In cloud-native systems, Kubernetes clusters with interdependent services often face challenges to their operational resilience due to poor workload management issues such as resource blocking, bottlenecks, or continuous pod crashes. These vulnerabilities are further amplified in adversarial scenarios, such as Distributed Denial-of-Service attacks (DDoS). Conventional Horizontal Pod Autoscaling (HPA) approaches struggle to address such dynamic conditions, while reinforcement learning-based methods, though more adaptable, typically optimize single goals like latency or resource usage, neglecting broader failure scenarios. We propose decomposing the overarching goal of maintaining operational resilience into failure-specific sub-goals delegated to collaborative agents, collectively forming an HPA Multi-Agent System (MAS). We introduce an automated, four-phase online framework for HPA MAS design: 1) modeling a digital twin built from cluster traces; 2) training agents in simulation using roles and missions tailored to failure contexts; 3) analyzing agent behaviors for explainability; and 4) transferring learned policies to the real cluster. Experimental results demonstrate that the generated HPA MASs outperform three state-of-the-art HPA systems in sustaining operational resilience under various adversarial conditions in a proposed complex cluster.

As cyber-attacks show to be more and more complex and coordinated, cyber-defenders strategy through multi-agent approaches could be key to tackle against cyberattacks as close as entry points in a networked system. This paper presents a Markovian modeling and implementation through a simulator of fighting cyber-attacker agents and cyberdefender agents deployed on host network nodes. It aims to provide an experimental framework to implement realistically based coordinated cyber-attack scenarios while assessing cyberdefenders dynamic organizations. We abstracted network nodes by sets of properties including agents' ones. Actions applied by agents model how the network reacts depending in a given state and what properties are to change. Collective choice of the actions brings the whole environment closer or farther from respective cyber-attackers and cyber-defenders goals. Using the simulator, we implemented a realistically inspired scenario with several behavior implementation approaches for cyber defenders and cyber-attackers.

In the context of the circular economy, we conceived a new decision support system. Its goal is to design new products from highly variable post-used components. This system is open, allowing components to be added or removed at anytime. Components are avatarized as virtual agents and seen as a multiagent system. A user can require the system to design products. Therefore agents of the system must form coalitions that represent a product. We propose a new socially inspired agent model adapted to this system to tackle a coalition structure generation problem. This agent-centered model allows agents to find out with which agent to form a coalition based on the Humanities and Social Sciences principles of attraction.

The cyber-physical systems deployment takes up an ever-increasing number of fields of application. The variety of platforms used for implementing cyber-physical systems and the dynamic interconnection of their components make the resilience issue of these systems very important. Our work aims to use the knowledge from human and social sciences, particularly to inspire from the emotional processes for creating an agent architecture that increases the cyber-physical systems resilience. Emotional inspiration occurs in the individual decision-making processes as well as in the social coordination mechanisms. In addition, one of the principal hypotheses in our research is that the multi-agent paradigm is suitable for integrating emotional processes into cyber-physical systems.

The concept of resilience is popular and studied in different domains like human and social sciences (psychology, psychiatry, sociology etc.) and recently in cognitive science, biology, ecology and computer science. The objective of this article is to present our research that aims to incorporate knowledge from human and social sciences in computer science to solve cyber-physical systems resilience problems. For us humans, emotions are considered as an important process in responding to unanticipated events, for that, emotions are important for our resilience. Our work aims to inspire from emotional processes to create an agent model that increases the resilience of cyber-physical systems. This agent model will integrate individual and collective processes. In addition, one of our principal hypotheses in our research is that the multi-agent paradigm is suitable to integrate emotion-like processes into cyber-physical systems.

International audience

In this article we propose an agent-based architecture to create an ecosystem of Intelligent Environments (IE's). The distinctive characteristics of this proposal are: it takes into account the changing needs of the environment's dwellers, that is, the IE is sensitive to the context, and it allows the user to move among the ecosystem's elements freely. With these features, the ecosystem manages all required information to satisfy the users' needs, even when they are in transit between two IE's. The key element in our proposal is an ontology that is used in three main processes in our ecosystem. First of all, it works together with sensor information to provide the sensitivity of the context, second, it is used to decide whether local resources are enough to fulfill user needs or if it is necessary to use the ecosystem's resources, and third, the ontology is also useful in the interaction among different IE's to satisfy some users' requirements.

This paper proposes an approach which aims at improving information access (skills, services) in large networks. It proposes a self-organized multiagent analysis of the problem allowing to reduce the number of messages transmitted for a skill search.

This paper deals with the design and the simulation of decentralized artificial intelligent systems (multiagent systems) in the context of distributed automation and decentralized control systems. It presents an original tool and its interesting modular and customizable physical model of a house.

Designing a real world complex system requires to use decentralized artificial intelligence. Such a system involves self-organization techniques, advanced interaction models, real-time architectures... To simplify the design step, it can be necessary to reproduce the behavior of the whole designed system in a virtual environment before to deploy the designed application on the real world devices. In such a deployment, the used models must be degraded in respect with device constraints (memory, CPU) and other physical constraints (thermal dissipation, energy consumption...). This paper proposes a new approach and a tool to support such a deployment.

This paper introduces an approach which aims at improving information access (skills, services) in large networks. A self-organized multiagent analysis of the problem, reducing the number of messages transmitted for a skill search is proposed. The MWAC (Multi-Wireless-Agent Communication) model is extended here to take some specificities of social networks into account, like the information held by the social network members and their connections with the others.

Web services (WS) are software components that offer functionalities for users through Internet. WS composition is a way to fulfill a complex request which can't be achieved by a single service. WS composition's approaches do not offer a dynamic and adaptative composition. In this paper, we defend recursive multi-agent systems as efficient tools to solve this problem and we propose a recursive multi-agent system based model for dynamic and adaptative WS composition.

The Multi-Wireless-Agent-Communication (MWAC) model uses a multiagent self-organization process through to manage communications in wireless sensor networks. This paper presents a first improvement of the MWAC model using an ant colony optimization. We introduce the use of ant colony optimization in the context of multiagent systems and wireless sensor networks. Ants are not simply deployed on the multiagent society. Each ant interacts with agents to obtain the best compromise on the routes choice to enable a reliable communication.

Multiagent systems are well suited to specify and to design requirements for hardware / software collaborative systems. Designing embedded multiagent systems require skills in hardware design and in software design. However, up to now, no method allows to build software/hardware hybrid multiagent systems. This paper introduces the DIAMOND hardware/software codesign method.

Multiagent systems (MAS) are well suited to specify requirements far open physical complex systems. However, up to now, no method allows to actually build software/hardware hybrid MAS. This paper presents an original method far designing embedded MAS.

An onboard multi-agent system for mobile robots requires improvements in reactivity and real time efficiency. The authors present a real time computing model blackboard with specific features to guarantee satisfactory behavior of a low level robotic control. They detail a mobile robot controller to evaluate the improvement offered by such a system in a real application. They present simulation software based on an essential event and state concept.

In order to specify intelligent control for robotic systems, a generic blackboard based methodology is proposed. It is based on a functional analysis of the problem of control. The control is viewed as a distributed multishape blackboard decomposition. Specific functionalities of blackboard architectures are specialized for each of the aspects of control: a graphical blackboard for interaction, a classical one for reasoning, and a parallel real-time one for control. An original parallel blackboard architecture which integrates the whole system is presented. A model is supplied to express activity and genericity of this tool.>