Antonia Alomar Adrover and Emma Fraxanet shared the 2023 Maria de Maeztu Prize for Interdisciplinarity in Research, part of the DTIC Doctoral Student Workshop. The award was selected by a commission composed by members of the MdM Executive Board and the UPF Center for Studies on Planetary Wellbeing.
Antonia's research
Perinatal 3D face reconstruction and analysis for early stage diagnosis of craniofacial anomalies
Abstract: Even for expert clinicians some of syndromes are difficult to identify as some of the patterns are really subtle and there exists thousands of genetic syndromes with different symptoms and phenotypes. Our aim is to help clinicians in the detection of abnormal faces with some underlying condition in early stages of life. We propose performing a face base genetic screening using 3D ultrasound and 2D images. The idea is to develop a multimodal deep neural network that reconstructs the facial 3d morphology and analyze if there exist or not an abnormality.
DTIC-UPF supervisors: Gemma Piella Fenoy (Simulation, Imaging and Modelling for Biomedical Systems, Barcelona Med-Tech) and Federico Sukno (Cognitive Media Technologies Research Group)
External collaborators:
Institut Hospital del Mar d'Investigacions Mèdiques (IMIM), BCN Natal, and Children's National Hospital
Emma's research
Polarization Dynamics from Signed Networks: an Application to Misinformation Crowdsourcing on Twitter
Abstract: The study of online polarization over time presents challenges such as identifying appropriate time scales and developing a framework that can accurately describe the various mechanisms at play. The aim is not only to measure the degree of ideological separation between communities but also to understand their evolution and response to endogenous and exogenous events. Inspired by balance theory, we define a signed network from aggregated votes which is partitioned according to maximal balance (Aref et al., 2016). We extend the existing computational framework by proposing a new Signed Polarization Index (SPI), based on quantifying the relevance of individual votes that are not in agreement with our optimal partition within a given time window. Unlike previous balance indices, this index has a natural applicability to temporal networks and displays stronger statistical properties. We apply this framework to Birdwatch, a crowd-based fact-checking platform linked to Twitter in which users add notes to Tweets stating their reliability. As a proxy for ground truth communities (Republicans, R, and Democrats, D), we infer political ideology of the original tweet's authors (Barberá et al., 2015) and analyse the aggregated notes. We find that Birdwatch is strongly polarized and fluctuates significantly over time. We analyse the existence of such variations by means of internal cohesion within a group (cohesiveness) and external division amongst groups (divisiveness) and find that polarization peaks have a complex interplay between these mechanisms. For example, peaks in SPI that map to Covid-19 vaccination discussions are related to higher divisiveness between the groups. This work has important implications to improve our understanding of online polarization and find conciliation strategies, in order to create platforms with healthier discussions.
DTIC-UPF supervisor: Vicenç Gómez (AI and machine learning research group)
External collaborators: David Garcia (Computational Social Sciences, University of Konstanz), Max Pellert (Cognitive sciences/Economics, University of Mannheim), Simon Schweighofer (Sociology, XJTLU Suzhou), Fabrizio Germano (Economics, UPF) Gael Le Mens (Economics, UPF) Jula Lühring (Comm. Science, CSH Vienna) Jana Lasser (Physics, TU Graz)
