transfer modules supporting plugability and modularity, complemented with self-organization capabilities to achieve a truly self-reconfigurable system.

In this demo work we develop a method to plan and coordinate a multi-agent team to gather information on demand. The data is periodically requested by a static Operation Center (OC) from changeable goals locations. The mission of the team is to reach these locations, taking measurements and delivering the data to the OC. Due to the limited communication range as well as signal attenuation because of the obstacles, the agents must travel to the OC, to upload the data. The agents can play two roles: ones as workers gathering data, the others as collectors traveling invariant paths for collecting the data of the workers to re-transmit it to the OC. The refreshing time of the delivered information depends on the number of available agents as well as of the scenario. The proposed algorithm finds out the best balance between the number of collectors-workers and the partition of the scenario into working areas in the planning phase, which provides the minimum refreshing time and will be the one executed by the agents.

Reducing peak electricity consumption is important to maximise use of renewable energy sources, and reduce the total amount of capacity required on a grid. Most approaches use a centralised optimisation algorithm run by a utility company. Here we develop a decentralised approach, where agents represent the interests of a household, and negotiate over when to run various appliances. We have developed an experimental framework that allows users’ perceived fairness of different negotiation algorithms to be evaluated.

This paper presents the EMiR robot, which is based on the RobElf robotic platform. EMiR has been developed as a cognitive assistant robot which is able to detect and to classify the emotional state of the human with whom it interacts. Moreover, EMiR integrates a powerful recommendation module that allows the robot to suggest activities to be done by the humans taking into account the emotional states among other aspects.

In this paper we propose the adoption of Agent Process Modelling, a theoretical-practical framework for the orchestration of agent behaviours running process models. The agent model is revisited from a distributed, cloud-native perspective and agents’ capabilities are externalized as microservices. Agent logic is represented in process models that orchestrate services execution by using a standard process notation, BPMN, increasing explainability and verification of agents decisions. As a preliminary work, we demonstrate a possible workflow to design and run agent-processes by instantiating the FIPA ContractNet protocol.

In this Demo we wish to demonstrate that it is not possible to evaluate a recommender’s ability on the basis of how good it is in carrying out the task we are interested in. On the contrary, a recommender should be evaluated as such. Although this mechanism is often used in literature, it unavoidably leads to incorrect results.

We present an agent based framework for improving multi-stakeholder optimisation problems, which we define as optimisation problems where the solution is utilised by a number of stakeholders who have their own local preferences. We explore our ideas within the domain of the University Timetabling Problem, demonstrating how a solution created by traditional timetabling methods may be further improved from the perspective of individual stakeholders (students) by agent based methods. We also note that this approach lends itself to increasing the level of trust in such systems by potentially allowing the stakeholders to view the actions taken by agents on their behalf.