Contribuciones a congresos

Permanent URI for this collectionhttps://hdl.handle.net/10486/723460

Browse

Now showing 1 - 20 of 1137

Transforming open edX into the next on-campus LMS: Results of the UniDigital Project
(IEEE, 2024-11-13) Despujol Zabala, Ignacio; Turró Ribalta, Carlos; Busquets Mataix, Jaime; Montoro Manrique, Germán; Departamento de Ingeniería Informática; Escuela Politécnica Superior
Open edX is an open-source platform known for delivering MOOCs and SPOCs to huge numbers of students, but its current features lack the detailed management required for individual student management in on-campus courses. This paper presents the results of an initiative by six public universities in the region of Madrid and the Universitat Politècnica de València, as part of a national project called UniDigital. Funded by Spain's Ministry of Universities and the EU's Recovery, Transformation, and Resilience Plan, this project, led by UC3M, UPV, and UAM, has invested over half a million euros to enhance Open edX for on-campus use.After partnering with Edunext, one of Open edX's top-tier partners, the project involved close collaboration with the Open edX community to integrate these enhancements into future platform versions. Key improvements include real-time analytics, advanced dashboards, customized feedback systems, enhanced exams and tasks, extended grading options, improved user interfaces for students and teachers, expanded email functionalities, revamped file management, H5P content integration, mind map tools, a student risk detection system, and the development of advanced voice assistants and a gamification extension for mobile apps. These enhancements aim to transform Open edX from a MOOC platform into a sophisticated on-campus Learning Management System (LMS), suitable for both online and campus-based education.We will discuss the initial goals and the real accomplishments, highlighting the complexities of negotiating with an open-source community with diverse interests and varying timeframes, adding layers of complexity to ensuring that developments are incorporated into the main source code.
Variational linearized laplace approximation for bayesian deep learning
(MLResearchPress, 2024-07-21) Ortega Andrés, Luis Antonio; Rodríguez Santana, Simón; Hernández Lobato, Daniel; Departamento de Ingeniería Informática; Escuela Politécnica Superior
The Linearized Laplace Approximation (LLA) has been recently used to perform uncertainty estimation on the predictions of pre-trained deep neural networks (DNNs). However, its widespread application is hindered by significant computational costs, particularly in scenarios with a large number of training points or DNN parameters. Consequently, additional approximations of LLA, such as Kronecker-factored or diagonal approximate GGN matrices, are utilized, potentially compromising the model’s performance. To address these challenges, we propose a new method for approximating LLA using a variational sparse Gaussian Process (GP). Our method is based on the dual RKHS formulation of GPs and retains as the predictive mean the output of the original DNN. Furthermore, it allows for efficient stochastic optimization, which results in sub-linear training time in the size of the training dataset. Specifically, its training cost is independent of the number of training points. We compare our proposed method against accelerated LLA (ELLA), which relies on the Nyström approximation, as well as other LLA variants employing the sample-then-optimize principle. Experimental results, both on regression and classification datasets, show that our method outperforms these already existing efficient variants of LLA, both in terms of the quality of the predictive distribution and in terms of total computational time
SDFR: Synthetic Data for Face Recognition competition
(IEEE, 2024-07-11) Shahreza, Hatef Otroshi; de Andrés Tamé, Iván Ioel; Tolosana Moranchel, Rubén; Vera Rodríguez, Rubén; Fiérrez Aguilar, Julián; Departamento de Tecnología Electrónica y de las Comunicaciones; Escuela Politécnica Superior
Large-scale face recognition datasets are collected by crawling the Internet and without individuals' consent, raising legal, ethical, and privacy concerns. With the recent advances in generative models, recently several works proposed generating synthetic face recognition datasets to mitigate concerns in web-crawled face recognition datasets. This paper presents the summary of the Synthetic Data for Face Recognition (SDFR) Competition held in conjunction with the 18th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2024) and established to investigate the use of synthetic data for training face recognition models. The SDFR competition was split into two tasks, allowing participants to train face recognition systems using new synthetic datasets and/or existing ones. In the first task, the face recognition backbone was fixed and the dataset size was limited, while the second task provided almost complete freedom on the model backbone, the dataset, and the training pipeline. The submitted models were trained on existing and also new synthetic datasets and used clever methods to improve training with synthetic data. The submissions were evaluated and ranked on a diverse set of seven benchmarking datasets. The paper gives an overview of the submitted face recognition models and reports achieved performance compared to baseline models trained on real and synthetic datasets. Furthermore, the evaluation of submissions is extended to bias assessment across different demography groups. Lastly, an outlook on the current state of the research in training face recognition models using synthetic data is presented, and existing problems as well as potential future directions are also discussed
Recommendation fairness in e-participation: Listening to minority, vulnerable and NIMBY citizens
(Springer, 2024-03-15) Alonso Cortés, Marina; Cantador Gutiérrez, Iván; Bellogin Kouki, Alejandro; Departamento de Ingeniería Informática; Escuela Politécnica Superior
E-participation refers to the use of digital technologies and online platforms to engage citizens and other stakeholders in democratic and government decision-making processes. Recent research work has explored the application of recommender systems to e-participation, focusing on the development of algorithmic solutions to be effective in terms of personalized content retrieval accuracy, but ignoring underlying societal issues, such as biases, fairness, privacy and transparency. Motivated by this research gap, on a public e-participatory budgeting dataset, we measure and analyze recommendation fairness metrics oriented to several minority, vulnerable and NIMBY (Not In My Back Yard) groups of citizens. Our empirical results show that there is a strong popularity bias (especially for the minority groups) due to how content is presented and accessed in a reference e-participation platform; and that hybrid algorithms exploiting user geolocation information in a collaborative filtering fashion are good candidates to satisfy the proposed fairness conceptualization for the above underrepresented citizen collectives
SaFL: Sybil-aware Federated Learning with application to face recognition
(IEEE, 2024-12-02) Ghafourian, Mahdi; Fiérrez Aguilar, Julián; Vera Rodríguez, Rubén; Tolosana Moranchel, Rubén; Morales Moreno, Aythami; Biometrics and Data Pattern Analytics - BiDa Lab; Escuela Politécnica Superior
Federated Learning (FL) is a machine learning paradigm to conduct collaborative learning among clients on a joint model. The primary goal is to share clients' local training parameters with an integrating server while preserving their privacy. This method permits to exploit the potential of massive mobile users' data for the benefit of machine learning models' performance while keeping sensitive data on local devices. On the downside, FL raises security and privacy concerns that have just started to be studied. To address some of the key threats in FL, researchers have proposed to use secure aggregation methods (e.g. homomorphic encryption, secure multiparty computation, etc.). These solutions improve some security and privacy metrics, but at the same time bring about other serious threats such as poisoning attacks, backdoor attacks, and free running attacks. This paper proposes a new defense method against poisoning attacks in FL called SaFL (Sybil-Aware Federated Learning) that minimizes the effect of sybils with a novel time-variant aggregation scheme
Synthmanticlidar: A synthetic dataset for semantic segmentation on Lidar imaging
(IEEE, 2024-09-27) Montalvo Rodrigo, Javier; Carballeira López, Pablo; García Martín, Álvaro; Video Processing and Understanding Labgrupo de Tratamiento e Interpretación de Vídeo; Departamento de Tecnología Electrónica y de las Comunicaciones; Escuela Politécnica Superior
Semantic segmentation on LiDAR imaging is increasingly gaining attention, as it can provide useful knowledge for perception systems and potential for autonomous driving. However, collecting and labeling real LiDAR data is an expensive and time-consuming task. While datasets such as SemanticKITTI [1] have been manually collected and labeled, the introduction of simulation tools such as CARLA [2], has enabled the creation of synthetic datasets on demand. In this work, we present a modified CARLA simulator designed with LiDAR semantic segmentation in mind, with new classes, more consistent object labeling with their counterparts from real datasets such as SemanticKITTI, and the possibility to adjust the object class distribution. Using this tool, we have generated SynthmanticLiDAR, a synthetic dataset for semantic segmentation on LiDAR imaging, designed to be similar to SemanticKITTI, and we evaluate its contribution to the training process of different semantic segmentation algorithms by using a naive transfer learning approach. Our results show that incorporating SynthmanticLiDAR into the training process improves the overall performance of tested algorithms, proving the usefulness of our dataset, and therefore, our adapted CARLA simulator
VAAD: Visual Attention Analysis Dashboard applied to e-learning
(IEEE, 2024-07-29) Navarro, Miriam; Becerra, Álvaro; Daza García, Roberto; Cobos Pérez, Ruth; Morales Moreno, Aythami; Fiérrez Aguilar, Julián; Departamento de Ingeniería Informática; Departamento de Tecnología Electrónica y de las Comunicaciones; Escuela Politécnica Superior
In this paper, we present an approach in the Multimodal Learning Analytics field. Within this approach, we have developed a tool to visualize and analyze eye movement data collected during learning sessions in online courses. The tool is named VAAD -an acronym for Visual Attention Analysis Dashboard-. These eye movement data have been gathered using an eye-tracker and subsequently processed and visualized for interpretation. The purpose of the tool is to conduct a descriptive analysis of the data by facilitating its visualization, enabling the identification of differences and learning patterns among various learner populations. Additionally, it integrates a predictive module capable of anticipating learner activities during a learning session. Consequently, VAAD holds the potential to offer valuable insights into online learning behaviors from both descriptive and predictive perspectives
The Financial Document Causality Detection Shared Task (FinCausal 2025): What are the FinCausal series?
(2025-01-24) Moreno Sandoval, Antonio; Carbajo Coronado, Blanca; Porta Zamorano, Jordi; Torterolo Orta, Yanco Amor; Samy, Doaa Ahmed; UAM. Departamento de Lingüística General, Lógica y Filosofía de la Ciencia, Lenguas Modernas, Teoría de la Literatura y Literatura Comparada y Estudios de Asia Oriental; Laboratorio de Lingüística Informática; Departamento de Ingeniería Informática; Departamento de Lingüística General, Lógica y Filosofía de la Ciencia, Lenguas Modernas, Teoría de la Literatura y Literatura Comparada, Estudios de Asia Oriental; Escuela Politécnica Superior; Facultad de Filosofía y Letras
The Financial Document Causality Detection Shared Task (FinCausal 2025)
(Association for Computational Linguistics, 2025-01-20) Moreno Sandoval, Antonio; Porta Zamorano, Jordi; Carbajo Coronado, Blanca; Torterolo Orta, Yanco Amor; Samy, Doaa Ahmed; UAM. Departamento de Lingüística General, Lógica y Filosofía de la Ciencia, Lenguas Modernas, Teoría de la Literatura y Literatura Comparada y Estudios de Asia Oriental; Laboratorio de Lingüística Informática; Departamento de Ingeniería Informática; Departamento de Lingüística General, Lógica y Filosofía de la Ciencia, Lenguas Modernas, Teoría de la Literatura y Literatura Comparada, Estudios de Asia Oriental; Escuela Politécnica Superior; Facultad de Filosofía y Letras
We present the Financial Document Causality Detection Task (FinCausal 2025), a multilingual challenge to identify causal relationships within financial texts. This task comprises English and Spanish subtasks, with datasets compiled from British and Spanish annual reports. Participants were tasked with identifying and generating answers to questions about causes or effects within specific text segments. The dataset combines extractive and generative question-answering (QA) methods, with abstractly formulated questions and directly extracted answers from the text. Systems performance is evaluated using exact matching and semantic similarity metrics. The challenge attracted submissions from 10 teams for the English subtask and 10 teams for the Spanish subtask. FinCausal 2025 is part of the 6th Financial Narrative Processing Workshop (FNP 2025), hosted at COLING 2025 in Abu Dhabi
Graph-based interface for explanations by examples in recommender systems: A user study
(Springer Nature, 2024-07-10) Caro Martínez, Marta; Jorro Aragoneses, José Luis; Díaz Agudo, Belén; Recio García, Juan A.; Departamento de Ingeniería Informática; Escuela Politécnica Superior
In recent years, recommender systems have used advanced machine-learning techniques to improve recommendation precision. However, many of these approaches, like deep-learning, are considered black-box models; in other words, users struggle to comprehend why the system recommends a particular item, affecting the user’s confidence in the provided recommendation. Some recommender systems include methods to explain a recommendation. A common technique is applying explanations-by-examples, but sometimes users do not understand the link between the recommendation and the examples. This paper describes a user study where we have evaluated a novel method to create graph-based explanations. It uses Formal Concept Analysis to extract the most relevant attributes to relate the recommendation with each example. Next, this method shows an interactive graph to users that explains the recommendation based on the links extracted before. Results show a high level of satisfaction regarding the explanations and their visualisation compared with other approaches found in the literature
Long-term geo-positioned re-identification dataset of urban elements
(IEEE, 2024-09-27) Moral De Eusebio, Paula; García Martín, Álvaro; Martínez Sánchez, José María; Video Processing and Understanding Labgrupo de Tratamiento e Interpretación de Vídeo; Departamento de Tecnología Electrónica y de las Comunicaciones; Escuela Politécnica Superior
This paper introduces UrbAM-ReID, a new long-term geo-positioned urban ReID dataset. It is composed by four sub-datasets recording the same trajectory at the UAM Campus, each one recorded in different seasons and including an inverse direction recording. While most of the current datasets in the state-of-the-art focus on person re-identification, with vehicles as the second most explored object, our work specifically addresses urban objects re-identification, currently, waste containers, rubbish bins, and crosswalks. The dataset provides different attributes of the annotated objects, like their classes, their foreground or background status and the geoposition. Several evaluation configurations can be defined to simulate realistic scenarios that may arise in actual situations within the management of urban elements, considering the utilization of just visual data, or incorporating additional attributes, providing different complexity levels. Finally, the dataset is used for defining a benchmark where two state-of-the-art systems are evaluated
Learning analytics tools to analyze progress and results with Moodle LMS Data
(IEEE, 2024-07-08) Alonso Fernández, Cristina; Jorro Aragoneses, José Luis; Alaiz Gudín, Carlos María; Rodríguez Marín, Pilar; Departamento de Ingeniería Informática; Escuela Politécnica Superior
Teachers can benefit from the information provided by learning analytics data for multiple purposes. Visual learning analytics dashboards provide near real-time information while more complex offline tools are commonly used to synthesize and transform the data gathered into interpretable information for teachers. The extended use of Learning Management Systems in universities, such as Moodle or Canvas, provides a rich environment to capture learning analytics data from students' interactions while they are progressing in their courses. In this paper, we present two different learning analytics tools aimed at teachers to obtain information about students' progress and results using data from the Moodle LMS at different stages of their learning process: (1) a progress visualization plugin for Moodle, which provides teachers with real-time information about the progress achieved by students in their courses, and the different goals set for their plans; and (2) an analytics Jupyter Notebook tool with a pre-defined set of analysis and visualizations to apply to data gathered from default activities in Moodle. The plugin is in an initial validation stage, while the analysis tool has been tested in a case study in a university course. Combined, both contributions can enrich the information that teachers have during and after the academic year, adapting their classes to better fit students' progress and needs, as well as providing overall results and comparison between groups after the course has finished
A generative AI-based personalized guidance tool for enhancing the feedback to MOOC learners
(IEEE, 2024-07-08) Becerra, Álvaro; Mohseni, Zeynab; Sanz, Javier; Cobos Pérez, Ruth; Departamento de Ingeniería Informática; Escuela Politécnica Superior
The widespread adoption of Massive Open Online Courses (MOOCs) has profoundly influenced higher education by granting learners access to an extensive array of educational materials. However, the substantial volume of data generated by MOOCs presents a considerable challenge for instructors who aim to assess and facilitate effective learner support. In this study, we introduce an innovative GenAI-based (Generative Artificial Intelligence) tool designed to assist and guide MOOC learners in understanding their progress in the course to enhance their performance and prevent dropout. Our proposed approach takes advantage of GenAI's capabilities to analyze and understand anonymized learner educational data, including aspects such as course progression, assignment results, time spent on different types of content, timestamps, and other pertinent information. By applying natural language processing techniques, GenAI identifies patterns and trends within the data, enabling it to provide personalized guidance to learners to help them develop better learning strategies and enhance their performance in the course. The proposed tool, named GePeTo (Generative AI-based Personalized Guidance Tool), not only streamlines the process of analyzing large volumes of educational data but also equips instructors with practical insights into their learners' performance and difficulties. GePeTo offers a promising solution for higher education institutions aiming to leverage the potential of MOOC data for effective learner assessment and support. Automating the analysis of educational data and delivering personalized guidance to learners will also facilitate instructors in making data-driven decisions. Ultimately, this will improve learning outcomes and educational experiences for learners in the digital age of education
ERC-Language: A student-centered learning platform for improving non-native' English Reading Comprehension skills
(IEEE, 2023-05-22) Amaya, Argemiro; Echeverria, Leovy; Cobos Pérez, Ruth; Ardila, Jorge Enrique; Departamento de Ingeniería Informática; Escuela Politécnica Superior
The main objective of this research is to propose a student-centered learning platform called ERC (English Reading Comprehension)-Language, for the developing of textual and topic-based knowledge. The proposed platform facilitates literal reading comprehension processes, the construction of linguistic and thematic knowledge, and improves students' reading performance. With the purpose to know the students' perceptions about the implementation of the ERC-Language platform, a perception study was performed. In this study 18 students of the Electronic Engineering bachelor program, from Universidad Pontificia Bolivariana (UPB Monteria Sectional) participated. The main results let us corroborate the necessity to develop a student-centered learning platform to improve non-native' ERC skills
The learning analytics system that improves the teaching-learning experience of MOOC instructors and students
(Springer Nature, 2023-05-26) Cobos Pérez, Ruth; Departamento de Ingeniería Informática; Escuela Politécnica Superior
Great learning opportunities are provided through MOOCs. However, MOOCs provide a number of challenges for students. Many students find it difficult to successfully finish MOOCs due to a variety of factors, including feelings of loneliness, a lack of support, and a lack of feedback. Additionally, the instructors of these courses are highly concerned about this situation and want to reduce these difficulties for their students. Due to the large number of students registered in these courses, this is not a simple task. To help both instructors and students, we created edX-LIMS, a learning analytics (LA) system that allows MOOC instructors to monitor the progress of their students and carry out an intervention strategy in their students’ learning thanks to a Web-based Instructor Dashboard. Furthermore, this LA system provides MOOC students with detailed feedback on their course performance as well as advice on how to improve it thanks to Web-based Learner Dashboards. This LA system have been used for more than two year in a MOOC at edX. During this period the Dashboards supported by the system have been improved, and as a result, MOOC students now appreciate the fact that they feel guided, engagement and motivated to complete the course, among other feelings. MOOC instructor have improved their student monitoring tasks and are better able to identify students who need assistance. Moreover thanks to the services that the intervention strategy supported by the LA system offer to them, now students and instructors feel that are connected
Optical Design of MAAT: an IFU for the GTC OSIRIS Spectrograph
(SPIE, 2022-08-29) Content, Robert; Prada, Francisco; Pérez, Enrique; Domínguez Tagle, Carlos; Abril, Manuela; Gómez, Gabriel; Henríquez, Kilian; Lawrence, Jon; González de Rivera Peces, Guillermo José; Goobar, Ariel; Hjorth, Jens; Pérez García, Ángeles; Agnello, Adriano; Jones, David; Departamento de Tecnología Electrónica y de las Comunicaciones; Escuela Politécnica Superior
The Mirror-slicer Array for Astronomical Transients (MAAT) is a new IFU for the OSIRIS spectrograph on the 10.4-m Gran Telescopio CANARIAS (GTC) at La Palma, spectrograph that has been recently upgraded with a new detector and moved to the Cassegrain focus. Funding has been secured to build MAAT. We present the nearly final design, its expected performances, the different options that were studied, and an analysis of the spectrograph aberrations. MAAT will take advantage of the OSIRIS mask cartridge for multi-object spectroscopy. The IFU will be in a box that will take the place of a few masks. It is based on the Advanced Image Slicer (AIS) concept as are MUSE and KMOS on the VLT (among many others). The field is 10" x 7" with 23 slices 0.305" wide giving a spaxel size of 0.254" x 0.305". The wavelength range is 360 nm to 1000 nm. The small space envelope, the maximum weight of the mask holder, and the curvature and tilt of the slit created additional design challenges. The spectral resolution will be about 1.6 times larger than with a standard slit of 0.6" because of the smaller size of the slices. All the eleven VPHs and grisms will be available to provide a broad spectral coverage with low to intermediate resolution (R=600 to 4100). To maximize the resolution of a spectrograph designed for a slit twice the width of the slices, we are in the process of measuring the wavefront of the spectrograph aberrations by using 2 out-of-focus masks with pinholes along the slit. We will then correct some of these aberrations with MAAT
Enhanced emergency operations: Leveraging UAV fleets for comprehensive response
(Springer Nature, 2024-08-04) Gómez Muñoz, Carlos Quiterio; González de Rivera Peces, Guillermo José; Garrido Salas, Javier; García Vellisca, Mariano Alberto; Gallego, Micael; Rodríguez Sánchez, María Cristina; Hardware and Control Technology Laboratory; Departamento de Tecnología Electrónica y de las Comunicaciones; Escuela Politécnica Superior
The study introduces the creation and execution of an innovative emergency management system that leverages unmanned aerial vehicles (UAVs) and a terrestrial computer system. The objective of the project is to establish a comprehensive platform that delivers instant, crucial data to first responders in emergencies, such as a building fire. To accomplish this goal, the system utilizes a variety of sensors to collect information on air quality, the detection of hazardous chemicals, and imagery through standard and thermal imaging cameras mounted on the UAVs. This collected data is then stored in a database for subsequent analysis and the generation of alert notifications. Communication between the UAV and the ground-based system is facilitated through a WiFi network architecture, ensuring the immediate transfer and display of data. This system is designed to enhance the efficiency and capabilities of emergency services in handling crisis situations
Impacto social de la investigación: Nuevos desarrollos
(2025-05-28) Rodríguez Pomeda, Jesús
El análisis del contexto general de la investigación científica en el ámbito de la economía y empresa muestra una crisis marcada por la mercantilización del conocimiento y el estancamiento de los estudios. El impulso de las políticas de Ciencia Abierta realizadas por organismos internacionales como Unesco, OECD o la Unión Europea suponen una evolución hacia un producción y evaluación del conocimiento más responsable, analizando para ello los principios RRBM (Responsible Research for Business and Management) de investigación responsable en economía y empresa
Drone collaboration using OLSR protocol in a FANET network for traffic monitoring in a smart city environment
(Springer Nature, 2023-05-01) Salazar, Franklin; Guamán-Molina, Jesús; Romero-Mediavilla, Juan; Arias-Espinoza, Cristian; Zurita, Marco; Jhonny, Carchi; Martínez García, María Sofía; Castro Martín, Ángel de; Departamento de Tecnología Electrónica y de las Comunicaciones; Escuela Politécnica Superior
A Flying Ad Hoc Network (FANET) is a new type of network derived from MANET, these networks use as nodes of unmanned aerial vehicles (UAV) that can be equipped with positioning, vision, or other systems. Currently, they present some advantages such as collaborative work between UAVs improving efficiency compared to single UAV systems. FANET’s capacity to cover large geographical areas has already positioned it as one of the best alternatives for traffic monitoring and control in smart cities. On the other hand, these networks present some challenges and problems that must be considered at the time of their design. The analysis of these networks using two-dimensional models, in some cases propose the use of a UAV in a stationary way in a specific area. Therefore, in this work we propose the implementation of a 3D model, allowing to generate a realistic environment to the movement of a UAV, using the Gauss Markov mobility model evaluating Ad Hoc routing protocols OLSR, AODV and DSDV. For the analysis of the results, the Network Simulator version 3 (Ns-3) software is used to simulate a FANET network, evaluating the efficiency of the routing protocols. The results obtained from the simulation and implementation of the proposed network showed that the OLSR protocol presents better efficiency maximizing the scope of traffic monitoring. Finally, the FANET network with 4 heterogenous drones, allowed to improve the coverage of a larger geographical area with an adequate performance of the protocol making it suitable for remote traffic monitoring in a smart city environment
Synthetic data for the mitigation of demographic biases in face recognition
(IEEE, 2024-03-01) Melzi, Pietro; Rathgeb, Christian; Tolosana Moranchel, Rubén; Vera Rodríguez, Rubén; Morales Moreno, Aythami; Lawatsch, Dominik; Domin, Florian; Schaubert, Maxim; Departamento de Tecnología Electrónica y de las Comunicaciones; Escuela Politécnica Superior
This study investigates the possibility of mitigating the demographic biases that affect face recognition technologies through the use of synthetic data. Demographic biases have the potential to impact individuals from specific demographic groups, and can be identified by observing disparate performance of face recognition systems across demographic groups. They primarily arise from the unequal representations of demographic groups in the training data. In recent times, synthetic data have emerged as a solution to some problems that affect face recognition systems. In particular, during the generation process it is possible to specify the desired demographic and facial attributes of images, in order to control the demographic distribution of the synthesized dataset, and fairly represent the different demographic groups. We propose to fine-tune with synthetic data existing face recognition systems that present some demographic biases. We use synthetic datasets generated with GANDiffFace, a novel framework able to synthesize datasets for face recognition with controllable demographic distribution and realistic intra-class variations. We consider multiple datasets representing different demographic groups for training and evaluation. Also, we fine-tune different face recognition systems, and evaluate their demographic fairness with different metrics. Our results support the proposed approach and the use of synthetic data to mitigate demographic biases in face recognition

Browse

Recent Submissions