Network Traffic Classification (NTC) increasingly relies on data-driven models, yet its practical deployment is often constrained by limited labeled data, strict privacy requirements, and the cost of collecting representative traffic traces. While Network Traffic Generation (NTG) provides an effective means to mitigate data scarcity, conventional generative methods struggle to model the complex temporal dynamics of modern traffic and often incur high computational costs. In this article, we investigate lightweight Generative Artificial Intelligence (GenAI) architectures for practical NTG. Rather than generating raw packet bytes or relying on large foundation models, we synthesize compact flow-level traffic representations derived from early packet-header information, enabling transformer-based, state-space, and diffusion models with only a few million parameters. We present a modular GenAI pipeline for NTG and evaluate it along four complementary axes: (i) synthetic traffic fidelity, (ii) synthetic-only training for privacy-preserving NTC, (iii) data augmentation under low-data regimes, and (iv) computational efficiency. Experiments on two heterogeneous datasets show that lightweight transformer-based and state-space models preserve both static and temporal traffic characteristics, while providing useful synthetic data for downstream NTC. Among them, transformer-based models offer the best fidelity-efficiency trade-off, combining high-quality traffic generation with moderate computational overhead.

Localizing knowledge within Large Language Models (LLMs) is crucial for interpreting their mechanisms and outcomes. Whereas knowledge attribution has so far provided local sample-level explanations, in this work we argue that whenever LLMs are used for classification tasks, a class-level explanation is preferable. We therefore define broader concept areas, i.e., regions of the LLM comprising a small set of neurons that contains the most salient knowledge pertaining to each class and propose methods to identify such areas. We apply our methodology to BERT-based LLMs fine-tuned for downstream classification tasks such as sentiment analysis and attack classification: our results show that it is possible to (i) identify crucial sets of neurons that determine the behaviour of fine-tuned LLMs for explanation purposes, as well as (ii) exploit such concept areas to improve their classification outcomes-yielding up to 6% macro F1-Score improvement on sentiment analysis (public dataset) and 2% on attack classification (private dataset) without requiring further fine-tuning.

Troubleshooting WiFi networks is knowingly difficult due to the variability of the wireless medium. Complementary to existing works that focus on detecting short-term fluctuations of radio signals (i.e., anomalies), we tackle the problem of reliably detecting long-term changes in statistical properties of WiFi networks. We propose a new method to reliably gain insights on such environmental changes, which we refer to as Random Access Point Forest (RAPF). RAPF identifies the changes from a forest of individual learners, each of them consisting of a random tree approximating the signal of a specific pair of APs. The biased selection of APs in a distributed manner along with the stochastic construction of each individual tree ensure its robustness to noise and biases. We conduct a measurement campaign on a real WLAN by collecting the path loss among pairs of APs in a network for which labels are available and perform an extensive comparison of our methodology against state-of-the-art change point methodologies, which conclusively shows RAPF to yield the most robust detection capabilities.

In Wireless Local Area Networks (WLANs), Access point (AP) transmit power influences (i) received signal quality for users and thus user throughput, (ii) user association and thus load across APs and (iii) AP coverage ranges and thus interference in the network. Despite decades of academic research, transmit power levels are still, in practice, statically assigned to satisfy uniform coverage objectives. Yet each network comes with its unique distribution of users in space, calling for a power control that adapts to users' probabilities of presence, for example, placing the areas with higher interference probabilities where user density is the lowest. Although nice on paper, putting this simple idea in practice comes with a number of challenges, with gains that are difficult to estimate, if any at all. This paper is the first to address these challenges and evaluate in a production network serving thousands of daily users the benefits of a user-aware transmit power control system. Along the way, we contribute a novel approach to reason about user densities of presence from historical IEEE 802.11k data, as well as a new machine learning approach to impute missing signal-strength measurements. Results of a thorough experimental campaign show feasibility and quantify the gains: compared to state-of-the-art solutions, the new system can increase the median signal strength by 15dBm, while decreasing airtime interference at the same time. This comes at an affordable cost of a 5dBm decrease in uplink signal due to lack of terminal cooperation.

Over the last years we witnessed a renewed interest toward Traffic Classification (TC) captivated by the rise of Deep Learning (DL). Yet, the vast majority of TC literature lacks code artifacts, performance assessments across datasets and reference comparisons against Machine Learning (ML) methods. Among those works, a recent study from IMC'22 [16] is worth of attention since it adopts recent DL methodologies (namely, few-shot learning, self-supervision via contrastive learning and data augmentation) appealing for networking as they enable to learn from a few samples and transfer across datasets. The main result of [16] on the UCDAVIS, ISCXVPN and ISCXTOR datasets is that, with such DL methodologies, 100 input samples are enough to achieve very high accuracy using an input representation called "flowpic'' (i.e., a per-flow 2d histograms of the packets size evolution over time). In this paper (i) we reproduce[16] on the same datasets and (ii) we replicate its most salient aspect (the importance of data augmentation) on three additional public datasets (MIRAGEA, MIRAGEB and UTMOBILENET). While we confirm most of the original results, we also found a ≈ 20% accuracy drop on some of the investigated scenarios due to a data shift in the original dataset that we uncovered. Additionally, our study validates that the data augmentation strategies studied in[16] perform well on other datasets too. In the spirit of reproducibility and replicability we make all artifacts (code and data) available to the research community at https://tcbenchstack.github.io/tcbench/.

Machine learning is increasingly used to automate networking tasks, in a paradigm known as zero touch network and service management (ZSM). In particular, deep reinforcement learning (DRL) techniques have recently gained much attention for their ability to learn taking complex decisions in different fields. In the ZSM context, DRL is an appealing candidate for tasks such as dynamic resource allocation, which are generally formulated as hard optimization problems. At the same time, successful training and deployment of DRL agents in real-world scenarios face a number of challenges that we outline and address in this article. Tackling the case of wireless local area network radio resource management, we report guidelines that extend to other usecases and more general contexts.

Recent years have witnessed the promise that reinforcement learning, coupled with Graph Neural Network (GNN) architectures, could learn to solve hard combinatorial optimization problems: given raw input data and an evaluator to guide the process, the idea is to automatically learn a policy able to return feasible and high-quality outputs. Recent work have shown promising results but the latter were mainly evaluated on the travelling salesman problem (TSP) and similar abstract variants such as Split Delivery Vehicle Routing Problem (SDVRP). In this paper, we analyze how and whether recent neural architectures can be applied to graph problems of practical importance. We thus set out to systematically "transfer" these architectures to the Power and Channel Allocation Problem (PCAP), which has practical relevance for, e.g., radio resource allocation in wireless networks. Our experimental results suggest that existing architectures (i) are still incapable of capturing graph structural features and (ii) are not suitable for problems where the actions on the graph change the graph attributes. On a positive note, we show that augmenting the structural representation of problems with Distance Encoding is a promising step towards the still-ambitious goal of learning multi-purpose autonomous solvers.

Airtime interference is a key performance indicator for WLANs, measuring, for a given time period, the percentage of time during which a node is forced to wait for other transmissions before to transmitting or receiving. Being able to accurately estimate interference resulting from a given state change (e.g., channel, bandwidth, power) would allow a better control of WLAN resources, assessing the impact of a given configuration before actually implementing it. In this paper, we adopt a principled approach to interference estimation in WLANs. We first use real data to characterize the factors that impact it, and derive a set of relevant synthetic workloads for a controlled comparison of various deep learning architectures in terms of accuracy, generalization and robustness to outlier data. We find, unsurprisingly, that Graph Convolutional Networks (GCNs) yield the best performance overall, leveraging the graph structure inherent to campus WLANs. We notice that, unlike e.g. LSTMs, they struggle to learn the behavior of specific nodes, unless given the node indexes in addition. We finally verify GCN model generalization capabilities, by applying trained models on operational deployments unseen at training time.

With the advent of 5G and the evolution of Internet protocols, industrial applications are moving from vertical solutions to general purpose IP-based infrastructures that need to meet deterministic Quality of Service (QoS) requirements. The IETF DetNet working group aims at providing an answer to this need with support for (i) deterministic worst-case latency and jitter, and (ii) zero packet loss for time-sensitive traffic. In this paper we focus on the joint routing and scheduling problem in large-scale deterministic networks using Cycle Specified Queuing and Forwarding (CSQF), an extension of Cyclic Queuing and Forwarding (CQF) with multiple transmission queues and support of segment routing. In this context, we present two centralized algorithms to maximize traffic acceptance for network planning and online flow admission. We propose an effective solution based on column generation and dynamic programming. Thanks to the reinforcement of the model with valid inequalities, we improve the upper bound and the solution. We demonstrate on realistic instances that we reach an optimality gap smaller than 10% in a few seconds. Finally, we also derive an ultra-fast adaptive greedy algorithm to solve the problem at the cost of a small extra gap.

Today's WLANs rely on a centralized Access Controller (AC) entity for managing distributed wireless Access Points (APs) to which user devices connect. The availability of real-time analytics at the AC opens the possibility to automate the allocation of scarce radio resources, continuously adapting to changes in traffic demands. Often, the allocation problem is formulated in terms of weighted graph coloring, which is NP-hard, and custom heuristics are used to find satisfactory solutions. In this paper, we contrast solutions that are based on (and even improve) state of the art heuristics to a data-driven solution that leverages Deep Reinforcement Learning (DRL). Based on both simulation results as well as experiments in a real deployment, we show that our DRL-based scheme not only learns to solve the complex combinatorial problem in bounded time, outperforming heuristics, but it also exhibits appealing generalization properties, e.g. to different network sizes and densities.

Today’s Wireless Local Area Networks (WLANs) rely on a centralized Access Controller (AC) entity for managing a fleet of Access Points (APs). Real-time analytics enable the AC to optimize the radio resource allocation (i.e. channels) online in response to sudden traffic shifts. Deep Reinforcement Learning (DRL) relieves the pressure of finding good optimization heuristics by learning a policy through interactions with the environment. However, it is not granted that DRL will behave well in unseen conditions. Tools such as the WiFi Dynoscope introduced here are necessary to gain this trust. In a nutshell, this demo dissects the dynamics of WLAN networks, both simulated and from real large-scale deployments, by (i) comparatively analyzing the performance of different algorithms on the same deployment at high level and (ii) getting low-level details and insights into algorithmic behaviour.

—Deep Reinforcement Learning (DRL) techniques have recently gathered much attention for their ability to learn taking complex decisions in different fields: as such, they are an appealing candidate for network Operation and Management (O&M). In particular, DRL can become a fundamental item in the toolbox of the so called “self-driving networks”, especially for tasks such as dynamic resource allocation, that is generally formulated and solved as complex optimization problems. Yet, training and deployment of DRL agents in real-world scenarios face important challenges, that we illustrate in this article using Wireless LANs as a relevant deployment example

Caching popular content at the wireless edge promises performance benefits as well as business perspectives. In this paper, we study the following arrangement: a mobile network operator (MNO) pre-installs memory on its wireless equipment, sets a price, and invites a unique content provider (CP) to invest. The CP leases memory space and places its content; the MNO then associates network users to stations, aiming for offloading CP traffic or not. We formulate an optimization problem, which maximizes offloading with minimum leasing costs for the CP. This is an NP-hard mixed-integer non-linear optimization problem. We present an iterative exact solution using Generalized Benders decomposition into a content-related master problem and a user-association slave problem. Master is integer linear. Slave is convex for various association policies, including: 1) join-the-closest cache and 2) cache-aware association. For slave, we introduce a distributed exact solution, named generalized bucket-filling. Extensive simulations illustrate the performance benefits under different association policies. The solution helps to determine the optimal leasing price for the MNO, and the optimal investment budget for the CP. As a general conclusion, both actors profit when the MNO association supports CP decisions.

Caching popular content at the wireless edge is recently proposed as a means to reduce congestion at the backbone of cellular networks. The two main actors involved are Mobile Network Operators (MNOs) and Content Providers (CPs). In this work, we consider the following arrangement: an MNO pre-installs memory on its wireless equipment (e.g. Base Stations) and invites a unique CP to use them, with monetary cost. The CP will lease memory space and place its content; the MNO will associate network users to stations. For a given association policy, the MNO may help (or not) the CP to offload traffic, depending on whether the association takes into account content placement. We formulate an optimization problem from the CP perspective, which aims at maximizing traffic offloading with minimum leasing costs. This is a joint optimization problem that can include any association policy, and can also derive the optimal one. We present a general exact solution using Benders decomposition. It iteratively updates decisions of the two actors separately and converges to the global optimum. We illustrate the optimal CP leasing/placement strategy and hit probability gains under different association policies. Performance is maximised when the MNO association follows CP actions.

The massive increase in cellular traffic poses serious challenges to all actors concerned with wireless content delivery. While network densification provides access to additional users, high-speed and high-capacity backhaul connections are expensive. Caching popular content at the network edge promises to offload user traffic from these congestion prone connections as well as from the data centers in the backbone network. This thesis proposes a business model in which a mobile network operator (MNO) pre-installs and maintains caches at its wireless equipment (Cache-equipped Base Stations, CBSs). Memory space together with computational capabilities is then leased to content providers (CPs) that want to bring their content closer to the user. For a financial compensation, a CP can then offload traffic from its data center and improve user Quality of Service. The CP makes content placement decisions based on predictive user traffic and content popularity data. In the delivery phase, users can be served from the caches in case they are associated to stations that have the requested content cached. This work investigates three aspects of the proposed business model: The first research question focuses on user association as a central element to the edge caching scheme. Cache-aware user association policies can allow for users in coverage overlap areas to be associated to a CBS that holds the requested content rather than conventionally to the one that provides the strongest signal. The thesis proposes an original decentralized algorithm for user association called Generalized Bucket-filling that allows gains beyond maximizing the hit ratio. Performance metrics such as network throughput and load balancing of users among CBSs are taken into account. Experiments show that cache-aware user association a) increases the hit ratio b) without overloading single CBSs while c) providing high system throughput. The second problem treated considers a single CP that needs to decide how much cache space to lease at each CBS for a fixed price, and what content to place. Its choices should be based on estimates of file popularity as well as MNO user association policy. The cache leasing and content placement problem is formulated as a non-linear mixed-integer problem (NLMIP). In its solution, the problem is separated into a linear discrete CP subproblem and a nonlinear continuous subproblem using Benders decomposition. The CP and the MNO cooperate, helping the CP to make optimal decisions that benefit both parties: The CP maximizes its savings from caching while the MNO can find the optimal cache price and receive the maximum financial compensation. A third research question widens the focus to the interaction between several CPs and one MNO. Now, the MNO does not set a fixed price per memory unit but instead reacts to CP demands for memory space that depend on the savings they can achieve from caching.

Caching of popular content on wireless nodes is recently proposed as a means to reduce congestion in the backbone of cellular networks and to improve Quality of Service. From a network point of view, the goal is to offload as many users as possible from the backbone network to the wireless caches while at the same time offering good service to cache-unrelated users. Aggressive offloading can lead to an unbalanced user association. Some wireless nodes can be overloaded by cache-related traffic while the resources of others remain underused. Given a fixed content placement, this work proposes an efficient distributed algorithm to control and balance the association of cache-related traffic among cellular cache memories. The algorithm allows the network to achieve the globally optimal solution and can be executed on base stations using a limited amount of information exchange between them. It is based on a novel algorithm we call Bucket-filling. The solution limits the cache-users per node by balancing the total load among the nodes in a fair way. The improvement compared to common user assignment policies is highlighted for single- as well as for multi-tier random networks.

We investigate the Robust Multiperiod Network Design Problem, a generalization of the classical Capacitated Network Design Problem that additionally considers multiple design periods and provides solutions protected against traffic uncertainty. Given the intrinsic difficulty of the problem, which proves challenging even for state-of-the art commercial solvers, we propose a hybrid primal heuristic based on the combination of ant colony optimization and an exact large neighborhood search. Computational experiments on a set of realistic instances from the SNDlib show that our heuristic can find solutions of extremely good quality with low optimality gap.

Base station cooperation in the downlink of cellular systems has been recently suggested as a promising concept towards a better exploitation of the communication system's physical resources. It may offer a high gain in capacity through interference mitigation. This however, comes at a cost of high information exchange between cooperating entities and a high computational burden. Clustering of base stations into subgroups is an alternative to guarantee such cooperation benefits in a lower scale. The optimal definition of clusters, however, and a systematic way to find a solution to such problem is not yet available. In this work, we highlight the combinatorial nature of the problem, exploit this to describe the system of users and base stations as a graph and formulate a pure 0-1 program. Its solution suggests a cost optimal way to form clusters and assign user subsets to them.

Base station cooperation in the downlink of cellular systems has been recently suggested as a promising concept towards a better exploitation of the communication system physical resources. It may offer a high gain in capacity through interference mitigation. This however, comes at a cost of high information exchange between cooperating entities and a high computational burden. Clustering of base stations into subgroups is an alternative to guarantee such cooperation benefits in a lower scale. The optimal definition of clusters, however, and a systematic way to find a solution to such problem is not yet available. In this work, we highlight the combinatorial nature of the problem, exploit this to describe the system of users and base stations as a graph and formulate a pure 0-1 program. Its solution suggests a cost optimal way to form clusters and assign user subsets to them.