Thrasyvoulos Spyropoulos

Marina Costantini, Giovanni Neglia, Thrasyvoulos Spyropoulos

Federated learning (FL) has enabled training machine learning models exploiting the data of multiple agents without compromising privacy. However, FL is known to be vulnerable to data heterogeneity, partial device participation, and infrequent communication with the server, which are nonetheless three distinctive characteristics of this framework. While much of the recent literature has tackled these weaknesses using different tools, only a few works have explored the possibility of exploiting inter-agent communication to improve FL's performance. In this work, we present FedDec, an algorithm that interleaves peer-to-peer communication and parameter averaging (similar to decentralized learning in networks) between the local gradient updates of FL. We analyze the convergence of FedDec under the assumptions of non-iid data distribution, partial device participation, and smooth and strongly convex costs, and show that inter-agent communication alleviates the negative impact of infrequent communication rounds with the server by reducing the dependence on the number of local updates $H$ from $O(H^2)$ to $O(H)$. Furthermore, our analysis reveals that the term improved in the bound is multiplied by a constant that depends on the spectrum of the inter-agent communication graph, and that vanishes quickly the more connected the network is. We confirm the predictions of our theory in numerical simulations, where we show that FedDec converges faster than FedAvg, and that the gains are greater as either $H$ or the connectivity of the network increase.

Marina Costantini, Nikolaos Liakopoulos, Panayotis Mertikopoulos et al.

In decentralized optimization environments, each agent $i$ in a network of $n$ nodes has its own private function $f_i$, and nodes communicate with their neighbors to cooperatively minimize the aggregate objective $\sum_{i=1}^n f_i$. In this setting, synchronizing the nodes' updates incurs significant communication overhead and computational costs, so much of the recent literature has focused on the analysis and design of asynchronous optimization algorithms, where agents activate and communicate at arbitrary times without needing a global synchronization enforcer. However, most works assume that when a node activates, it selects the neighbor to contact based on a fixed probability (e.g., uniformly at random), a choice that ignores the optimization landscape at the moment of activation. Instead, in this work we introduce an optimization-aware selection rule that chooses the neighbor providing the highest dual cost improvement (a quantity related to a dualization of the problem based on consensus). This scheme is related to the coordinate descent (CD) method with the Gauss-Southwell (GS) rule for coordinate updates; in our setting however, only a subset of coordinates is accessible at each iteration (because each node can communicate only with its neighbors), so the existing literature on GS methods does not apply. To overcome this difficulty, we develop a new analytical framework for smooth and strongly convex $f_i$ that covers the class of set-wise CD algorithms -- a class that directly applies to decentralized scenarios, but is not limited to them -- and we show that the proposed set-wise GS rule achieves a speedup factor of up to the maximum degree in the network (which is in the order of $Θ(n)$ for highly connected graphs). The speedup predicted by our analysis is validated in numerical experiments with synthetic data.

Konstantinos Mylonas, Thrasyvoulos Spyropoulos

Nowadays, social media is the ground for political debate and exchange of opinions. There is a significant amount of research that suggests that social media are highly polarized. A phenomenon that is commonly observed is the echo chamber structure, where users are organized in polarized communities and form connections only with similar-minded individuals, limiting themselves to consume specific content. In this paper we explore a way to decrease the polarization of networks with two echo chambers. Particularly, we observe that if some users adopt a moderate opinion about a topic, the polarization of the network decreases. Based on this observation, we propose an efficient algorithm to identify a good set of K users, such that if they adopt a moderate stance around a topic, the polarization is minimized. Our algorithm employs a Graph Neural Network and thus it can handle large graphs more effectively than other approaches

Theodoros Giannakas, Pavlos Sermpezis, Anastasios Giovanidis et al.

As mobile traffic is dominated by content services (e.g., video), which typically use recommendation systems, the paradigm of network-friendly recommendations (NFR) has been proposed recently to boost the network performance by promoting content that can be efficiently delivered (e.g., cached at the edge). NFR increase the network performance, however, at the cost of being unfair towards certain contents when compared to the standard recommendations. This unfairness is a side effect of NFR that has not been studied in literature. Nevertheless, retaining fairness among contents is a key operational requirement for content providers. This paper is the first to study the fairness in NFR, and design fair-NFR. Specifically, we use a set of metrics that capture different notions of fairness, and study the unfairness created by existing NFR schemes. Our analysis reveals that NFR can be significantly unfair. We identify an inherent trade-off between the network gains achieved by NFR and the resulting unfairness, and derive bounds for this trade-off. We show that existing NFR schemes frequently operate far from the bounds, i.e., there is room for improvement. To this end, we formulate the design of Fair-NFR (i.e., NFR with fairness guarantees compared to the baseline recommendations) as a linear optimization problem. Our results show that the Fair-NFR can achieve high network gains (similar to non-fair-NFR) with little unfairness.

Savvas Kastanakis, Pavlos Sermpezis, Vasileios Kotronis et al.

Joint caching and recommendation has been recently proposed as a new paradigm for increasing the efficiency of mobile edge caching. Early findings demonstrate significant gains for the network performance. However, previous works evaluated the proposed schemes exclusively on simulation environments. Hence, it still remains uncertain whether the claimed benefits would change in real settings. In this paper, we propose a methodology that enables to evaluate joint network and recommendation schemes in real content services by only using publicly available information. We apply our methodology to the YouTube service, and conduct extensive measurements to investigate the potential performance gains. Our results show that significant gains can be achieved in practice; e.g., 8 to 10 times increase in the cache hit ratio from cache-aware recommendations. Finally, we build an experimental testbed and conduct experiments with real users; we make available our code and datasets to facilitate further research. To our best knowledge, this is the first realistic evaluation (over a real service, with real measurements and user experiments) of the joint caching and recommendations paradigm. Our findings provide experimental evidence for the feasibility and benefits of this paradigm, validate assumptions of previous works, and provide insights that can drive future research.

Pavlos Sermpezis, Savvas Kastanakis, João Ismael Pinheiro et al.

In this paper we propose that recommendation systems (RSs) for multimedia services should be "QoS-aware", i.e., take into account the expected QoS with which a content can be delivered, to increase the user satisfaction. Network-aware recommendations have been very recently proposed as a promising solution to improve network performance. However, the idea of QoS-aware RSs has been studied from the network perspective. Its feasibility and performance performance advantages for the content-provider or user perspective have only been speculated. Hence, in this paper we aim to provide initial answers for the feasibility of the concept of QoS-aware RS, by investigating its impact on real user experience. To this end, we conduct experiments with real users on a testbed, and present initial experimental results. Our analysis demonstrates the potential of the idea: QoS-aware RSs could be beneficial for both the users (better experience) and content providers (higher user engagement). Moreover, based on the collected dataset, we build statistical models to (i) predict the user experience as a function of QoS, relevance of recommendations (QoR) and user interest, and (ii) provide useful insights for the design of QoS-aware RSs. We believe that our study is an important first step towards QoS-aware recommendations, by providing experimental evidence for their feasibility and benefits, and can help open a future research direction.

Spyridon Vassilaras, Luigi Vigneri, Nikolaos Liakopoulos et al.

Future 5G wireless networks will rely on agile and automated network management, where the usage of diverse resources must be jointly optimized with surgical accuracy. A number of key wireless network functionalities (e.g., traffic steering, power control) give rise to hard optimization problems. What is more, high spatio-temporal traffic variability coupled with the need to satisfy strict per slice/service SLAs in modern networks, suggest that these problems must be constantly (re-)solved, to maintain close-to-optimal performance. To this end, we propose the framework of Online Network Optimization (ONO), which seeks to maintain both agile and efficient control over time, using an arsenal of data-driven, online learning, and AI-based techniques. Since the mathematical tools and the studied regimes vary widely among these methodologies, a theoretical comparison is often out of reach. Therefore, the important question `what is the right ONO technique?' remains open to date. In this paper, we discuss the pros and cons of each technique and present a direct quantitative comparison for a specific use case, using real data. Our results suggest that carefully combining the insights of problem modeling with state-of-the-art AI techniques provides significant advantages at reasonable complexity.