IMDEA Networks Institute Digital Repository
Not a member yet
1915 research outputs found
Sort by
Getting the Best Out of Both Worlds: Algorithms for Hierarchical Inference at the Edge
No full textWe consider a resource-constrained Edge Device (ED), such as an IoT sensor or a microcon-
troller unit, embedded with a small-size ML model (S-ML) for a generic classification application and an Edge Server (ES) that hosts a large-size ML model (L-ML). Since the inference accuracy of S-ML is lower than that of the L-ML, offloading all the data samples to the ES results in high inference accuracy, but it defeats the purpose of embedding S-ML on the ED and deprives the benefits of reduced latency, bandwidth savings, and energy efficiency of doing local inference. In order to get the best out of both worlds, i.e., the benefits of doing inference on the ED and the benefits of doing inference on ES, we explore the idea of Hierarchical Inference (HI), wherein S-ML inference is only accepted when it is correct, otherwise, the data sample is offloaded for L-ML inference. However, the ideal implementation of HI is infeasible as the correctness of the S-ML inference is not known to the ED. We thus propose an online meta-learning framework that the ED can use to predict the correctness of the S-ML inference. In particular, we propose to use the probability corresponding to the maximum probability class output by S-ML for a data sample and decide whether to offload it or not. The resulting online learning problem turns out to be a Prediction with Expert Advice (PEA) problem with continuous expert space. For a full feedback scenario, where the ED receives feedback on the correctness of the S-ML once it accepts the inference, we propose the HIL-F algorithm and prove a sublinear regret bound√n ln(1/λmin)/2 without any assumption on the smoothness of the loss function, where n is the number of data samples and λmin is the minimum difference between any two distinct maximum probability values across the data samples. For a no-local feedback scenario, where the ED does not receive the ground truth for the classification, we propose the HIL-N algorithm and prove that it has O (n2/3 ln1/3(1/λmin)) regret bound. We evaluate and benchmark the performance of the proposed algorithms for image classification application using four datasets, namely, Imagenette and Imagewoof, MNIST, and CIFAR-10.EUTRUEpu
In the Room Where It Happens: Characterizing Local Communication and Threats in Smart Homes
Get PDFThe network communication between Internet of Things (IoT) devices on the same local network has significant implications
for platform and device interoperability, security, privacy, and correctness. Yet, the analysis of local home Wi-Fi network traffic
and its associated security and privacy threats have been largely ignored by prior literature, which typically focuses on studying
the communication between IoT devices and cloud end-points, or detecting vulnerable IoT devices exposed to the Internet. In this
paper, we present a comprehensive and empirical measurement study to shed light on the local communication within a smart
home deployment and its threats. We use a unique combination of passive network traffic captures, protocol honeypots, dynamic
mobile app analysis, and crowdsourced IoT data from participants to identify and analyze a wide range of device activities on
the local network. We then analyze these datasets to characterize local network protocols, security and privacy threats associated
with them. Our analysis reveals vulnerable devices, insecure use of network protocols, and sensitive data exposure by IoT devices.
We provide evidence of how this information is exfiltrated to remote servers by mobile apps and third-party SDKs, potentially for
household fingerprinting, surveillance and cross-device tracking. We make our datasets and analysis publicly available to support
further research in this area.TRUEpu
Generation of Genuine Multipartite Entangled States via Indistinguishability of Identical Particles
Get PDFIndistinguishability of identical particles is a resource for quantum information processing and has been utilized to generate entanglement from independent particles that spatially overlap only at the detection stage. Here, we introduce a controllable scheme capable of generating via three steps, i.e., initialization, deformation, and postselection, different classes of multipartite entangled states starting from a product state of ��� spatially distinguishable identical qubits. While our scheme is generalizable to any class of entangled bosonic and fermionic systems, we provide an explicit recipe for the generation of ���, Dicke, GHZ, and cluster states, which are resource states for quantum information processing. Using graph-based representations within the framework of spatially localized operations and classical communication (sLOCC), we mathematically demonstrate a direct translation of the generation schemes of specific entangled states into colored, complex, and weighted digraphs, each corresponding to a given experimental setup. We also show that this graph-theoretical approach allows for the optimization of the generation efficiency of specific multipartite entangled states by exploring a variety of generation schemes. The presented theoretical approach, while already implementable with current linear optics architectures, has the potential to bring clear advantages over existing technologies, such as in quantum computing search algorithms and in the design of new experiments in quantum optics or other platforms.TRUEpu
In-Band Quality Notification from Users to ISPs
Get PDFWhile ISPs (Internet service providers) strive to improve QoE (quality of experience) for end users, end-to-end traffic encryption by OTT (over-the-top) providers undermines independent inference of QoE by an ISP. Due to the economic and technological complexity of the modern Internet, ISP-side QoE inference based on OTT assistance or out-of-band signaling sees low adoption. This paper presents IQN (in-band quality notification), a novel mechanism for signaling QoE impairments from an automated agent on the end-user device to the server-to-client ISP responsible for QoE-impairing congestion. Compatible with multi-ISP paths, asymmetric routing, and other Internet realities, IQN does not require OTT support and induces the OTT server to emit distinctive packet patterns that encode QoE information, enabling ISPs to infer QoE by monitoring these patterns in network traffic. We develop a prototype system, YouStall, which applies IQN signaling to ISP-side inference of YouTube stalls. Cloud-based experiments with YouStall on YouTube Live streams validate IQN’s feasibility and effectiveness, demonstrating its potential for accurate user-assisted ISP-side QoE inference from encrypted traffic in real Internet environments.MICIU/AEI/10.13039/501100011033 and ERDF, EUAustrian Federal Ministry for Digital and Economic AffairsNational Foundation for Research, Technology and Development, AustriaChristian Doppler Research AssociationTRUEpu
Explainability Analysis: An In-Depth Comparison between Fuzzy Cognitive Maps and LAMDA
Get PDFCurrently, it has become very relevant that machine learning techniques can provide an explanation of the results they generate, which is even more relevant in certain domains, called critical, such as health and energy, among others. For this reason, several explainability methods have been generated in the literature. Some of them are Local Interpretable Model-Agnostic Explanations (LIME) and Feature Importance, which have been used in a wide range of problems. The objective of this work is to analyze the explainability capacity of the Learning Algorithm for Multivariate Data Analysis (LAMDA) and Fuzzy Cognitive Maps (FCM), which have been used with great interest due to their interpretability, management of uncertainty during the inference process, simplicity in its use, among other things. For the development of this work, data from two critical domains have been considered, health (COVID-19 and Dengue datasets) and energy (energy price dataset), for which prediction/classification models have been developed using LAMDA and FCM techniques. Afterward, two explainability techniques were used to analyze the explainability provided by each method, one based on the LIME method and the other on the feature importance method, the latter adapted to our work by being based on the permutation of values. Finally, the work proposes two new explainability methods, one based on causal inference and the other on the degrees of membership of the variables to the classes. The latter, in particular, allows doing an explainability analysis by class. The new explainability methods reproduce the results of well-known explainability methods in the literature such as LIME and feature importance, with less execution cost, and also, with an explainability analysis by class. This work opens the doors to new work on class explainability. Furthermore, we see that machine learning approaches based on causal or fuzzy relationships are quite self-explanatory, but specific explainability methods such as those we propose in the work allow us to study particular aspects, such as highly important variables, which general explainability methods do not allow us to do, such as LIME and feature importance.SocialProbingTRUEpu
Exact Resource Allocation for Fair Wireless Relay
Get PDFIn relay-enabled cellular networks, the intertwined nature of network agents calls for complex schemes to allocate wireless resources. Resources need to be distributed among mobile users while considering how relay resources are allocated, and constrained by the traffic rate achievable by base stations and over backhaul links. In this letter, we derive an exact resource allocation scheme that achieves max–min fairness across mobile users, found with a linear complexity with respect to the number of mobile users and relays. The results reveal that the proposed scheme remarkably outperforms current solutions.Ministry of Economic Affairs and Digital Transformation of Spain and the European Union NextGeneration-EU in the framework of the Spanish Recovery, Transformation and Resilience PlanUniversidad Cardenal Herrera-CEU, CEU UniversitiesTRUEpu
Transformer-Based Quantification of the Echo Chamber Effect in Online Communities
Get PDFAn Echo Chamber on social media refers to the environment where like-minded people hear the echo of each others' voices, opinions, or beliefs, which reinforce their own. Echo Chambers can turn social media platforms into collaborative venues that polarize and radicalize users rather than broadening their exposure to diverse information. Having a quantified metric for measuring the Echo Chamber effect can aid moderators and policymakers in tracking and mitigating online polarization and radicalization. Existing methods for Echo Chamber detection are either one-dimensional, only considering the network behavior of users while ignoring their semantic behavior, or require demanding supervised labeling, which is both expensive and less generalizable.
This paper proposes a new metric to quantify the Echo Chamber effect using Transformer models for context-sensitive processing of natural language (NLP). Our metric quantifies (1) the effect of an Echo Chamber through the inverse effect of user diversity, and (2) polarization by means of user separability between two Echo Chambers in a topic. Leveraging this metric, we further propose an NLP-based embedding that represents the users' activity. Our model is simultaneously effective, computationally cheap, and unsupervised. As our method is unsupervised, it makes existing collaborative moderation efforts to thwart Echo Chamber effects more efficient by addressing the problem of identifying narrow information bases for algorithmic biases and misinformation detection. We run our analysis on three recent highly controversial political topics and a non-controversial topic: Russo-Ukrainian War, Abortion, Gun-Control, and SXSW music festival. Our results offer data-driven findings such as a higher Echo Chamber effect among Republicans over Democrats and diverse explicit support for Ukraine, especially among Democrats. We also observe a direct relationship between the Echo Chamber effect and polarization while observing that the low Echo Chamber effect for the Russo-Ukraine war is accompanied by a low polarization; and vice versa for Gun-Control.ESF Investing in your futureSpanish Ministry of Science and InnovationUK's Research Centre on Privacy, Harm Reduction & Adversarial Influence onlineEuropean Union-NextGenerationEUTRUEinpres
Emotions as implicit feedback for adapting difficulty in tutoring systems based on reinforcement learning
Get PDFIn tutoring systems, a pedagogical policy, which decides the next action for the tutor to take, is important because it determines how well students will learn. An effective pedagogical policy must adapt its actions according to the student’s features, such as knowledge, error patterns, and emotions. For adapting difficulty, it is common to consider student knowledge but not the other features as emotions. Reinforcement learning (RL), which is a machine learning framework, fits well for adapting to difficulty; however, the known ways of considering emotions into RL like through states or reward-shaping functions are not enough. Then, to find the pedagogical policy that maximizes the student learning gain, we propose considering emotions as implicit feedback through both the reward and the exploration-exploitation strategy, using the circumplex model to represent emotions and the flow theory to select the appropriate difficulty level. Our approach follows three design considerations: pursuing positive emotions, managing unwanted (anxiety and boredom) emotions, and anticipating unwanted emotions. We simulate interactions with users based on real data from publicly available datasets to quantitatively compare our approach with others that adapt difficulty. Also, we qualitatively compare our approach with others that consider emotions in different contexts. Quantitative results show that our approach is better than the others that adapt difficulty to foster learning gain in students because it allows getting higher values all the studied time (200 tasks). Qualitative comparisons show that although other approaches pursue positive emotions or manage unwanted emotions, our approach does so as well and additionally anticipates unwanted emotions. We conclude that our approach is useful in tutoring systems for adapting difficulty because it allows high learning gains in students in a few interactions.TRUEpu
AMECOS: A Modular Event-based Framework for Concurrent Object Specification
Get PDFIn this work, we introduce a modular framework for specifying distributed systems that we call AMECOS. Specifically, our framework departs from the traditional use of sequential specification, which presents limitations both on the specification expressiveness and implementation efficiency of inherently concurrent objects, as documented by Castañeda, Rajsbaum and Raynal in CACM 2023. Our framework focuses on the interactions between the various system components, specified as concurrent objects. Interactions are described with sequences of object events. This provides a modular way of specifying distributed systems and separates legality (object semantics) from other issues, such as consistency. We demonstrate the usability of our framework by (i) specifying various well-known concurrent objects, such as registers, shared memory, message-passing, reliable broadcast, and consensus, (ii) providing hierarchies of ordering semantics (namely, consistency hierarchy, memory hierarchy, and reliable broadcast hierarchy), and (iii) presenting a novel axiomatic proof of the impossibility of the well-known Consensus problem.FALSEinpres
Different Transfer Learning Approaches for Insect Pest Classification in Cotton
Get PDFBoll weevil is an important pest that affects cotton crops worldwide, causing significant economic losses. The classification of the boll-weevil population is crucial for developing effective pest management strategies. However, the low availability of data and features makes classification a challenging task. This study aimed to investigate the use of Transfer Learning (TL) techniques to improve the classification of boll weevil populations. Three types of TL techniques, instance-based, feature-based, and parameter-based, were studied to improve the classification performance of the machine learning algorithms. This work used data from two domains, one with few instances and the other with few features, to test the proposed approaches. Also, climate variables (temperature, humidity, and rainfall) were incorporated as features to predict the level of the boll-weevil attack. The most relevant results of this work are that define 1) How to measure and quantify the similarity or relationship between tasks of different domains; 2) How to select, align, or adapt the relevant features, instances, or models from the source task/domain to the target task/domain; 3) How to reuse parameter settings from the source domain; and 4) How to evaluate and validate the performance and robustness of the TL model on the target task/domain. The proposed approach achieved significant improvements in classification over previous results in the metrics of accuracy and F-measure. For example, in the case with few instances reached an accuracy of 90.79%, while in the case with few features reached an accuracy of 96.28%. Thus, the results demonstrate the effectiveness of TL techniques in improving the classification of boll-weevil populations in cotton crops when few data and/or features are available.TRUEpu