IMDEA Networks Institute Digital Repository
Not a member yet
1915 research outputs found
Sort by
Demonstrating Distributed Inference in the User Plane with DUNE
Full text linkDeploying Machine Learning (ML) models in the user plane enables low-latency and scalable in-network inference, but integrating them into programmable devices faces stringent constraints in terms of memory resources and computing capabilities. In this demo, we show how the newly proposed DUNE, a novel framework for distributed user-plane inference across multiple programmable network devices by automating the decomposition of large ML models into smaller sub-models, mitigates the limitations of traditional monolithic ML designs. We run experiments on a testbed with Intel Tofino switches using measurement data and show how DUNE not only improves the accuracy that the traditional single-device monolithic approach gets but also maintains a comparable per-switch latency.TRUEinpres
Middle-Output Deep Image Prior for Blind Hyperspectral and Multispectral Image Fusion
Full text linkObtaining a low-spatial-resolution hyperspectral image (HS) or low-spectral-resolution multispectral (MS) image from a high-resolution (HR) spectral image is straightforward with knowledge of the acquisition models. However, the reverse process, from HS and MS to HR, is an ill-posed problem known as spectral image fusion.} Although recent fusion techniques based on supervised deep learning have shown promising results, these methods require large training datasets involving expensive acquisition costs and long training times. In contrast, unsupervised HS and MS image fusion methods \baccac{have emerged as an alternative to data demand issues; however, they rely on the knowledge of the linear degradation models for optimal performance.} To overcome these challenges, we propose the Middle-Output Deep Image Prior (MODIP) for unsupervised blind HS and MS image fusion. \baccac{MODIP is adjusted for the HS and MS images, and the HR fused image is estimated at} an intermediate layer within the network. The architecture comprises two convolutional \baccac{networks} that reconstruct the HR spectral image from HS and MS inputs, along with two networks that appropriately downscale the estimated HR image to match the available MS and HS \baccac{images}, learning the non-linear degradation models. The network parameters of MODIP are jointly and iteratively adjusted by minimizing a proposed loss function. This approach can handle scenarios where the degradation operators are unknown or partially estimated. To evaluate the performance of MODIP, we test the fusion approach on three simulated spectral image datasets (Pavia University, Salinas Valley, and CAVE) and a real dataset obtained through a testbed implementation in an optical lab. Extensive simulations demonstrate that MODIP outperforms other unsupervised model-based image fusion methods \baccac{by up to 6 dB in PNSR.TRUEpu
SYMBXRL: Symbolic Explainable Deep Reinforcement Learning for Mobile Networks
Full text linkThe operation of future 6th-generation (6G) mobile networks will increasingly rely on the ability of Deep Reinforcement Learning (DRL) to optimize network decisions in real-time. DRL yields demonstrated efficacy in various resource allocation problems, such as joint decisions on user scheduling and
antenna allocation or simultaneous control of computing resources and modulation. However, trained DRL agents are closed-boxes and inherently difficult to explain, which hinders their adoption in production settings. In this paper, we make a step towards removing this critical barrier by presenting SYMBXRL, a novel technique for EXplainable Reinforcement Learning (XRL) that synthesizes human-interpretable explanations for DRL agents. SYMBXRL leverages symbolic AI to produce explanations where key concepts and their relationships are described via intuitive symbols and rules; coupling such a representation with logical reasoning exposes the decision process of DRL agents and offers more comprehensible descriptions of their behaviors compared to existing approaches. We validate SYMBXRL in practical network management use cases supported by DRL, proving that it not only improves the semantics of the explanations but also paves the way for explicit agent control: for instance, it enables intent-based programmatic action steering that improves by 12% the median cumulative reward over a pure DRL solution.TRUEinpres
Practical and General-Purpose Flow-Level Inference with Random Forests in Programmable Switches
Full text linkIntegrating machine learning (ML) models directly in the network user plane enables inference on data traffic at line rate, and can dramatically reduce the latency and improve the scalability of key functionalities like traffic classification or intrusion detection. Yet, the hardware that can be used to this purpose, in particular programmable switches, present stringent constraints in terms of limited memory and little support for mathematical operations or data types that render ML model deployment a substantial technical challenge. In this paper, we make a step forward in user-plane ML by introducing Flowrest, a solution that redefines the state of the art in flow-level inference for programmable switches. Flowrest allows implementing general-purpose Random Forest (RF) models in industry-grade switches by (i) suitably handling stateful flow-level (FL) features in the switch ASIC, (ii) achieving low-collision flow management, and (iii) customizing RF models right from the design phase for in-switch operation. We develop Flowrest as an open-source software using the P4 language and evaluate its performance in an experimental testbed with Intel Tofino switches. Experiments with inference tasks of varying complexity prove that our solution improves accuracy by over 10 percent points on average with respect to the second-best competitor out to five recent approaches for RF-based in-switch inference, while maintaining sub-microsecond latency.TRUEinpres
Interactive Explanation and Steering of DRL Agents for Massive MIMO Scheduling with SYMBXRL
Full text linkFuture 6th-generation (6G) mobile networks will increasingly rely on Deep Reinforcement Learning (DRL) for real-time decision optimization. However, DRL’s opaque nature
hinders its adoption, as operators need to understand and control
these complex systems, necessitating explainability tools to reveal
the model’s reasoning. This paper demonstrates SYMBXRL,
an EXplainable Reinforcement Learning ( XRL) framework
that translates DRL’s internal logic into human-interpretable
symbolic representations and enables intent-based action steering.
We introduce a novel interactive dashboard that enhances
transparency and control by providing a real-time view of the
DRL agent’s operation. Our demonstration showcases how SYM-
BXRL i) generates human-readable explanations using symbolic
Artificial Intelligence (AI) and knowledge graphs, (ii) enables
operator-defined, intent-based action steering for performance
improvement, and (iii) provides real-time visualization of agent
behavior and network metrics. We demonstrate SYMBXRL using
a DRL agent that schedules users in a Massive MIMO scenario,
leveraging real-world channel measurements from a 64-antenna
testbed to maximize spectral efficiency while maintaining fairness.TRUEinpres
An Enhanced Virtualized Control and Key Management Model for QKD Networks
Full text linkThe advent of softwarization and disaggregated architectures has transformed modern communication
networks, sparking innovation by separating network functionalities from the underlying hardware. Fol-
lowing this trend, in future quantum networks, the virtualization and softwarization of critical compo-
nents will be essential to achieve global interoperability and enhance adaptability. Virtualization allows
for the abstraction of hardware resources, making it easier to manage and scale networks dynamically,
while softwarization promotes flexibility, reconfigurability, and interoperability by enabling the use of stan-
dardized, software-defined protocols that can be easily updated and modified to work across diverse and
evolving network environments. In this paper, we introduce and examine an operational model for QKD
networks that leverages the virtualization of control and key management functionalities. We detail its el-
ements and procedures to optimize QKD services. The design paves the way for the integration of quan-
tum networks with functions already established in modern mobile networks while adhering to estab-
lished QKD network standards. The proposed operational model has been validated at the 5G Telefonica
Open Network Innovation Centre (5TONIC) using an environment that deploys digital twins of QKD net-
works.TRUEpu
Your Signal, Their Data: An Empirical Privacy Analysis of Wireless-scanning SDKs in Android
Full text linkMobile apps frequently use Bluetooth Low Energy (BLE) and WiFi scanning permissions to discover nearby devices like peripherals and connect to WiFi Access Points (APs). However, wireless interfaces also serve as a covert proxy for geolocation data, enabling continuous user tracking and profiling. This includes technologies like BLE beacons, which are BLE devices broadcasting unique identifiers to determine devices’ indoor physical locations; such beacons are easily found in shopping centres. Despite the widespread use of wireless scanning APIs and their potential for privacy abuse, the interplay between commercial mobile SDKs with wireless sensing and beaconing technologies remains largely unexplored. In this work, we conduct the first systematic analysis of 52 wireless-scanning SDKs, revealing their data collection practices and privacy risks. We develop a comprehensive analysis pipeline that enables us to detect beacon scanning capabilities, inject wireless events to trigger app behaviors, and monitor runtime execution on instrumented devices. Our findings show that 86% of apps integrating these SDKs collect at least one sensitive data type, including device and user identifiers such as AAID, email, along with GPS coordinates, WiFi and Bluetooth scan results. We uncover widespread SDK-to-SDK data sharing and evidence of ID bridging, where persistent and resettable identifiers are shared and synchronized within SDKs embedded in applications to potentially construct detailed mobility profiles, compromising user anonymity and enabling long-term tracking. We provide evidence of key actors engaging in these practices and conclude by proposing mitigation strategies such as stronger SDK sandboxing, stricter enforcement of platform policies, and improved transparency mechanisms to limit unauthorized tracking.TRUEinpres
Deanonymizing Ethereum Validators: The P2P network has a privacy issue
Full text linkMany blockchain networks aim to preserve the anonymity of validators in the \textit{peer-to-peer (P2P)} network, ensuring that no adversary can link a validator's identifier to the IP address of a peer due to associated privacy and security concerns. This work demonstrates that the Ethereum P2P network does not offer this anonymity. We present a methodology that enables any node in the network to identify validators hosted on connected peers and empirically verify the feasibility of our proposed method. Using data collected from four nodes over three days, we locate more than 15% of Ethereum validators in the P2P network. The insights gained from our deanonymization technique provide valuable information on the distribution of validators across peers, their geographic locations, and hosting organizations. We further discuss the implications and risks associated with the lack of anonymity in the P2P network and propose methods to help validators protect their privacy. The Ethereum Foundation has awarded us a bug bounty, acknowledging the impact of our results.TRUEinpres
Steady-state coherence in multipartite quantum systems: its connection with thermodynamic quantities and impact on quantum thermal machines
Full text linkUnderstanding how coherence of quantum systems affects thermodynamic quantities, such as work and heat, is essential for harnessing quantumness effectively in thermal quantum technologies. Here, we study the unique contributions of quantum coherence among different subsystems of a multipartite system, specifically in non-equilibrium steady states, to work and heat currents. Our system comprises two coupled ensembles, each consisting of N particles, interacting with two baths of different temperatures, respectively. The particles in an ensemble interact with their bath either simultaneously or sequentially, leading to non-local dissipation and enabling the decomposition of work and heat currents into local and non-local components. We find that the non-local heat current, as well as both the local and non-local work currents, are linked to the system quantum coherence. We provide explicit expressions of coherence-related quantities that determine the work currents under various intrasystem interactions. Our scheme is versatile, capable of functioning as a refrigerator, an engine, and an accelerator, with its performance being highly sensitive to the configuration settings. These findings establish a connection between thermodynamic quantities and quantum coherence, supplying valuable insights for the design of quantum thermal machines.TRUEinpres
Sentiment Analysis on Social Networks for Defining Innovation Problems in Organizations
No full textIn recent years, social networks have transformed into dynamic platforms where individuals express personal thoughts, share emotions, and generate content on virtually any topic. To harness the potential of this information, we propose a sentiment analysis system for social networks grounded in the autonomic cycle of data analysis tasks paradigm, aimed at identifying innovation challenges within organizations. This autonomic cycle comprises a series of tasks that systematically collect and manage large volumes of unstructured social media data, facilitating the identification of innovation problems through sentiment analysis. These tasks involve steps such as filtering negative tweets, identifying key terms, and clustering these tweets to analyze centroids for additional insights aligned with the five W-model questions: What, Where, When, Why, and Who, which are essential for problem definition. The final stage centers on defining customer-driven innovation challenges based on the clustered data. The paper concludes with a case study analyzing tweets from a fashion enterprise, in which very promising results are obtained.TRUEpu