IMDEA Networks Institute Digital Repository
Not a member yet
1915 research outputs found
Sort by
AI in the Gray: Exploring Moderation Policies in Dialogic Large Language Models vs. Human Answers in Controversial Topics
Get PDFThe introduction of ChatGPT and the subsequent improvement of Large Language Models (LLMs) have prompted more and more individuals to turn to the use of ChatBots, both for information and assistance with decision-making. However, the information the user is after is often not formulated by these ChatBots objectively enough to be provided with a definite, globally accepted answer.
Controversial topics, such as "religion", "gender identity", "freedom of speech", and "equality", among others, can be a source of conflict as partisan or biased answers can reinforce preconceived notions or promote disinformation. By exposing ChatGPT to such debatable questions, we aim to understand its level of awareness and if existing models are subject to socio-political and/or economic biases. We also aim to explore how AI-generated answers compare to human ones. For exploring this, we use a dataset of a social media platform created for the purpose of debating human-generated claims on polemic subjects among users, dubbed Kialo.
Our results show that while previous versions of ChatGPT have had important issues with controversial topics, more recent versions of ChatGPT (gpt-3.5-turbo) are no longer manifesting significant explicit biases in several knowledge areas. In particular, it is well-moderated regarding economic aspects. However, it still maintains degrees of implicit libertarian leaning toward right-winged ideals which suggest the need for increased moderation from the socio-political point of view. In terms of domain knowledge on controversial topics, with the exception of the "Philosophical" category, ChatGPT is performing well in keeping up with the collective human level of knowledge. Finally, we see that sources of Bing AI have slightly more tendency to the center when compared to human answers. All the analyses we make are generalizable to other types of biases and domains.Spanish Ministry of Science and InnovationESF Investing in your futureUK's Research centre on Privacy, Harm Reduction & Adversarial Influence onlineTRUEinpres
Consistent Comparison of Symptom-based Methods for COVID-19 Infection Detection (Extended Abstract)
Get PDFDuring the global pandemic crisis, several COVID-19 diagnosis methods based on survey information have been proposed with the purpose of providing medical staff with quick detection tools that allow them to efficiently plan the limited healthcare resources. In general, these methods have been developed to detect COVID-19-positive cases from a particular combination of self-reported symptoms. In addition, these methods have been evaluated using datasets extracted from different studies with different characteristics. On the other hand, the University of Maryland, in partnership with Facebook, launched the Global COVID-19 Trends and Impact Survey (UMD-CTIS), the largest health surveillance tool to date that has collected information from 114 countries/territories from April 2020 to June 2022. This survey collected information on various individual features including gender, age groups, self-reported symptoms, isolation measures, and mental health status, among others. In this paper, we compare the performance of different COVID-19 diagnosis methods using the information collected by UMD-CTIS, for the years 2020 and 2021, in six countries: Brazil, Canada, Israel, Japan, Turkey, and South Africa. The evaluation of these methods with homogeneous data across countries and years provides a solid and consistent comparison among them.Spanish Ministry of Science and InnovationEuropean Union "Next Generation EU"TRUEinpres
AI-driven Orchestration for 6G Networking: the Hexa-X vision
Get PDFMobile networks are adopting disaggregation and modularisation to support flexibility. However, large modular networks with a wide range of heterogeneous components have many degrees of freedom, making its Management and Orchestration complex. The use of Machine Learning techniques is expected to improve the efficiency of the operation of 6G networks, by introducing data-driven approaches into their Management and Orchestration. In this paper, we review the current best practices of ML usage to support Management and Orchestration, and we present the H2020 European project Hexa-X Management and Orchestration architecture. We then identify the main challenges ahead to fully embrace a Machine Learning driven operation.TRUEpu
Empowerment of atypical viewers via low-effort personalized modeling of video streaming quality
Get PDFThis is a CoNEXT 2023 conference item published in the journal.Quality of Experience (QoE) and QoE models are of an increasing importance to networked systems. The traditional QoE modeling for video streaming applications builds a one-size-fits-all QoE model that underserves atypical viewers who perceive QoE differently. To address the problem of atypical viewers, this paper proposes iQoE (individualized QoE), a method that employs explicit, expressible, and actionable feedback from a viewer to construct a personalized QoE model for this viewer. The iterative iQoE design exercises active learning and combines a novel sampler with a modeler. The chief emphasis of our paper is on making iQoE sample-efficient and accurate. By leveraging the Microworkers crowdsourcing platform, we conduct studies with 120 subjects who provide 14,400 individual scores. According to the subjective studies, a session of about 22 minutes empowers a viewer to construct a personalized QoE model that, compared to the best of the 10 baseline models, delivers the average accuracy improvement of at least 42% for all viewers and at least 85% for the atypical viewers. The large-scale simulations based on a new technique of synthetic profiling expand the evaluation scope by exploring iQoE design choices, parameter sensitivity, and generalizability.Spanish Ministry of Science and InnovationAustrian Federal Ministry for Digital and Economic Affairs, National Foundation for Research, Technology and Development, and Christian Doppler Research AssociationTRUEpu
ATHENA: Machine Learning and Reasoning for Radio Resources Scheduling in vRAN systems
Get PDFNext-generation mobile networks will rely on their autonomous operation. Virtual Network Functions empowered by Artificial Intelligence (AI) and Machine Learning (ML) can adapt to varying environments that encompass both network conditions and the cloud platform executing them. In this view, it becomes paramount to understand why AI/ML algorithms made a decision, to be able to reason upon those decisions and, eventually, take further decisions related to e.g., network orchestration. In this paper, we present ATHENA, an ML-based radio resource scheduler for virtualized Radio Access Network (RAN) system. Our real-software implementation shows that the proposed ML-based approach can outperform the baseline solution. We discuss how additional re-orchestration actions can be taken by analyzing our scheduling decisions and learning from the past.TRUEpu
A Primer on Online Convex Optimization
No full textWe will start the presentation with the general Online Learning (OL) procedure. We will present one of the most important algorithms within the Online Cover Optimization (OCO) framework, the Online Mirror Descent (OMD), and we will give a brief intuitive connection to how OMD can adapt to different geometries of the feasible space in order to give solutions with performance guarantees. Then we will present Regret, one of the main metric considered in OCO, and the main OL benchmarks against which online algorithms are compared. We will see some results on the no-regret property (no-regret, i.e. asymptotic optimality, and negative results). We will also see some specific research works on which OCO has been applied as a solution and conclude by wrapping up with some use cases and discuss possible future directions where OCO could be a good tool and those where it is not an adequate tool, focusing on problems that arise within networking.FALSEpu
Handling Data Handoff of AI-based Applications in Edge Computing Systems
Get PDFEdge computing aims at better supporting low-latency applications. One of its key techniques is computation offloading, the process that outsources computing tasks from resourced-constrained mobile devices and moves them to edge data centers. In this paper, we tackle an emerging problem within the umbrella of computation offloading, i.e., migration of offloaded inference tasks of Artificial Intelligence (AI) trained models. Such context tailors migration aspects of data-sensitive services where i) the value of the updates is inversely proportional to the data age and ii) outage is highly detrimental to accuracy. To tackle this challenge, we propose Mobile Edge Data-handoff (MED) a framework able to relocate inference or online training tasks from one edge datacenter to another by moving only the necessary data to minimize any accuracy drop during the process. We implemented MED in a well-known edge computing emulator, openLEON, and experimentally verified its performance with an AI-based Industry 4.0 application that forecasts the gas flow in a chemical plant. For our experiments, we use a real, open-source dataset that contains sensors readings. Collected results show that MED, employing proactive data handoff algorithms, is able to minimize the packet loss during the handoff thereby providing guarantees on the inference accuracy.AEITRUEpu
EXPLORA: AI/ML EXPLainability for the Open RAN
Get PDFThe Open Radio Access Network (RAN) paradigm is transforming cellular networks into a system of disaggregated, virtualized, and software-based components. These self-optimize the network through programmable, closed-loop control, leveraging Artificial Intelligence (AI) and Machine Learning (ML) routines. In this context, Deep Reinforcement Learning (DRL) has shown great potential in addressing complex resource allocation problems. However, DRL-based solutions are inherently hard to explain, which hinders their deployment and use in practice. In this paper, we propose EXPLORA, a framework that provides explainability of DRL-based control solutions for the Open RAN ecosystem. EXPLORA synthesizes network-oriented explanations based on an attributed graph that produces a link between the actions taken by a DRL agent (i.e., the nodes of the graph) and the input state space (i.e., the attributes of each node). This novel approach allows EXPLORA to explain models by providing information on the wireless context in which the DRL agent operates. EXPLORA is also designed to be lightweight for real-time operation. We prototype EXPLORA and test it experimentally on an O-RAN-compliant near-real-time RIC deployed on the Colosseum wireless network emulator. We evaluate EXPLORA for agents trained for different purposes and showcase how it generates clear network-oriented explanations. We also show how explanations can be used to perform informative and targeted intent-based action steering and achieve median transmission bitrate improvements of 4% and tail improvements of 10%Ministerio de Ciencia e InnovaciónMinisterio de Asuntos Económicos y Transformación DigitalTRUEpu
Rethinking LiFi for Carbon Neutral Sunlight-based Communication
Get PDFWireless communication has to be redesigned to achieve sustainable communication infrastructure providing net-zero carbon dioxide emission. This paper presents a novel Light Fidelity (LiFi) design that exploits the concept of tubular light guides that can be installed in new buildings. While these guides bring natural sunlight indoors, our proposed system LiFiTube modulates it with low-power liquid crystal (LC) shutters to provide wireless data services without using any power hungry electronic components. Preliminary results show that our proposed LiFiTube system can provide a self-sustainable communication technology and can guarantee demands of Internet of Things networks, while illuminating indoors. We also draw some future research directions.European UnionTRUEpu
Analysis of security and data control in smart personal assistants from the user’s perspective
Get PDFAdvances in the fields of the Internet of Things, Speech Recognition and Artificial Intelligence have facilitated the development of Smart Personal Assistants. As a result, Smart Personal Assistants currently allow requesting a wide range of tasks naturally and intuitively through voice interaction. Their wide popularity, together with the high technological complexity of their environments, have made them an attractive target from a security point of view. Recent works have shown some of the security and privacy issues they stand upon. In this work, we propose a methodology to carry out a systematic security analysis of Smart Personal Assistants using a comprehensive set of tests designed to measure issues around the installation, the interaction, key functionality, and overall Security and Privacy controls. We apply this methodology to analyze security and data control in predominant commercial Smart Personal Assistants (SPA), including Apple HomePod, Google Home and Nest, Amazon Echo (Show and Dot), and Facebook Portal. The main findings of our research are: (i) SPA are not resilient to voice replay attacks; (ii) their skills activation mechanisms can be significantly improved to be more reliable in multi-user households; (iii) the users’ control to restrict the collection and access of Personally Identifiable Information can be also improved; (iv) they lack configurations adapted to minors, which should be included to make them more appropriate for a segment of users who interact more and more with them and have especially high regulatory requirements regarding security and data protection. Among the many hot research topics within this area, we find voice authentication and authorization especially interesting since they may push the usability of Smart Personal Assistants further, as long as they are robust enough from the security perspective.This work has been funded by the Horizon 2020 program of the European Union through the RAYUELA project (contract no. 882828). Guillermo Suarez-Tangil was partially funded by the “Ramon y Cajal” Fellowship RYC-2020- 029401-I and supported by TED2021-132900A-I00.TRUEpu