Andrea Passarella

Biancamaria Bombino, Chiara Boldrini, Andrea Passarella et al.

Inferring the sign of social relationships from online interactions is a fundamental challenge in social network analysis. Existing approaches typically rely on sentiment analysis to label individual interactions as positive or negative, then aggregate these labels to assign a sign to the relationship. However, sentiment analysis captures the valence of the content being discussed rather than the nature of the relational exchange itself, a conflation that can lead to systematic misclassification. In this paper, we propose a methodology that addresses this limitation by leveraging Large Language Models (LLMs) in a zero-shot setting to identify interaction-level relational signals (specifically, personal praise and personal attacks directed at the interlocutor) as more direct indicators of positive and negative social ties. We evaluate four models spanning open-weight and proprietary architectures (Qwen2.5:7b, Gemma2:9b, GPT-4o, GPT-5.4-mini) across three prompt designs of increasing complexity, on two human-annotated datasets of approximately 298 and 340 texts respectively. Results show that zero-shot LLMs achieve good classification performance on both tasks without any task-specific training data, establishing a practical baseline for relational annotation. Performance differs across tasks: attack detection is robust to prompt design and model choice, while praise detection is more sensitive to both, reflecting the greater subjectivity of positive relational gestures. These findings lay the groundwork for integrating LLM-based relational annotation into sign prediction pipelines.

Dino Pedreschi, Luca Pappalardo, Emanuele Ferragina et al.

Human-AI coevolution, defined as a process in which humans and AI algorithms continuously influence each other, increasingly characterises our society, but is understudied in artificial intelligence and complexity science literature. Recommender systems and assistants play a prominent role in human-AI coevolution, as they permeate many facets of daily life and influence human choices on online platforms. The interaction between users and AI results in a potentially endless feedback loop, wherein users' choices generate data to train AI models, which, in turn, shape subsequent user preferences. This human-AI feedback loop has peculiar characteristics compared to traditional human-machine interaction and gives rise to complex and often ``unintended'' social outcomes. This paper introduces Coevolution AI as the cornerstone for a new field of study at the intersection between AI and complexity science focused on the theoretical, empirical, and mathematical investigation of the human-AI feedback loop. In doing so, we: (i) outline the pros and cons of existing methodologies and highlight shortcomings and potential ways for capturing feedback loop mechanisms; (ii) propose a reflection at the intersection between complexity science, AI and society; (iii) provide real-world examples for different human-AI ecosystems; and (iv) illustrate challenges to the creation of such a field of study, conceptualising them at increasing levels of abstraction, i.e., technical, epistemological, legal and socio-political.

Francesco Di Cursi, Chiara Boldrini, Marco Conti et al.

Email archives offer a rare view of social relationships through repeated communication, but it remains unclear how well classical ego network layering applies to digital interaction data. This paper compares two public email archives with sharply contrasting structures: Enron, a workplace corpus involving around 150 users, and Jmail, a single-ego archive centered on an exceptionally active focal actor whose communication volume is more than twenty times higher than the average Enron user. We ask, in each case, whether Dunbar-like layered organization is recoverable from email communication frequency and how it should be interpreted. For Jmail, we show that extreme communication intensity causes standard layering methods (whether clustering-based or threshold-based) to break down. Jmail is not a broad communication environment with many occasional contacts, but a selective pool of high-interest alters operating on a much higher frequency scale than ordinary email. Once the Dunbar frequency ladder is anchored to the empirical support-clique boundary, a clearer layered structure emerges. Reciprocity analysis confirms that the recovered layers reflect genuine bidirectional relationships rather than artifacts of the focal actor's outgoing activity. Enron serves as a workplace benchmark that grounds the comparison: its ego networks partially reproduce Dunbar-like organization, with stable inner circles and an outermost recovered layer corresponding to Dunbar's affinity group ($\sim50$), confirming that layered structure is recoverable from ordinary organizational email. Overall, the findings show that Dunbar-like organization can be meaningfully studied in email archives, but that selective high-frequency archives require frequency normalization before the layered structure becomes interpretable.

Ravi Dhiman, Andrea Passarella, Mirco Tribastone et al.

Neural network compression is commonly achieved by pruning parameters based on local importance scores, e.g., magnitude-based pruning. We propose a complementary approach that compresses models by aggregating neurons with similar functional behavior rather than removing weights independently. Our method encodes a trained network as a polynomial ODE system and applies a lumping method called Approximate Forward Differential Equivalence to identify neurons with approximately matching induced dynamics. A single tolerance parameter, $\varepsilon$, controls the compression level and induces a smooth trade-off between model size and predictive accuracy. We evaluate the method on synthetic datasets derived from nonlinear dynamical systems with known ground-truth behavior and on public regression benchmarks. Across both settings, the proposed approach achieves substantial parameter reduction while preserving accuracy, and consistently compares favorably with magnitude-based pruning and Wanda at similar compression levels. These results suggest that differential equivalence-based aggregation is a principled and effective alternative to conventional weight-centric pruning.

Andrew Fuchs, Andrea Passarella, Marco Conti

With humans interacting with AI-based systems at an increasing rate, it is necessary to ensure the artificial systems are acting in a manner which reflects understanding of the human. In the case of humans and artificial AI agents operating in the same environment, we note the significance of comprehension and response to the actions or capabilities of a human from an agent's perspective, as well as the possibility to delegate decisions either to humans or to agents, depending on who is deemed more suitable at a certain point in time. Such capabilities will ensure an improved responsiveness and utility of the entire human-AI system. To that end, we investigate the use of cognitively inspired models of behavior to predict the behavior of both human and AI agents. The predicted behavior, and associated performance with respect to a certain goal, is used to delegate control between humans and AI agents through the use of an intermediary entity. As we demonstrate, this allows overcoming potential shortcomings of either humans or agents in the pursuit of a goal.

Andrew Fuchs, Andrea Passarella, Marco Conti

There is a clear desire to model and comprehend human behavior. Trends in research covering this topic show a clear assumption that many view human reasoning as the presupposed standard in artificial reasoning. As such, topics such as game theory, theory of mind, machine learning, etc. all integrate concepts which are assumed components of human reasoning. These serve as techniques to attempt to both replicate and understand the behaviors of humans. In addition, next generation autonomous and adaptive systems will largely include AI agents and humans working together as teams. To make this possible, autonomous agents will require the ability to embed practical models of human behavior, which allow them not only to replicate human models as a technique to "learn", but to to understand the actions of users and anticipate their behavior, so as to truly operate in symbiosis with them. The main objective of this paper it to provide a succinct yet systematic review of the most important approaches in two areas dealing with quantitative models of human behaviors. Specifically, we focus on (i) techniques which learn a model or policy of behavior through exploration and feedback, such as Reinforcement Learning, and (ii) directly model mechanisms of human reasoning, such as beliefs and bias, without going necessarily learning via trial-and-error.

Andrew Fuchs, Andrea Passarella, Marco Conti

As we discussed in Part I of this topic, there is a clear desire to model and comprehend human behavior. Given the popular presupposition of human reasoning as the standard for learning and decision-making, there have been significant efforts and a growing trend in research to replicate these innate human abilities in artificial systems. In Part I, we discussed learning methods which generate a model of behavior from exploration of the system and feedback based on the exhibited behavior as well as topics relating to the use of or accounting for beliefs with respect to applicable skills or mental states of others. In this work, we will continue the discussion from the perspective of methods which focus on the assumed cognitive abilities, limitations, and biases demonstrated in human reasoning. We will arrange these topics as follows (i) methods such as cognitive architectures, cognitive heuristics, and related which demonstrate assumptions of limitations on cognitive resources and how that impacts decisions and (ii) methods which generate and utilize representations of bias or uncertainty to model human decision-making or the future outcomes of decisions.

Kilian Ollivier, Chiara Boldrini, Andrea Passarella et al.

Well-established cognitive models coming from anthropology have shown that, due to the cognitive constraints that limit our "bandwidth" for social interactions, humans organize their social relations according to a regular structure. In this work, we postulate that similar regularities can be found in other cognitive processes, such as those involving language production. In order to investigate this claim, we analyse a dataset containing tweets of a heterogeneous group of Twitter users (regular users and professional writers). Leveraging a methodology similar to the one used to uncover the well-established social cognitive constraints, we find regularities at both the structural and semantic level. At the former, we find that a concentric layered structure (which we call ego network of words, in analogy to the ego network of social relationships) very well captures how individuals organise the words they use. The size of the layers in this structure regularly grows (approximately 2-3 times with respect to the previous one) when moving outwards, and the two penultimate external layers consistently account for approximately 60% and 30% of the used words, irrespective of the number of the total number of layers of the user. For the semantic analysis, each ring of each ego network is described by a semantic profile, which captures the topics associated with the words in the ring. We find that ring #1 has a special role in the model. It is semantically the most dissimilar and the most diverse among the rings. We also show that the topics that are important in the innermost ring also have the characteristic of being predominant in each of the other rings, as well as in the entire ego network. In this respect, ring #1 can be seen as the semantic fingerprint of the ego network of words.

Quang-Phong Tran, Claudio Cicconetti, Marco Conti et al.

Sharing entanglement among multiple users remains a central challenge for scalable quantum networks. Recent work proposed an on-demand entanglement packet architecture in which a controller uses a Time Division Multiple Access (TDMA) approach to allocate network resources. Quantum nodes are assigned a periodic schedule that probabilistically fulfills application requests for end-to-end entanglements. The schedule is recomputed periodically using well-known algorithms, such as Earliest Deadline First (EDF). However, a static schedule offers limited flexibility when outcomes are stochastic and arrivals are asynchronous. To overcome this limitation, we propose an online scheduler that dynamically schedules, defers, retries, or drops entanglement distribution reservations. In our simulations, the dynamic scheduler achieves lower completion time, higher completion ratio, and higher throughput than the static baseline. Furthermore, when the network is overloaded, the dynamic scheduler continues to construct deadline-feasible schedules and degrades gracefully.

Luigi Palmieri, Lorenzo Valerio, Chiara Boldrini et al.

In a decentralized machine learning system, data is typically partitioned among multiple devices or nodes, each of which trains a local model using its own data. These local models are then shared and combined to create a global model that can make accurate predictions on new data. In this paper, we start exploring the role of the network topology connecting nodes on the performance of a Machine Learning model trained through direct collaboration between nodes. We investigate how different types of topologies impact the "spreading of knowledge", i.e., the ability of nodes to incorporate in their local model the knowledge derived by learning patterns in data available in other nodes across the networks. Specifically, we highlight the different roles in this process of more or less connected nodes (hubs and leaves), as well as that of macroscopic network properties (primarily, degree distribution and modularity). Among others, we show that, while it is known that even weak connectivity among network components is sufficient for information spread, it may not be sufficient for knowledge spread. More intuitively, we also find that hubs have a more significant role than leaves in spreading knowledge, although this manifests itself not only for heavy-tailed distributions but also when "hubs" have only moderately more connections than leaves. Finally, we show that tightly knit communities severely hinder knowledge spread.

Luigi Palmieri, Chiara Boldrini, Lorenzo Valerio et al.

Fully decentralized learning enables the distribution of learning resources and decision-making capabilities across multiple user devices or nodes, and is rapidly gaining popularity due to its privacy-preserving and decentralized nature. Importantly, this crowdsourcing of the learning process allows the system to continue functioning even if some nodes are affected or disconnected. In a disaster scenario, communication infrastructure and centralized systems may be disrupted or completely unavailable, hindering the possibility of carrying out standard centralized learning tasks in these settings. Thus, fully decentralized learning can help in this case. However, transitioning from centralized to peer-to-peer communications introduces a dependency between the learning process and the topology of the communication graph among nodes. In a disaster scenario, even peer-to-peer communications are susceptible to abrupt changes, such as devices running out of battery or getting disconnected from others due to their position. In this study, we investigate the effects of various disruptions to peer-to-peer communications on decentralized learning in a disaster setting. We examine the resilience of a decentralized learning process when a subset of devices drop from the process abruptly. To this end, we analyze the difference between losing devices holding data, i.e., potential knowledge, vs. devices contributing only to the graph connectivity, i.e., with no data. Our findings on a Barabasi-Albert graph topology, where training data is distributed across nodes in an IID fashion, indicate that the accuracy of the learning process is more affected by a loss of connectivity than by a loss of data. Nevertheless, the network remains relatively robust, and the learning process can achieve a good level of accuracy.

Mirko Nardi, Lorenzo Valerio, Andrea Passarella

Federated learning (FL) is proving to be one of the most promising paradigms for leveraging distributed resources, enabling a set of clients to collaboratively train a machine learning model while keeping the data decentralized. The explosive growth of interest in the topic has led to rapid advancements in several core aspects like communication efficiency, handling non-IID data, privacy, and security capabilities. However, the majority of FL works only deal with supervised tasks, assuming that clients' training sets are labeled. To leverage the enormous unlabeled data on distributed edge devices, in this paper, we aim to extend the FL paradigm to unsupervised tasks by addressing the problem of anomaly detection in decentralized settings. In particular, we propose a novel method in which, through a preprocessing phase, clients are grouped into communities, each having similar majority (i.e., inlier) patterns. Subsequently, each community of clients trains the same anomaly detection model (i.e., autoencoders) in a federated fashion. The resulting model is then shared and used to detect anomalies within the clients of the same community that joined the corresponding federated process. Experiments show that our method is robust, and it can detect communities consistent with the ideal partitioning in which groups of clients having the same inlier patterns are known. Furthermore, the performance is significantly better than those in which clients train models exclusively on local data and comparable with federated models of ideal communities' partition.

Andrew Fuchs, Andrea Passarella, Marco Conti

In the context of humans operating with artificial or autonomous agents in a hybrid team, it is essential to accurately identify when to authorize those team members to perform actions. Given past examples where humans and autonomous systems can either succeed or fail at tasks, we seek to train a delegating manager agent to make delegation decisions with respect to these potential performance deficiencies. Additionally, we cannot always expect the various agents to operate within the same underlying model of the environment. It is possible to encounter cases where the actions and transitions would vary between agents. Therefore, our framework provides a manager model which learns through observations of team performance without restricting agents to matching dynamics. Our results show our manager learns to perform delegation decisions with teams of agents operating under differing representations of the environment, significantly outperforming alternative methods to manage the team.

Andrew Fuchs, Andrea Passarella, Marco Conti

Given an increasing prevalence of intelligent systems capable of autonomous actions or augmenting human activities, it is important to consider scenarios in which the human, autonomous system, or both can exhibit failures as a result of one of several contributing factors (e.g. perception). Failures for either humans or autonomous agents can lead to simply a reduced performance level, or a failure can lead to something as severe as injury or death. For our topic, we consider the hybrid human-AI teaming case where a managing agent is tasked with identifying when to perform a delegation assignment and whether the human or autonomous system should gain control. In this context, the manager will estimate its best action based on the likelihood of either (human, autonomous) agent failure as a result of their sensing capabilities and possible deficiencies. We model how the environmental context can contribute to, or exacerbate, the sensing deficiencies. These contexts provide cases where the manager must learn to attribute capabilities to suitability for decision-making. As such, we demonstrate how a Reinforcement Learning (RL) manager can correct the context-delegation association and assist the hybrid team of agents in outperforming the behavior of any agent working in isolation.

Alessio Mora, Lorenzo Valerio, Paolo Bellavista et al.

Federated Learning (FL) systems enable the collaborative training of machine learning models without requiring centralized collection of individual data. FL participants should have the ability to exercise their right to be forgotten, ensuring their past contributions can be removed from the learned model upon request. In this paper, we propose FedQUIT, a novel algorithm that uses knowledge distillation to scrub the contribution of the data to forget from an FL global model while preserving its generalization ability. FedQUIT directly works on client devices that request to leave the federation, and leverages a teacher-student framework. The FL global model acts as the teacher, and the local model works as the student. To induce forgetting, FedQUIT tailors the teacher's output on local data (the data to forget) penalizing the prediction score of the true class. Unlike previous work, our method does not require hardly viable assumptions for cross-device settings, such as storing historical updates of participants or requiring access to proxy datasets. Experimental results on various datasets and model architectures demonstrate that (i) FedQUIT outperforms state-of-the-art competitors in forgetting data, (ii) has the exact computational requirements as a regular FedAvg round, and (iii) reduces the cumulative communication costs by up to 117.6$\times$ compared to retraining from scratch to restore the initial generalization performance after unlearning.

Mirko Nardi, Lorenzo Valerio, Andrea Passarella

Federated Learning (FL) enables decentralized machine learning while preserving data privacy, making it ideal for sensitive applications where data cannot be shared. While FL has been widely studied in supervised contexts, its application to unsupervised learning remains underdeveloped. This work introduces FedCRef, a novel unsupervised federated learning method designed to uncover all underlying data distributions across decentralized clients without requiring labels. This task, known as Federated Clustering, presents challenges due to heterogeneous, non-uniform data distributions and the lack of centralized coordination. Unlike previous methods that assume a one-cluster-per-client setup or require prior knowledge of the number of clusters, FedCRef generalizes to multi-cluster-per-client scenarios. Clients iteratively refine their data partitions while discovering all distinct distributions in the system. The process combines local clustering, model exchange and evaluation via reconstruction error analysis, and collaborative refinement within federated groups of similar distributions to enhance clustering accuracy. Extensive evaluations on four public datasets (EMNIST, KMNIST, Fashion-MNIST and KMNIST49) show that FedCRef successfully identifies true global data distributions, achieving an average local accuracy of up to 95%. The method is also robust to noisy conditions, scalable, and lightweight, making it suitable for resource-constrained edge devices.

Elisabetta Biondi, Chiara Boldrini, Andrea Passarella et al.

Online social networks (OSNs) have transformed the way individuals fulfill their social needs and consume information. As OSNs become increasingly prominent sources for news dissemination, individuals often encounter content that influences their opinions through both direct interactions and broader network dynamics. In this paper, we propose the Friedkin-Johnsen on Cascade (FJC) model, which is, to the best of our knowledge, is the first attempt to integrate information cascades and opinion dynamics, specifically using the very popular Friedkin-Johnsen model. Our model, validated over real social cascades, highlights how the convergence of socialization and sharing news on these platforms can disrupt opinion evolution dynamics typically observed in offline settings. Our findings demonstrate that these cascades can amplify the influence of central opinion leaders, making them more resistant to divergent viewpoints, even when challenged by a critical mass of dissenting opinions. This research underscores the importance of understanding the interplay between social dynamics and information flow in shaping public discourse in the digital age.

Adnan Ahmad, Chiara Boldrini, Lorenzo Valerio et al.

Decentralized Federated Learning (DFL) is a serverless collaborative machine learning paradigm where devices collaborate directly with neighbouring devices to exchange model information for learning a generalized model. However, variations in individual experiences and different levels of device interactions lead to data and model initialization heterogeneities across devices. Such heterogeneities leave variations in local model parameters across devices that leads to slower convergence. This paper tackles the data and model heterogeneity by explicitly addressing the parameter level varying evidential credence across local models. A novel aggregation approach is introduced that captures these parameter variations in local models and performs robust aggregation of neighbourhood local updates. Specifically, consensus weights are generated via approximation of second-order information of local models on their local datasets. These weights are utilized to scale neighbourhood updates before aggregating them into global neighbourhood representation. In extensive experiments with computer vision tasks, the proposed approach shows strong generalizability of local models at reduced communication costs.

Francesco Di Cursi, Chiara Boldrini, Marco Conti et al.

We evaluate large language models (LLMs) for automatic personality prediction from text under the binary Five Factor Model (BIG5). Five models -- including GPT-4 and lightweight open-source alternatives -- are tested across three heterogeneous datasets (Essays, MyPersonality, Pandora) and two prompting strategies (minimal vs. enriched with linguistic and psychological cues). Enriched prompts reduce invalid outputs and improve class balance, but also introduce a systematic bias toward predicting trait presence. Performance varies substantially: Openness and Agreeableness are relatively easier to detect, while Extraversion and Neuroticism remain challenging. Although open-source models sometimes approach GPT-4 and prior benchmarks, no configuration yields consistently reliable predictions in zero-shot binary settings. Moreover, aggregate metrics such as accuracy and macro-F1 mask significant asymmetries, with per-class recall offering clearer diagnostic value. These findings show that current out-of-the-box LLMs are not yet suitable for APPT, and that careful coordination of prompt design, trait framing, and evaluation metrics is essential for interpretable results.

Giacomo Lanciano, Filippo Galli, Tommaso Cucinotta et al.

Cloud auto-scaling mechanisms are typically based on reactive automation rules that scale a cluster whenever some metric, e.g., the average CPU usage among instances, exceeds a predefined threshold. Tuning these rules becomes particularly cumbersome when scaling-up a cluster involves non-negligible times to bootstrap new instances, as it happens frequently in production cloud services. To deal with this problem, we propose an architecture for auto-scaling cloud services based on the status in which the system is expected to evolve in the near future. Our approach leverages on time-series forecasting techniques, like those based on machine learning and artificial neural networks, to predict the future dynamics of key metrics, e.g., resource consumption metrics, and apply a threshold-based scaling policy on them. The result is a predictive automation policy that is able, for instance, to automatically anticipate peaks in the load of a cloud application and trigger ahead of time appropriate scaling actions to accommodate the expected increase in traffic. We prototyped our approach as an open-source OpenStack component, which relies on, and extends, the monitoring capabilities offered by Monasca, resulting in the addition of predictive metrics that can be leveraged by orchestration components like Heat or Senlin. We show experimental results using a recurrent neural network and a multi-layer perceptron as predictor, which are compared with a simple linear regression and a traditional non-predictive auto-scaling policy. However, the proposed framework allows for the easy customization of the prediction policy as needed.

Luigi Palmieri, Chiara Boldrini, Lorenzo Valerio et al.

Fully decentralized learning is gaining momentum for training AI models at the Internet's edge, addressing infrastructure challenges and privacy concerns. In a decentralized machine learning system, data is distributed across multiple nodes, with each node training a local model based on its respective dataset. The local models are then shared and combined to form a global model capable of making accurate predictions on new data. Our exploration focuses on how different types of network structures influence the spreading of knowledge - the process by which nodes incorporate insights gained from learning patterns in data available on other nodes across the network. Specifically, this study investigates the intricate interplay between network structure and learning performance using three network topologies and six data distribution methods. These methods consider different vertex properties, including degree centrality, betweenness centrality, and clustering coefficient, along with whether nodes exhibit high or low values of these metrics. Our findings underscore the significance of global centrality metrics (degree, betweenness) in correlating with learning performance, while local clustering proves less predictive. We highlight the challenges in transferring knowledge from peripheral to central nodes, attributed to a dilution effect during model aggregation. Additionally, we observe that central nodes exert a pull effect, facilitating the spread of knowledge. In examining degree distribution, hubs in Barabasi-Albert networks positively impact learning for central nodes but exacerbate dilution when knowledge originates from peripheral nodes. Finally, we demonstrate the formidable challenge of knowledge circulation outside of segregated communities.

Lorenzo Valerio, Chiara Boldrini, Andrea Passarella et al.

Federated Learning (FL) is a well-known framework for successfully performing a learning task in an edge computing scenario where the devices involved have limited resources and incomplete data representation. The basic assumption of FL is that the devices communicate directly or indirectly with a parameter server that centrally coordinates the whole process, overcoming several challenges associated with it. However, in highly pervasive edge scenarios, the presence of a central controller that oversees the process cannot always be guaranteed, and the interactions (i.e., the connectivity graph) between devices might not be predetermined, resulting in a complex network structure. Moreover, the heterogeneity of data and devices further complicates the learning process. This poses new challenges from a learning standpoint that we address by proposing a communication-efficient Decentralised Federated Learning (DFL) algorithm able to cope with them. Our solution allows devices communicating only with their direct neighbours to train an accurate model, overcoming the heterogeneity induced by data and different training histories. Our results show that the resulting local models generalise better than those trained with competing approaches, and do so in a more communication-efficient way.

Andrew Fuchs, Andrea Passarella, Marco Conti

When humans and autonomous systems operate together as what we refer to as a hybrid team, we of course wish to ensure the team operates successfully and effectively. We refer to team members as agents. In our proposed framework, we address the case of hybrid teams in which, at any time, only one team member (the control agent) is authorized to act as control for the team. To determine the best selection of a control agent, we propose the addition of an AI manager (via Reinforcement Learning) which learns as an outside observer of the team. The manager learns a model of behavior linking observations of agent performance and the environment/world the team is operating in, and from these observations makes the most desirable selection of a control agent. We restrict the manager task by introducing a set of constraints. The manager constraints indicate acceptable team operation, so a violation occurs if the team enters a condition which is unacceptable and requires manager intervention. To ensure minimal added complexity or potential inefficiency for the team, the manager should attempt to minimize the number of times the team reaches a constraint violation and requires subsequent manager intervention. Therefore our manager is optimizing its selection of authorized agents to boost overall team performance while minimizing the frequency of manager intervention. We demonstrate our manager performance in a simulated driving scenario representing the case of a hybrid team of agents composed of a human driver and autonomous driving system. We perform experiments for our driving scenario with interfering vehicles, indicating the need for collision avoidance and proper speed control. Our results indicate a positive impact of our manager, with some cases resulting in increased team performance up to ~187% that of the best solo agent performance.

Luigi Palmieri, Chiara Boldrini, Lorenzo Valerio et al.

In the vibrant landscape of AI research, decentralised learning is gaining momentum. Decentralised learning allows individual nodes to keep data locally where they are generated and to share knowledge extracted from local data among themselves through an interactive process of collaborative refinement. This paradigm supports scenarios where data cannot leave local nodes due to privacy or sovereignty reasons or real-time constraints imposing proximity of models to locations where inference has to be carried out. The distributed nature of decentralised learning implies significant new research challenges with respect to centralised learning. Among them, in this paper, we focus on robustness issues. Specifically, we study the effect of nodes' disruption on the collective learning process. Assuming a given percentage of "central" nodes disappear from the network, we focus on different cases, characterised by (i) different distributions of data across nodes and (ii) different times when disruption occurs with respect to the start of the collaborative learning task. Through these configurations, we are able to show the non-trivial interplay between the properties of the network connecting nodes, the persistence of knowledge acquired collectively before disruption or lack thereof, and the effect of data availability pre- and post-disruption. Our results show that decentralised learning processes are remarkably robust to network disruption. As long as even minimum amounts of data remain available somewhere in the network, the learning process is able to recover from disruptions and achieve significant classification accuracy. This clearly varies depending on the remaining connectivity after disruption, but we show that even nodes that remain completely isolated can retain significant knowledge acquired before the disruption.

Andrew Fuchs, Andrea Passarella, Marco Conti

We anticipate increased instances of humans and AI systems working together in what we refer to as a hybrid team. The increase in collaboration is expected as AI systems gain proficiency and their adoption becomes more widespread. However, their behavior is not error-free, making hybrid teams a very suitable solution. As such, we consider methods for improving performance for these teams of humans and AI systems. For hybrid teams, we will refer to both the humans and AI systems as agents. To improve team performance over that seen for agents operating individually, we propose a manager which learns, through a standard Reinforcement Learning scheme, how to best delegate, over time, the responsibility of taking a decision to any of the agents. We further guide the manager's learning so they also minimize how many changes in delegation are made resulting from undesirable team behavior. We demonstrate the optimality of our manager's performance in several grid environments which include failure states which terminate an episode and should be avoided. We perform our experiments with teams of agents with varying degrees of acceptable risk, in the form of proximity to a failure state, and measure the manager's ability to make effective delegation decisions with respect to its own risk-based constraints, then compare these to the optimal decisions. Our results show our manager can successfully learn desirable delegations which result in team paths near/exactly optimal with respect to path length and number of delegations.

Samuele Sabella, Chiara Boldrini, Lorenzo Valerio et al.

Decentralized federated learning (DFL) enables devices to collaboratively train models over complex network topologies without relying on a central controller. In this setting, local data remains private, but its quality and quantity can vary significantly across nodes. The extent to which a fully decentralized system is vulnerable to poor-quality or corrupted data remains unclear, but several factors could contribute to potential risks. Without a central authority, there can be no unified mechanism to detect or correct errors, and each node operates with a localized view of the data distribution, making it difficult for the node to assess whether its perspective aligns with the true distribution. Moreover, models trained on low-quality data can propagate through the network, amplifying errors. To explore the impact of low-quality data on DFL, we simulate two scenarios with degraded data quality -- one where the corrupted data is evenly distributed in a subset of nodes and one where it is concentrated on a single node -- using a decentralized implementation of FedAvg. Our results reveal that averaging-based decentralized learning is remarkably robust to localized bad data, even when the corrupted data resides in the most influential nodes of the network. Counterintuitively, this robustness is further enhanced when the corrupted data is concentrated on a single node, regardless of its centrality in the communication network topology. This phenomenon is explained by the averaging process, which ensures that no single node -- however central -- can disproportionately influence the overall learning process.

Lorenzo Valerio, Raffaele Bruno, Andrea Passarella

The widespread diffusion of mobile phones is triggering an exponential growth of mobile data traffic that is likely to cause, in the near future, considerable traffic overload issues even in last-generation cellular networks. Offloading part of the traffic to other networks is considered a very promising approach and, in particular, in this paper, we consider offloading through opportunistic networks of users' devices. However, the performance of this solution strongly depends on the pattern of encounters between mobile nodes, which should therefore be taken into account when designing offloading control algorithms. In this paper, we propose an adaptive offloading solution based on the Reinforcement Learning framework and we evaluate and compare the performance of two well-known learning algorithms: Actor-Critic and Q-Learning. More precisely, in our solution the controller of the dissemination process, once trained, is able to select a proper number of content replicas to be injected into the opportunistic network to guarantee the timely delivery of contents to all interested users. We show that our system based on Reinforcement Learning is able to automatically learn a very efficient strategy to reduce the traffic on the cellular network, without relying on any additional context information about the opportunistic network. Our solution achieves a higher level of offloading with respect to other state-of-the-art approaches, in a range of different mobility settings. Moreover, we show that a more refined learning solution, based on the Actor-Critic algorithm, is significantly more efficient than a simpler solution based on Q-learning.

Matteo Mordacchini, Andrea Passarella, Marco Conti

It is commonly agreed that data will be one of the cornerstones of Future Internet systems. In this context, mobile Opportunistic Networks (ONs) are one of the key paradigms to support, in a self-organising and decentralised manner, the growth of data generated by localized interactions between users mobile devices, and between them and nearby devices such as IoT nodes. In ONs, the spontaneous collaboration among mobile devices is exploited to disseminate data toward interested users. However, the limited resources and knowledge available at each node, and the vast amount of data available, make it difficult to devise efficient schemes to accomplish this task. Recent solutions propose to equip each device with data filtering methods derived from human data processing schemes, known as Cognitive Heuristics, i.e. very effective methods used by the brain to quickly drop useless information, while keeping the most relevant one. These solutions can become less effective when facing dynamic scenarios or situations where nodes cannot fully collaborate. One of the reasons is that the solutions proposed so far do not take take into account the social structure of the environment where the nodes move in. To be more effective, the selection of information performed by each node should take into consideration this dimension of the environment. In this paper we propose a social-based data dissemination scheme, based on the cognitive Social Circle Heuristic. This evaluation method exploits the structure of the social environment to make inferences about the relevance of discovered information. We show how the Social Circle Heuristic, coupled with a cognitive-based community detection scheme, can be exploited to design an effective data dissemination algorithm for ONs. We provide a detailed analysis of the performance of the proposed solution via simulation.

Matteo Mordacchini, Marco Conti, Andrea Passarella et al.

Data management using Device-to-Device (D2D) communications and opportunistic networks (ONs) is one of the main focuses of human-centric pervasive Internet services. In the recently proposed "Internet of People" paradigm, accessing relevant data dynamically generated in the environment nearby is one of the key services. Moreover, personal mobile devices become proxies of their human users while exchanging data in the cyber world and, thus, largely use ONs and D2D communications for exchanging data directly. Recently, researchers have successfully demonstrated the viability of embedding human cognitive schemes in data dissemination algorithms for ONs. In this paper, we consider one such scheme based on the recognition heuristic, a human decision-making scheme used to efficiently assess the relevance of data. While initial evidence about its effectiveness is available, the evaluation of its behaviour in large-scale settings is still unsatisfactory. To overcome these limitations, we have developed a novel hybrid modelling methodology, which combines an analytical model of data dissemination within small-scale communities of mobile users, with detailed simulations of interactions between different communities. This methodology allows us to evaluate the algorithm in large-scale city- and country-wide scenarios. Results confirm the effectiveness of cognitive data dissemination schemes, even when content popularity is very heterogenous.

Lorenzo Valerio, Andrea Passarella, Marco Conti

Due to the pervasive diffusion of personal mobile and IoT devices, many ``smart environments'' (e.g., smart cities and smart factories) will be, among others, generators of huge amounts of data. Currently, this is typically achieved through centralised cloud-based data analytics services. However, according to many studies, this approach may present significant issues from the standpoint of data ownership, and even wireless network capacity. One possibility to cope with these shortcomings is to move data analytics closer to where data is generated. In this paper, we tackle this issue by proposing and analyzing a distributed learning framework, whereby data analytics are performed at the edge of the network, i.e., on locations very close to where data is generated. Specifically, in our framework, partial data analytics are performed directly on the nodes that generate the data, or on nodes close by (e.g., some of the data generators can take this role on behalf of subsets of other nodes nearby). Then, nodes exchange partial models and refine them accordingly. Our framework is general enough to host different analytics services. In the specific case analysed in the paper, we focus on a learning task, considering two distributed learning algorithms. Using an activity recognition and a pattern recognition task, both on reference datasets, we compare the two learning algorithms between each other and with a central cloud solution (i.e., one that has access to the complete datasets). Our results show that using distributed machine learning techniques, it is possible to drastically reduce the network overhead, while obtaining performance comparable to the cloud solution in terms of learning accuracy. The analysis also shows when each distributed learning approach is preferable, based on the specific distribution of the data on the nodes.

Lorenzo Valerio, Marco Conti, Andrea Passarella

Due to the pervasive diffusion of personal mobile and IoT devices, many "smart environments" (e.g., smart cities and smart factories) will be, generators of huge amounts of data. Currently, analysis of this data is typically achieved through centralised cloud-based services. However, according to many studies, this approach may present significant issues from the standpoint of data ownership, as well as wireless network capacity. In this paper, we exploit the fog computing paradigm to move computation close to where data is produced. We exploit a well-known distributed machine learning framework (Hypothesis Transfer Learning), and perform data analytics on mobile nodes passing by IoT devices, in addition to fog gateways at the edge of the network infrastructure. We analyse the performance of different configurations of the distributed learning framework, in terms of (i) accuracy obtained in the learning task and (ii) energy spent to send data between the involved nodes. Specifically, we consider reference wireless technologies for communication between the different types of nodes we consider, e.g. LTE, Nb-IoT, 802.15.4, 802.11, etc. Our results show that collecting data through the mobile nodes and executing the distributed analytics using short-range communication technologies, such as 802.15.4 and 802.11, allows to strongly reduce the energy consumption of the system up to $94\%$ with a loss in accuracy w.r.t. a centralised cloud solution up to $2\%$.

Mustafa Toprak, Chiara Boldrini, Andrea Passarella et al.

Being able to recommend links between users in online social networks is important for users to connect with like-minded individuals as well as for the platforms themselves and third parties leveraging social media information to grow their business. Predictions are typically based on unsupervised or supervised learning, often leveraging simple yet effective graph topological information, such as the number of common neighbors. However, we argue that richer information about personal social structure of individuals might lead to better predictions. In this paper, we propose to leverage well-established social cognitive theories to improve link prediction performance. According to these theories, individuals arrange their social relationships along, on average, five concentric circles of decreasing intimacy. We postulate that relationships in different circles have different importance in predicting new links. In order to validate this claim, we focus on popular feature-extraction prediction algorithms (both unsupervised and supervised) and we extend them to include social-circles awareness. We validate the prediction performance of these circle-aware algorithms against several benchmarks (including their baseline versions as well as node-embedding- and GNN-based link prediction), leveraging two Twitter datasets comprising a community of video gamers and generic users. We show that social-awareness generally provides significant improvements in the prediction performance, beating also state-of-the-art solutions like node2vec and SEAL, and without increasing the computational complexity. Finally, we show that social-awareness can be used in place of using a classifier (which may be costly or impractical) for targeting a specific category of users.

Lorenzo Valerio, Andrea Passarella, Marco Conti

The exponential growth of devices and data at the edges of the Internet is rising scalability and privacy concerns on approaches based exclusively on remote cloud platforms. Data gravity, a fundamental concept in Fog Computing, points towards decentralisation of computation for data analysis, as a viable alternative to address those concerns. Decentralising AI tasks on several cooperative devices means identifying the optimal set of locations or Collection Points (CP for short) to use, in the continuum between full centralisation (i.e., all data on a single device) and full decentralisation (i.e., data on source locations). We propose an analytical framework able to find the optimal operating point in this continuum, linking the accuracy of the learning task with the corresponding network and computational cost for moving data and running the distributed training at the CPs. We show through simulations that the model accurately predicts the optimal trade-off, quite often an intermediate point between full centralisation and full decentralisation, showing also a significant cost saving w.r.t. both of them. Finally, the analytical model admits closed-form or numeric solutions, making it not only a performance evaluation instrument but also a design tool to configure a given distributed learning task optimally before its deployment.

Lorenzo Valerio, Franco Maria Nardini, Andrea Passarella et al.

Neural Networks (NN), although successfully applied to several Artificial Intelligence tasks, are often unnecessarily over-parametrised. In edge/fog computing, this might make their training prohibitive on resource-constrained devices, contrasting with the current trend of decentralising intelligence from remote data centres to local constrained devices. Therefore, we investigate the problem of training effective NN models on constrained devices having a fixed, potentially small, memory budget. We target techniques that are both resource-efficient and performance effective while enabling significant network compression. Our Dynamic Hard Pruning (DynHP) technique incrementally prunes the network during training, identifying neurons that marginally contribute to the model accuracy. DynHP enables a tunable size reduction of the final neural network and reduces the NN memory occupancy during training. Freed memory is reused by a \emph{dynamic batch sizing} approach to counterbalance the accuracy degradation caused by the hard pruning strategy, improving its convergence and effectiveness. We assess the performance of DynHP through reproducible experiments on three public datasets, comparing them against reference competitors. Results show that DynHP compresses a NN up to $10$ times without significant performance drops (up to $3.5\%$ additional error w.r.t. the competitors), reducing up to $80\%$ the training memory occupancy.