A5004766899- Complex Network Analysis Techniques
- Network Security and Intrusion Detection
- Opinion Dynamics and Social Influence
- Internet Traffic Analysis and Secure E-voting
- Advanced Malware Detection Techniques
- Distributed Sensor Networks and Detection Algorithms
- Control Systems and Identification
- Neural Networks and Applications
- Neural Networks Stability and Synchronization
- Chaos-based Image/Signal Encryption
- Software-Defined Networks and 5G
- Advanced Graph Neural Networks
- Game Theory and Applications
- Gaussian Processes and Bayesian Inference
- Spam and Phishing Detection
- Health and Well-being Studies
- Gene Regulatory Network Analysis
University of Salerno
2019-2023
Consorzio Nazionale Interuniversitario per le Telecomunicazioni
2021-2022
In a Distributed Denial of Service (DDoS) attack, network (botnet) dispersed agents (bots) sends requests to website saturate its resources. Since the are sent by automata, typical way detect them is look for some repetition pattern or commonalities between same user from different users. For this reason, recent DDoS variants exploit communication layers that offer broader possibility in terms admissible request patterns, such as, e.g., application layer. case, malicious can pick legitimate...
A network attacker wants to transmit Voice-over-IP (VoIP) traffic streams covertly. He tries evade the detection system by manipulating VoIP through padding, shifting, and splitting operations, so as conceal them amidst Internet traffic. defender detect manipulated streams. Tackling this problem from an adversarial perspective, we provide two contributions: 1) obtain a highly stylized representation of in terms transmission frequency F packet length L, characterize (F, L) region achievable...
This work examines a social learning problem, where dispersed agents connected through network topology interact locally to form their opinions ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">beliefs</i> ) as regards certain hypotheses of interest. These evolve over time, since the collect observations from environment, and update current beliefs by accounting for: past beliefs, innovation contained in new data, received neighbors. The...
In social learning, a group of agents linked by graph topology collect data and exchange opinions on some topic interest, represented finite set hypotheses. Traditional learning algorithms allow all in the network to gain full confidence true underlying hypothesis as number observations increases. Under partial information sharing, can only single hypothesis. This introduces significant challenges compared standard case opinion sharing. We propose novel strategy where each agent forms valid...
We consider the problem of identifying members a botnet under an application-layer (L7) DDoS attack, where target site is flooded with large number requests that emulate legitimate users' patterns. This challenging has been recently addressed reference to two simplified scenarios, either all bots pick from same emulation dictionary (total overlap), or they are divided in separate clusters corresponding distinct dictionaries (no overlap at all). However, over real networks these extreme...
This work examines the problem of learning topology a network (graph learning) from signals produced at subset nodes (partial observability). challenging was recently tackled assuming that is drawn according to an Erdős-Rényi model, for which it shown graph under partial observability achievable, exploiting in particular homogeneity across and independence edges. However, several real-world networks do not match optimistic assumptions homogeneity/independence, example, high het-erogeneity...
This work addresses the problem of learning topology a network from signals emitted by nodes. These are generated over time through linear diffusion process, where neighboring nodes exchange messages according to underlying topology, and aggregate them certain combination matrix. We consider demanding setting graph under partial observability, available only limited fraction nodes, we want establish whether these can be estimated faithfully, despite presence possibly many latent Recent...
Objective: Hyperuricemia can be both determined by overproduction (due to xanthine oxidase hyperactivity) or renal hypoexcretion. In this latter case, also diuretic use (particularly hydrochlorotiazhide) involved. While the relationship between hyperuricemia and CardioVascular (CV) events has been definitively linked in many studies, data on related are still lacking. The objective of analysis is assess induced CV events. Design method: URic acid Right for heArt Health study a nationwide,...
network attacker wants to transmit VoIP traffic streams covertly. He tries evade the detection system by manipulating through padding and shifting operations, so as conceal them amidst Internet traffic. A defender (the system) detect manipulated streams. Tackling this problem from an adversarial perspective, we provide two contributions: i) obtain a highly stylized representation of in terms transmission frequency $\mathcal{F}$ payload $\mathcal{L}$, characterize $(\mathcal{F},\...
This work deals with cyber-threat propagation across a communication network designed according to the network-slicing paradigm. Exploiting multi-dimensional Birth-Death-Immigration model, we examine threat percolation from vulnerable slice virtually secured slice. The analysis quantifies role played by slice-coupling on propagation, revealing how cross-slice attacks can be particularly dangerous in applications where attacker opens door some relative, e.g., ordinary services, breaking...
The detection of encrypted multimedia traffic (like VoIP or Video) is a crucial task for both TELCO operators and authorities involved in lawful interception issues. As an example, Skype that cannot be detected through classical methods as port-based (because random based choice port option) nor payload inspection encryption mechanisms adopted). Dwelling on Skype, the aim this work to propose novel technique that, by recasting regularities data streams terms recurrence plots (a...
This work examines a social learning problem, where dispersed agents connected through network topology interact locally to form their opinions (beliefs) as regards certain hypotheses of interest. These evolve over time, since the collect observations from environment, and update current beliefs by accounting for: past beliefs, innovation contained in new data, received neighbors. The distinguishing feature present is that are constrained share regarding only single hypothesis. We devise...
This work examines the problem of learning a network graph from signals emitted by nodes, according to diffusion model ruled Laplacian combination policy. The challenging regime partial observability is considered, where are collected limited subset and we wish estimate subgraph connections between these probed nodes. For static setting fixed during estimation process, examine sample complexity (number time samples necessary achieve consistent as size grows) Erdős-Rényi Bollobás-Riordan...
In this brief we investigate the generalization properties of a recently-proposed class non-parametric activation functions, kernel functions (KAFs). KAFs introduce additional parameters in learning process order to adapt nonlinearities individually on per-neuron basis, exploiting cheap expansion every value. While increase flexibility has been shown provide significant improvements practice, theoretical proof for its capability not addressed yet literature. Here, leverage recent literature...