Decentralized online planning, such as decentralized Monte Carlo tree search (MCTS), is an attractive paradigm for cooperative multi-agent systems in information-gathering tasks. However, current MCTS algorithms implicitly assume that agents are always available and actively contributing throughout the mission. In realistic, dynamic, and volatile environments, agent attrition is common and can severely degrade performance. In this paper, we demonstrate that agent attrition can cause current decentralized MCTS methods to perform arbitrarily worse than the optimum, particularly in applications with submodular reward functions. To address this issue, we propose Attritable Monte Carlo Tree Search (A-MCTS), a decentralized MCTS algorithm that adapts quickly and efficiently to reductions in the set of active agents. Our key idea is to have each agent build its search tree using the global utility function while coordinating with teammates through a regret-matching algorithm. Our theoretical analysis and extensive simulations show that A-MCTS maintains effective coordination among agents, even in high-attrition environments. We evaluate our approach in different information-gathering problems by modeling realistic reference scenarios. Results highlight that A-MCTS substantially improves over the main competing methods regarding global utility, scalability, and robustness.

This paper studies the performance of Floating Gossip, a novel decentralized approach for Gossip Learning at the network edge. Floating Gossip utilizes Floating Content to facilitate location-based probabilistic evolution of Machine Learning models, without external infrastructure support. We investigate dynamic scenarios requiring continuous learning, leveraging a mean field approach to analyze Floating Gossip's performance boundaries. Our focus is on the quantity of data that users can integrate into their models, as a function of key system parameters. Unlike previous studies that separately optimize communication or computational aspects of Gossip Learning, our methodology considers their combined effect. We validate our analysis through comprehensive simulations, demonstrating the high accuracy of our analytical model. Our methodology reveals Floating Gossip's effectiveness in training and updating Machine Learning models collaboratively, leveraging opportunistic exchanges between mobile users, while flexibly adapting to different user characteristics and mobility patterns. This research highlights Floating Gossip's potential for continuous, cooperative model training in dynamic, infrastructure-less environments, offering insight into its performance patterns and its potential in practical applications.

This paper introduces a novel analytical framework for evaluating energy-efficient, QoS-aware network-sharing strategies in cellular networks. Leveraging stochastic geometry, our framework enables the systematic assessment of network performance across a range of sharing paradigms, including both conventional single-operator scenarios and advanced hybrid strategies that enable full integration and cooperation among multiple mobile network operators. Our framework incorporates diverse user densities, rate requirements, and energy consumption models to ensure comprehensive analysis. Applying our results to real-world datasets from French mobile network operators, we demonstrate that hybrid network sharing can yield substantial energy savings, up to 35%, while maintaining QoS. Furthermore, our results allow us to characterize how the benefits of network sharing vary as a function of the geographical and functional characteristics of the deployment area. These findings highlight the potential of collaborative sharing strategies to enhance operational efficiency and sustainability in next-generation cellular networks.

Decentralized planning is a key element of cooperative multi-agent systems for information gathering tasks. However, despite the high frequency of agent failures in realistic large deployment scenarios, current approaches perform poorly in the presence of failures, by not converging at all, and/or by making very inefficient use of resources (e.g. energy). In this work, we propose Attritable MCTS (A-MCTS), a decentralized MCTS algorithm capable of timely and efficient adaptation to changes in the set of active agents. It is based on the use of a global reward function for the estimation of each agent’s local contribution, and regret matching for coordination. We evaluate its effectiveness in realistic data-harvesting problems under different scenarios. We show both theoretically and experimentally that A-MCTS enables efficient adaptation even under high failure rates. Results suggest that, in the presence of frequent failures, our solution improves substantially over the best existing approaches in terms of global utility and scalability.

In future radio access networks, machine learning (ML) based strategies for short-term forecasting of vehicular trajectories will be key for anticipatory resource allocation and management at the mobile edge. However, training ML models in a centralized fashion, over data collected from a massive heterogeneous and dynamic set of devices, poses significant scalability, reliability, and efficiency challenges, which are still open to date. In this article, we look at the specific issue of scalable and resource-efficient training of ML models in a vehicular environment. To address such a challenge, we propose a new Gossip Learning scheme, i.e., a fully distributed, collaborative training approach based on direct, opportunistic model exchanges via wireless device-to-device (D2D) communications with no centralized support. Our approach is based on constantly improving each node's own model instance through knowledge transfer among nodes, and on different strategies for estimating the potential contribution of neighboring nodes to the training process at a node. Extensive numerical assessments on a variety of measurement-based dynamic urban scenarios suggest that our schemes are able to converge rapidly and provide sufficiently accurate forecasts of vehicle position for time horizons which are typical of future 5 G/6 G dynamic resource allocation algorithms.

Coordinated multi-robot systems are an effective way to harvest data from sensor networks and implement active perception strategies. However, achieving efficient coordination in a way that guarantees a target QoS while adapting dynamically to changes (in the environment and/or in the system) is a key open issue. In this paper, we propose a novel decentralized Monte Carlo Tree Search (MCTS) algorithm for dynamic environments that allows agents to optimize their own actions while achieving some form of coordination. Its main underlying idea is to balance adaptively the exploration-exploitation trade-off to deal effectively with changes in the environment while filtering out outdated and irrelevant samples via a sliding window mechanism. We show both theoretically and through simulations that in dynamic environments our algorithm provides a log-factor (in terms of time steps) smaller regret than state-of-the-art decentralized multi-agent planning methods. We instantiate our approach to the problem of underwater data collection, showing in a variety of different settings that our approach greatly outperforms the best-competing approaches, both in terms of convergence speed and global utility.

In this paper we investigate the limit performance of Floating Gossip, a new, fully distributed Gossip Learning scheme which relies on Floating Content to implement location-based probabilistic evolution of machine learning models in an infrastructure-less manner.We consider dynamic scenarios where continuous learning is necessary, and we adopt a mean field approach to investigate the limit performance of Floating Gossip in terms of amount of data that users can incorporate into their models, as a function of the main system parameters. Different from existing approaches in which either communication or computing aspects of Gossip Learning are analyzed and optimized, our approach accounts for the compound impact of both aspects. We validate our results through detailed simulations, proving good accuracy. Our model shows that Floating Gossip can be very effective in implementing continuous training and update of machine learning models in a cooperative manner, based on opportunistic exchanges among moving users.

In dynamic settings, fully distributed gossip-based learning schemes have recently gained interest due to their better scalability, robustness, and enhanced privacy protection compared to server-based architectures. However, existing approaches to their performance characterization either assume stable connectivity among nodes or are ad-hoc for specific trace-based mobility patterns. Thus, in dynamic settings, there is currently a poor understanding of the conditions under which gossip-based learning schemes are feasible, and of their main performance tradeoffs. In this work, we start addressing this issue by performing a first baselining of Gossip Learning (GL) on random Time-Varying Graphs (TVG), to get a first-order characterization of their main performance patterns in dynamic settings. The use of random TVG enables a fine-grained and accurate characterization of GL effectiveness as a function of the main system parameters while abstracting from scenario-specific features of patterns of communication and mobility (e.g., induced by road grids or measured mobility traces). Our results suggest that GL schemes are robust to node mobility and comparable in accuracy and convergence speed to Federated Learning architectures, over a wide range of operational conditions. We show that the final model accuracy is robust against data dispersion across nodes as well as against very low rates of exchanges across nodes.

The role of fast yet reliable wireless communications in various application domains is getting ever more important. At the same time, as use cases are becoming more and more complex, application requirements are getting ever more stringent. One example is intelligent transportation, where the efficiency and reliability of wireless data delivery is essential for effective service support. As a consequence, in this context the adoption of AI techniques is widely considered crucial for enabling vehicular communications to adapt to dynamic changes of the environment. In this position paper, we discuss some representative applications of advanced AI tools in vehicular communications. In particular, we elaborate on the potential of distributed learning based on federated learning, of proactive service provisioning, and of graph neural network for enabling AI-native vehicular communications.

Natural and man-made disasters have often consequences on service availability in a wireless access network, provoking a progressively degraded performance or even the lack of connectivity. However, given the growing importance of situation awareness, telehealth, and advanced rescue teams coordination services for the affected population, it is key to restore these services in a timely fashion, while guaranteeing the required QoS levels. To this end, UAV-mounted base stations have been recently proposed as a key instrument to achieve this goal. Nonetheless, this gives rise to the key issue of how to deploy them in a resource efficient manner, in a post-disaster context typically characterized by lack of infrastructure support and of power supply. In this work, we tackle the issue of how to jointly optimize drones deployment and user association in a QOS aware manner to efficiently cater for coverage holes and QoS degradation in a cellular network after a disaster. We formulate a network optimization problem, and we provide a two-step genetic algorithm which iteratively tunes UAV position, base station transmit power and user association in order to minimize the number of employed drones. Initial results on a realistic measurement based scenario show that our approach is able to effectively minimize the number of deployed drones while achieving a target minimum QoS.

Nowcasting, i.e., short-term forecasting, of end user location is becoming increasingly important for anticipatory resource management in radio access networks (RAN). In this paper, we look at the case of vehicles moving in dense urban environments, and we tackle the location nowcasting problem with a particular class of machine learning (ML) algorithms that goes under the name Gossip Learning (GL). GL is a peer-to-peer machine learning approach based on direct, opportunistic exchange of models among nodes via wireless device-to-device (D2D) communications, and on collaborative model training. It has recently proven to scale efficiently to large numbers of static nodes, and to offer better privacy guarantees than traditional centralized learning architectures. We present new decentralized algorithms for GL, suitable for setups with dynamic nodes. In our approach, nodes improve their personalized model instance by sharing it with neighbors, and by weighting neighbors' contributions according to an estimate of their marginal utility. Our results show that the proposed GL algorithms are capable of providing accurate vehicle position predictions for time horizons of a few seconds, which are sufficient to implement effective anticipatory radio resource management.

Internet of Things (IoT) devices often have batteries of limited capacity, which are not easily replaced or recharged. This implies very short device lifetimes, and calls for a very careful device configuration to achieve the optimal trade-off between performance and power consumption. SWIPT (Simultaneous Wireless Information and Power Transfer) deals with this problem by harvesting energy at IoT devices from the received RF signals. Studying the efficiency of SWIPT in dealing with the energy and data transfer demands of IoT nodes leads to a number of open issues. In this paper, we devise an analytical model based on stochastic geometry for a SWIPT radio access network with a dense population of IoT users. With our model, it is possible to accurately study the impact of the system parameters on the key system performance indicators, while accounting in a realistic manner for device performance, and for the statistics of time scheduling at base stations. This allows us to understand (not without some surprise) what are the most effective strategies to minimize energy consumption in a SWIPT network, and what is their potential for energy savings.

Coordinated multi-robot systems are an effective way to harvest data from sensor networks and to implement active perception strategies. However, achieving efficient coordination in a way which guarantees a target QoS while adapting dynamically to changes (in the environment, due to sensors’ mobility, and/or in the value of harvested data) is to date a key open issue. In this paper, we propose a novel decentralized Monte Carlo Tree Search algorithm (MCTS) which allows agents to optimize their own actions while achieving some form of coordination, in a changing environment. Its key underlying idea is to balance in an adaptive manner the exploration-exploitation trade-off to deal effectively with abrupt changes caused by the environment and random changes caused by other agents’ actions. Critically, outdated and irrelevant samples - an inherent and prevalent feature in all multi-agent MCTS-based algorithms - are filtered out by means of a sliding window mechanism. We show both theoretically and through simulations that our algorithm provides a log-factor (in terms of time steps) smaller regret than state-of-the-art decentralized multi-agent planning methods. We instantiate our approach on the problem of underwater data collection, showing on a set of different models for changes that our approach greatly outperforms the best available algorithms for that setting, both in terms of convergence speed and of global utility.

Floating Content (FC) is a paradigm for localized infrastructure-less content dissemination, that aims at sharing information among nodes within a restricted geographical area by relying only on opportunistic content exchanges. FC provides the basis for the probabilistic spatial storage of shared information in a completely decentralized fashion, usually without support from dedicated infrastructure. One of the key open issues in FC is the characterization of its performance limits as functions of the system parameters, accounting for its reliance on volatile wireless exchanges and on limited user resources. This paper takes a first step towards tackling this issue, by elaborating a model for the storage capacity of FC, i.e., for the maximum amount of information that can be stored through the FC paradigm. The storage capacity of FC, and of similar probabilistic content dissemination systems, is evaluated with a powerful information theoretical approach, based on a mean field model of opportunistic information exchange. In addition, an extremely simple explicit approximate expression for storage capacity is derived. The numerical results generated by our analytical models are compared to the predictions of realistic simulations under different setups, proving the accuracy of our analytical approaches, and characterizing the properties of the FC storage capacity.

Although many target applications in VANETs are information-centric, the performance of Named Data Networking (NDN) in vehicular ad-hoc networks is severely hampered by persistent network partitioning, typical of many vehicular scenarios. Existing approaches try to address this issue by relying on opportunistic communications. However, they leave open the crucial issue of how to guarantee content persistence and tight QoS levels while optimizing the resource utilization in the vehicular environment. In this work we propose DeepNDN, a communication scheme based on the joint application of NDN and of probabilistic spatial content caching, which enables content retrieval in fragmented and dynamic network topologies with tight delay constraints. We present a data-based approach to DeepNDN management, based on locally modulating content replication and delivery in order to achieve a target hit ratio in a resource-efficient manner. Our management algorithm employs a Convolutional Neural Network (CNN) architecture for effectively capturing the complex relations between spatio-temporal patterns of mobility and content requests and DeepNDN performance. Its numerical assessment in realistic, measurement-based scenarios suggest that our management approach achieves its target set goals while outperforming a set of reference schemes.

Floating Content (FC) is a paradigmatic example of opportunistic infrastructure-less content sharing system where information is spread upon mobile node encounters within an area which is called the replication zone. FC allows the probabilistic spatial storage of information, even in the case of unreliable communications, with no support from dedicated servers. Given the large amount of communication and storage resources typically required to guarantee content persistence despite node mobility, a major open issue for the practical viability of FC and of similar distributed storage systems is the characterization of their storage capacity, i.e., of the maximum amount of information which can be stored for a given set of system parameters. In this paper, we propose a simple yet powerful information theoretical model of the storage capacity of probabilistic distributed storage systems such as FC, based on a mean field model of opportunistic information exchange. We evaluate numerically our results, and validate the model by means of realistic simulations, showing the accuracy of our mean field approach and characterizing the properties of the FC storage capacity versus the main system parameters.

In this work, we demonstrate the feasibility and the main functionalities of a low-cost, nomadic platform for Smart District services. The goal of the platform is to enable the extension of Smart City services to smaller cities, to towns and into the countryside, bypassing the natural barriers through the use of services and vectors which are naturally cross-domain, such as public transportation vehicles, and moving people. the platform is based on mobile and opportunistic sensing, on crowdsensing, and on strategies for community engagement and co-creation mediated by a map-based crowdsourcing application. The demo will showcase the main components of the system, as well as a sample set of Smart District services supported by the platform.

Floating Content (FC) has recently been proposed as an attractive application for mobile networks, such as VANETs, to operate opportunistic and distributed content sharing over a given geographic area, namely Anchor Zone (AZ). FC performances are tightly dependent on the AZ size, which in literature is classically chosen by the node that generates the floating message. In the present work, we propose a method to improve FC performances by optimizing the AZ size with the support of a Software Defined Network (SDN) controller, which collects mobility information, such as speed and position, of the vehicles in its coverage range.

The surging interest in autonomous coordinateddriving and in proactive safety services, exploiting the wealth ofsensing and computing resources which are gradually permeatingthe urban and vehicular environments, is making provisioning ofhigh levels of QoS in vehicular networks an urgent issue. At thesame time, the spreading model of a smart car, with a wealthof infotainment applications, calls for architectures for vehicularcommunications capable of supporting traffic with a diverse setof performance requirements. So far efforts have been revolvedtowards enabling a single specific QoS level. But the issues of howto support traffic with tight QoS requirements (no packet loss, and delays inferior to 1ms), and of designing a system capable atthe same time of efficiently sustaining such traffic together withtraffic from infotainment applications, are still open. In this paper we present the approach taken in the SNFCONTACT project in order to tackle these issues. The goal ofthe project is to investigate how an architecture for vehicularcommunications which integrates content-centric networking, software-defined networking as well as context aware floatingcontent schemes can properly support the very diverse set ofapplication and services currently envisioned for the vehicularenvironment.

Network densification is currently seen as one of the key techniques enabling Radio Access Networks (RANs) to meet the performance and functional requirements of the 5G paradigm in urban areas. Avoiding the connection of small cells to the power grid facilitates their deployment and reduces both capital and operational expenditures (CAPEX and OPEX). In this paper, we propose an approach to enable net-Zero Energy Networking (ZEN) in urban scenarios, based on dynamically managing the operating point of Base Stations (BSs), so as to drastically decrease their overall energy requirements. More specifically, we formalize the problem of optimizing the CAPEX of a ZEN, while guaranteeing quality of service (QoS) and a given level of service availability. Optimization is over power system parameters (solar panel area, battery capacity) as well as over BS power levels and user association over time. We propose a practical algorithm for the derivation of QoS-aware spatio-temporal strategies in terms of user association and BS power allocation, which, for a given expected pattern of renewable power generation, minimize the probability of service unavailability due to power shortage. Through extensive simulations using measured data, and realistic BS deployments, we show that our algorithm leads to substantial reduction in CAPEX, and to feasible power system requirements.

The IoT paradigm holds the promise to revolutionize the way we live and work by means of a wealth of new services, based on seamless interactions between a large amount of heterogeneous devices. After decades of conceptual inception of the IoT, in recent years a large variety of communication technologies has gradually emerged, reflecting a large diversity of application domains and of communication requirements. Such heterogeneity and fragmentation of the connectivity landscape is currently hampering the full realization of the IoT vision, by posing several complex integration challenges. In this context, the advent of 5G cellular systems, with the availability of a connectivity technology, which is at once truly ubiquitous, reliable, scalable, and cost-efficient, is considered as a potentially key driver for the yet-to emerge global IoT. In the present paper, we analyze in detail the potential of 5G technologies for the IoT, by considering both the technological and standardization aspects. We review the present-day IoT connectivity landscape, as well as the main 5G enablers for the IoT. Last but not least, we illustrate the massive business shifts that a tight link between IoT and 5G may cause in the operator and vendors ecosystem.

In this paper we present an analytic approach to performance analysis of ad hoc networks under non saturation conditions, which does not rely on any assumption on traffic statistics. Our approach assumes traffic to be constrained by leaky bucket arrival curves, and it relies on a coupled processors model to capture the dependencies between user achievable rates due to sharing of the wireless transmission medium. We derive sufficient conditions for stability of transmission queues in an ad hoc network, and we describe a method for the determination of a proportionally fair allocation of resources, which allows trading the fairness of the solution for computational complexity. We validate our results through simulations, showing how our approach allows deriving operating points which both increase the fairness of the allocation and the overall average utilization of network resources with respect to saturated models.

This work presents the first experimental evaluation of the Floating Content (FC) communication paradigm in a campus/large office setting. By logging information transfer events we have characterized mobility patterns, and we have assessed the performance of services implemented using the FC paradigm. Our results unveil the key relevance of group dynamics in user movements for the FC performance. Surprisingly, in such an environment, our results show that a relatively low user density is enough to guarantee content persistence over time, contrarily to predictions from available models. Based on these experimental findings, we develop a novel simple analytical model that accounts for the peculiarities of the mobility patterns in such a setting, and that can accurately predict the effectiveness of FC for the implementation of services in a campus/large office setting.

Context-awareness is a peculiar characteristic of an ever expanding set of applications that make use of a combination of restricted spatio-temporal locality and mobile communications, to deliver a variety of services to the end user. It is expected that by 2014 more than 1.5 billion people would be using applications based on local search (search restricted on the basis of spatio-temporal locality), and that mobile location based services will drive revenues of more than $15 billion worldwide. A common feature of such context-aware applications is the fact that their communication requirements significantly differ from ordinary applications. For most of them, the scope of generated content itself is local. This locally relevant content may be of little concern to the rest of the world, therefore moving this content from the user device to store it in a well-accessible centralized location and/or making this information available beyond its scope represents a clear waste of resources (connectivity, storage). Due to these specific requirements, opportunistic communication can play a special role when coupled with context-awareness. The benefit of opportunistic communications is that it naturally incorporates context as spatial proximity is closely associated with connectivity.

Context-awareness is a peculiar characteristic of an expanding set of applications that make use of a combination of restricted spatio-temporal locality and mobile communications, to deliver a variety of services. Opportunistic communications satisfy well the communication requirements of these applications, because they naturally incorporate context. Recently, an opportunistic communication paradigm called "Floating Content" (FC) was proposed, to support infrastructure-less, distributed content sharing. But how good is floating content in supporting context-aware applications? In this work, we present a simple approximate analytical model for the performance analysis of context-aware applications that use floating content. We estimate the "success probability" for a representative category of context-aware applications, and show how the system can be configured to achieve the application's target QoS. We validate our model using extensive simulations under different settings and mobility patterns, showing that our model-based predictions are highly accurate under a wide range of conditions.

We consider the problem of minimizing the energy consumed in a cellular access network, under loads that slowly vary over space and time, while guaranteeing quality of service (QoS). In particular, we formalize the problem of jointly optimizing the base stations (BS) power levels and the association of users to BSs, while guaranteeing a minimum throughput to each user, and a target value of blocking probability. We propose abstractions that enable tracking of long-term spatial load distributions, and a practical algorithm for energy efficient user association and base station power allocation. Our algorithm is applicable to arbitrary (planar) BS layouts, to settings with interference, to different BS energy models, and to arbitrary user distributions over the service area. Through extensive simulations using measured data, and realistic BS deployments, we show that our algorithm leads to substantial energy savings both with traditional BS designs and with energy-proportional equipment, and we demonstrate the potential of BS sleep modes to achieve network-level energy proportionality.

Despite file distribution applications are responsible for a major portion of the current Internet traffic, so far little effort has been dedicated to study file distribution from the point of view of energy efficiency. In this paper, we present the first extensive and detailed theoretical study for the problem of energy efficiency in file distribution. Specifically, we first demonstrate that the general problem of minimizing energy consumption in file distribution is NP-hard. For restricted versions of the problem, we derive tight lower bounds on energy consumption, and we design a family of algorithms that achieve these bounds. Our results prove that through collaborative p2p schemes up to 50% energy savings are achievable with respect to the best available centralized file distribution scheme. Through simulation, we show that even in heterogeneous settings (e.g., considering network congestion, and link variability across hosts) our collaborative algorithms always achieve significant energy savings with respect to the power consumption of centralized file distribution systems.

Aggregate scheduling is one of the most promising solutions to the issue of scalability in networks, like DiffServ networks and high speed switches, where hard QoS guarantees are required. For networks of FIFO aggregate schedulers, the main existing sufficient conditions for stability (the possibility to derive bounds to delay and backlog at each node) are of little practical utility, as they are either relative to specific topologies, or based on strong ATM-like assumptions on the network (the so-called "RIN" result), or they imply an extremely low node utilization. We use a deterministic approach to this problem. We identify a nonlinear operator on a vector space of finite (but large) dimension, and we derive a first sufficient condition for stability, based on the super-additive closure of this operator. Second, we use different upper bounds of this operator to obtain practical results. We find new sufficient conditions for stability, valid in an heterogeneous environment and without any of the restrictions of existing results. We present a polynomial time algorithm to test our sufficient conditions for stability. We show that with leaky bucket constrained flows the inner bound to the stability region derived with our algorithm is always larger than the one determined by all existing results. We prove that all the main existing results can be derived as special cases of our results. We also present a method to compute delay bounds in practical cases.