Video streaming has emerged as a dominant mode of entertainment and information dissemination. However, traditional approaches to video bitrate adaptation often fall short in non-stationary network environments, leading to suboptimal user experiences. To address these challenges, this work presents an innovative method leveraging Proximal Policy Optimization (PPO), a Reinforcement Learning (RL) algorithm, to optimize the Quality of Experience (QoE) in video streaming scenarios. A Network Digital Twin (NDT) environment, which allows to emulate production environments, and the end, facilitate high-fidelity training and evaluation while completely isolating live network operations from experimental risks. The PPO algorithm learns to make bitrate decisions based on a reward function that incorporates key QoE metrics, including blockiness and block loss, achieving an impressive 73.02% enhancement in overall performance. Experimental results demonstrate a significant reduction in block loss (66.87%) and an improvement in blockiness (6.40%), while also increasing the streaming bitrate (17.18%) from training to production deployment.

According to the EU State of Cybersecurity report by the European Union Agency for Cybersecurity (ENISA), the number of cybersecurity-related incidents will increase by 24 percent by 2025, with ransomware and DoS/DDoS attacks being the most common. The emergence of new threats [1] and the consolidation of existing ones require doubling of efforts in proactive prevention and a decisive increase in research dedicated to cybersecurity. CyberNEMO (End-to-end Cybersecurity to NEMO meta-OS) project emerges as an evolution of the NEMO (Next Generation Meta Operating System) platform, designed to provide a secure, trustworthy, and robust execution environment across the IoT-Edge-Cloud computing continuum. Leveraging NEMO modular meta-operating system (mOS) framework, CyberNEMO introduces advanced cybersecurity and privacy-preserving mechanisms, emphasizing Zero Trust principles. This paper presents the CyberNEMO architecture, details its core innovative technologies, and describes its validation strategy through diverse living labs-including Smart Energy, Smart Water, Smart Manufacturing, Healthcare, Multimedia Distribution, and Smart Farming scenarios-demonstrating end-to-end cybersecurity and real-time threat mitigation capabilities, aligned with Europe's strategic cybersecurity goals.

The massive growth of live streaming, especially gaming-focused content, has led to an overall increase in global bandwidth consumption. Certain services see their quality diminished at times of peak consumption, degrading the quality of the content. This trend generates new research related to optimizing image quality according to network and service conditions. In this work we present a gaming streaming use case optimization on a real multisite 5G environment. The paper outlines the virtualized workflow of the use case and provides a detailed description of the applications and resources deployed for the simulation. This simulation tests the optimization of the service based on the addition of Artificial Intelligence (AI) algorithms, assuring the delivery of content with good Quality of Experience (QoE) under different working conditions. The AI introduced is based on Deep Reinforcement Learning (DRL) algorithms that can adapt, in a flexible way, to the different conditions that the multimedia workflow could face. That is, adapt, through corrective actions, the streaming bitrate, in order to optimize the QoE of the content on a real-time multisite scenario. The results of this work demonstrate how we have been able to minimize content losses, as well as the fact of obtaining high audiovisual multimedia quality results with higher bitrates, compared to a service without an optimizer integrated in the system. In a multi-site environment, we have achieved an improvement of 20 percentage points in terms of blockiness efficiency and also 15 percentage points in block loss.

This paper presents an horizontal technique to classify the main directions of orientation of the planar surfaces that conform solid bodies. When mathematically representing these surfaces via 3D point clouds, a cascade or chain process using combined information of the point positions and of their detected normal vectors is applied to solidly identify the planar surfaces composing them and to retrieve their orientations. This technique distinguishes itself from others in the field due to (i) a specific filtering inspired by image processing techniques that aims to mitigate the effects that surface rugosity-induced mathematical noise can induce on the classification; and (ii) also due to the data representation on the unit-sphere using the von Mises-Fisher mixture model. The extents and limits of the proposed process are presented using several validation cases of increasing complexity, and its results and possibilities for horizontal applications over many kinds of real-life bodies are also analyzed.

Textual noise, such as typos or abbreviations, is a well-known issue that penalizes vanilla Transformers for most downstream tasks. We show that this is also the case for sentence similarity, a fundamental task in multiple domains, e.g. matching, retrieval or paraphrasing. Sentence similarity can be approached using cross-encoders, where the two sentences are concatenated in the input allowing the model to exploit the inter-relations between them. Previous works addressing the noise issue mainly rely on data augmentation strategies, showing improved robustness when dealing with corrupted samples that are similar to the ones used for training. However, all these methods still suffer from the token distribution shift induced by typos. In this work, we propose to tackle textual noise by equipping cross-encoders with a novel LExical-aware Attention module (LEA) that incorporates lexical similarities between words in both sentences. By using raw text similarities, our approach avoids the tokenization shift problem obtaining improved robustness. We demonstrate that the attention bias introduced by LEA helps cross-encoders to tackle complex scenarios with textual noise, specially in domains with short-text descriptions and limited context. Experiments using three popular Transformer encoders in five e-commerce datasets for product matching show that LEA consistently boosts performance under the presence of noise, while remaining competitive on the original (clean) splits. We also evaluate our approach in two datasets for textual entailment and paraphrasing showing that LEA is robust to typos in domains with longer sentences and more natural context. Additionally, we thoroughly analyze several design choices in our approach, providing insights about the impact of the decisions made and fostering future research in cross-encoders dealing with typos.