e-Repositori UPF :: Articles (Departament de Tecnologies de la Informació i les Comunicacions)

Recent Submissions

Beyond the adult mind: a developmental framework for predictive processing in infancy
(Wiley, 2025) Ward, Emma K.; Rutar, Danaja; Zaadnoordijk, Lorijn; Poli, Francesco; Hunnius, Sabine
Predictive Processing has been proposed as the single unifying computation underlying all of cognition, and proponents argue that all psychological phenomena can be explained as consequences of this principle. This theoretical framework has inspired many cognitive scientists and neuroscientists, but it currently has no developmental mechanism that would explain how infants begin to perceive and learn about the world. Rather, it treats human cognition as if it exists in a fully developed adult with a history of observations and world knowledge. In its current formulation, Predictive Processing only allows for perception of incoming stimuli given the existence of expectations based on previous experiences and as such does not allow for an infant to ever make a first observation. In this paper, we propose a possible starting point from which the infant can begin to develop predictive models, as well as a toolkit necessary to allow the infant to perform the range of cognitive operations on predictive models necessary for learning. The starting point we propose is a set of low-precision, low level-of-detail predictions with little or no hierarchical structure, which is very rapidly updated to reflect the infant's early environment. The toolkit contains a range of operations referred to collectively as structure learning, which are applied to models in order to allow for building adult-like hierarchical models. These modifications are necessary for developmental scientists to be able to adopt the Predictive Processing framework and benefit from its advantages, but also for Predictive Processing to be able to explain all human cognition, which inherently must include development.
Infants have rich visual categories in ventrotemporal cortex at 2-months of age
(Nature Research, 2026) O'Doherty, Cliona; Dineen, Áine T.; Truzzi, Anna; King, Graham; Zaadnoordijk, Lorijn; Harrison, Keelin; D'Arcy, Enna-Louise; White, Jessica; Caldinelli, Chiara; Holloway, Tamrin; Kravchenko, Anna; Diedrichsen, Jörn; Tarrant, Ailbhe; Byrne, Angela T.; Foran, Adrienne; Molloy, Eleanor J.; Cusack, Rhodri
What are the foundations of visual categories in the human brain? Although infant looking behavior characterizes the development of overt categorization, it cannot measure neural representation or distinguish the underlying mechanism. For this, we need rich neuroimaging from young infants and the capacity to apply advanced computational models of vision. In this study, we conducted an awake functional magnetic resonance imaging (fMRI) study of more than 100 2-month-old infants, with follow-ups at 9 months, finding that categorical structure is present in high-level visual cortex from 2 months of age. This precedes its emergence in lateral visual cortex, suggesting non-hierarchical development of category representations. A deep neural network model aligned with infants’ representational geometry, indicating that the features comprising infants’ category template span a range of complexities and can be learned from the statistics of visual input. Our results reveal the existence of complex function in ventral visual cortex at 2 months of age and describe the early development of category perception.
ChatGPT relies more heavily on consonants than on vowels to recognize words
(Ubiquity Press, 2026) Toro Soto, Juan Manuel, 1976-
Humans develop biases during language learning. For example, we rely more heavily on consonants than on vowels to identify words. Advances on artificial intelligence have allowed the development of proficient large language models that sometimes mimic humans' language use. They do so by tracking regularities in natural language datasets that are used to train them. Here we test the hypothesis that tracking such regularities is enough for the emergence of responses that resemble the consonant bias. We asked ChatGPT which of two nonsense words (one with a vowel and one with a consonant change) was more similar to a target word. We observed that the model uses more the consonants than the vowels to perform similarity judgments across words in the two languages that we tested (English and Spanish).
Binocular rivalry: evaluating the role of theta power as a neural index of conflict
(Wiley, 2026) Drew, Alice; San-Segundo Gonzalez, Jorge; Soto-Faraco, Salvador, 1970-; Torralba Cuello, Mireia
Binocular rivalry (BR) occurs when each eye is presented with mutually incompatible images, and the brain alternates between perceiving one image, the other or occasionally a mashup of both. Addressing a decades-old suggestion, it has been shown that the competition between alternative representations in BR induces a pattern of neural activation resembling that occurring in cognitive conflict, eventually leading to fluctuations between different perceptual outcomes in the case of steep competition. This is reflected by known signatures of conflict dynamics, namely, an increase in fronto-medial theta oscillatory power (5–7 Hz) in the EEG right before perceptual transitions and a decrease thereafter, as well as a decrease in parieto-occipital alpha oscillatory power (8–12 Hz) prior to perceptual transitions and an increase thereafter. However, according to a growing body of research, frontal activity during BR might be related to report processes rather than perception processing per se. Such conflation is related to the use of continuous report protocols. To circumvent this confound, here, we present a BR study using an onset rivalry (rather than continuous rivalry) protocol that dissociates the moment of report from the period of stimulus presentation. The findings revealed higher fronto-medial theta power for rivalrous than for non-rivalrous stimuli both resulting in equivalent perceptual classification, despite the absence of a motor confound. In addition, we found greater parieto-occipital alpha suppression for rivalrous stimuli. The results presented here advance our understanding of how cognitive conflict monitoring and resolution may influence perception in the event of competition from incompatible sensory patterns.
Automated ethical design of multi-agent reinforcement learning environments
(IOS Press, 2025) Mayoral-Macau, Arnau; Rodriguez-Soto, Manel; Marchesini, Enrico; Lopez-Sanchez, Maite; Sánchez-Fibla, Martí; Farinelli, Alessandro; Rodriguez-Aguilar, Juan Antonio
This paper introduces the Approximate Multi-Agent Ethical Embedding Process, an algorithm to ethically design reinforcement learning environments where agents learn behaviours aligned with a moral value, while pursuing their own goals. Building on Multi-Objective and Deep Reinforcement Learning, it extends a previously theory-driven method limited to small-scale problems. The new approach is tested in a scaled-up, ethically augmented version of the gathering game, demonstrating its effectiveness in managing increased complexity.
Identification of statistical critical area to discriminate osteoporotic hip fractures in women
(Elsevier, 2025) Morando, Nicole; Ruiz Wills, Carlos; Tassani, Simone
Osteoporotic hip fracture is a worldwide health problem, but its understanding is still out of reach. Finite Element models are often implemented to study the phenomenon, but the analysis of simulation's results is in discussion. The simple identification of maximum stress or strain might be misleading and only partially related to the development of the fracture. The aim of the present study is to identify regions with statistically significant differences between fractured and control patients using a rigorous methodology based on Random Field Theory. A cohort of 90 osteoporotic female subjects was used: 45 fractured and 45 controls. 3D FE models were built from Dual-energy X-ray Absorptiometry (DXA) acquisitions. The cohort included both neck and trochanteric fractures. Areas with statistical differences were selected through Random Field Theory. The suitability of the selected elements for the discrimination of a fracture event was validated through the area under the curve (AUC) methods, and binary logistic regression with leave-one-out validation. The FE models elements identified in such a way were below 7 % of the total elements. Major Principal Stress in the selected elements showed an AUC up to 0.95. Patients were classified with an accuracy of up to 84.2 %. The methodology explored focused the analysis on specific points. This approach not only allowed for reaching a relevant classification power, but also suggested a specific bone remodeling process, including reduction of variability and interacting behavior between cortical and trabecular bone. In conclusion, a novel approach to finite element model analysis is presented, showing good classification power and extraction of information about bone remodeling in osteoporotic subjects.
Influence of disc height and strain-dependent solute diffusivity on metabolic transport in patient-personalized intervertebral disc models
(Frontiers, 2025) Workineh, Zerihun Getahun; Muñoz-Moya, Estefano; Ruiz Wills, Carlos; Lialios, Dimitrios; Noailly, Jérôme
Introduction: Intervertebral disc (IVD) degeneration is a primary contributor to low back pain, with nutritional stress due to the IVD's avascularity recognized as a key factor. Solute transport within the disc relies predominantly on diffusion, which is governed by tissue morphology and mechanical deformation. However, the interplay between disc geometry, poro-mechanical strain, diffusion, and degeneration remains incompletely characterized. Previous specimen-specific models have captured inter-subject variability in metabolite transport, but the isolated effects of disc height and degeneration-dependent material composition have not been systematically assessed. Moreover, although strain-dependent diffusion coefficients are commonly modeled as porosity functions, the role of intra-element diffusivity gradients (Formula presented.), arising under large deformation, has been largely overlooked. Methods: The present study focuses on poro-mechanical finite element (FE) models of three patient-personalized L4-L5 lumbar IVD geometries, representing varying heights categorized as thin, medium, and tall IVDs. Three days of physiological mechanical load cycles, comprising 8 hours of rest and 16 hours of activity, were simulated, under both 'healthy' (Pfirrmann grade 1) and degenerated (Pfirrmann grade 3) tissue conditions. Results: Simulation outcomes demonstrated that a one-third reduction in disc height (relative to medium height) led to (Formula presented.) increases in oxygen and glucose concentrations and (Formula presented.) decreases in lactate levels, particularly in the nucleus and anterior regions. Conversely, a one-third height increase resulted in (Formula presented.) reductions in oxygen and glucose and a corresponding rise in lactate levels. These deviations were more pronounced in degenerated tissues, highlighting the synergistic role of morphology and matrix integrity in determining metabolic homeostasis. Importantly, the inclusion of (Formula presented.) in the diffusion-reaction model produced negligible changes in solute concentration profiles. Discussion: These findings underscore the predominant influence of disc geometry and matrix composition on IVD metabolic homeostasis, suggesting limited relevance of the (Formula presented.) term in practical simulations. Simplified diffusion models, without (Formula presented.), may be sufficient for future IVD mechano-transport FE modeling.
Analyzing the effect of osteoporosis drug treatments on femoral strength using 3D-DXA finite elements modelling
(Elsevier, 2026) Ruiz Wills, Carlos; Qasim, Muhammad; Winzenrieth, Renaud; Di Gregorio, Silvana; Del Rio, Luis; Humbert, Ludovic; Noailly, Jérôme
Osteoporotic hip fracture represents a high social and economic burden in western countries. Pharmacological treatments aim to limit/reverse the loss of bone mineral density (BMD). BMD is monitored through dual energy X-ray absorptiometry (DXA). Biomechanical analysis, through 3D-DXA finite element (FE) femur models, has been shown to potentially improve fracture risk prediction. Yet, the capability of 3D-DXA FE simulations to capture the effects of pharmacological treatments on bone strength remains unexplored. Thus, this study aims to evaluate simulated changes in bone strength in subjects with different osteoporosis treatments using 3D-DXA FE models. A cohort of 155 subjects was used to generate the patient-specific FE models. Osteoporosis treatments included Alendronate (AL, n = 54), Denosumab (DMAB, n = 33), Teriparatide (TPTD, n = 31), and Naïve (NAÏVE, n = 37). Bone was modelled as BMD-dependent elasto-plastic material. Lateral fall was simulated, and bone FE-strength changes from baseline were assessed. Integral FE-strength significantly increased by 3.1% and 4.0% in the AL and DMAB groups, respectively. Trabecular and cortical FE-strength significantly increased by 2.2% and 1.9%, respectively with DMAB. Load-bearing capacity increased in both the cortical and trabecular bone of the femoral neck with DMAB and AL, while it only increased in the trabecular bone with TPTD. 3D-DXA FE analysis might help clinicians to better monitor the effects of pharmacological treatments and potentially improve personalised treatment plans for subjects with osteoporosis.
Complex mechanical loading and pro-inflammatory cytokines in intervertebral disc degeneration
(Wiley, 2026) Crump, Katherine B.; Muñoz-Moya, Estefano; de Graaf, Kim; Bermúdez-Lekerika, Paola; Schwendener, Nicole; Zech, Wolf-Dieter; Le Maitre, Christine L.; Noailly, Jérôme; Gantenbein, Benjamin
Background: Intervertebral disc (IVD) degeneration is a major contributor to low back pain, yet its initiating factors remain unclear. While the individual effects of pro-inflammatory cytokines and mechanical loading on IVDs have been studied, their combined impact is poorly understood. This study investigated how dynamic compression and torsion interact with interleukin-1 beta (IL-1β) and its inhibitor, interleukin-1 receptor antagonist (IL-1Ra), using bovine IVDs in an ex vivo organ culture system. Methods: Whole bovine caudal IVDs were cultured for one week in a custom bioreactor applying diurnal dynamic compression (0.1–0.5 MPa) and torsion (±6°) under three media conditions: physiological, catabolic (10 ng/mL IL-1β), and inhibitory (10 ng/mL IL-1Ra). Static compression (0.1 MPa) served as control. 3 T magnetic resonance imaging (MRI) was used pre- and post-culture for imaging and segmentation using 3DSlicer. Subject-personalized finite element (FE) models were generated via morphing algorithms and coupled with a parallel network (PN) model to analyze metabolite transport and its impact on gene expression. Outcomes included disc height, glycosaminoglycan (GAG) content, qPCR, and cell metabolic activity. Results & Conclusions: Degenerative changes were detected in all treatment groups. Results of decreased disc height, hydration, and ACAN expression, alongside increased MMP-13, indicated that the applied loading was supraphysiological and induced catabolic responses. IL-1Ra, at the given dose, did not counteract degeneration. MRI-based FE modeling effectively captured patterns of tissue consolidation and degeneration, providing valuable insights into IVD responses under combined mechanical and inflammatory stress. This integrative platform highlights the importance of modeling complex IVD environments and may inform the design of improved anti-catabolic therapies.
The therapeutic loop: closed-loop epilepsy systems mirroring the read-write architecture of brain-computer interfaces
(MDPI, 2026) Montoya Gálvez, Justo; Ivankovic, Karla; Rocamora Zúñiga, Rodrigo Alberto; Principe, Alessandro
Drug-resistant epilepsy (DRE) remains a major therapeutic challenge, as a considerable proportion of epilepsy patients fail to achieve seizure control with conventional anti-seizure medications or surgical therapy. Closed-loop systems have emerged as a promising alternative, offering patient-specific, on-demand neuromodulation. Despite notable advances in the academic domain, clinical translation has stagnated, and surgical resection remains the intervention with the highest probability of achieving seizure freedom. In this review, we delineate the principal limitations currently constraining progress in epilepsy neuromodulation and conceptualise these systems as instantiations of the read-write architecture characteristic of brain–computer interfaces. The read component entails the continuous acquisition and analysis of neurophysiological signals to predict or detect imminent seizures. In contrast, the write component involves the delivery of targeted interventions to disrupt epileptiform dynamics and prevent clinical seizure manifestation. We outline the closed-loop processing pipeline, survey the current state of the art, and discuss key methodological and translational challenges, particularly in algorithm validation and long-term reliability. Finally, we address patients’ and caregivers’ perspectives on the acceptance and practical integration of such technologies. This work synthesises current advances in the field and delineates the path toward fully autonomous clinically effective closed-loop neuromodulation as a viable treatment paradigm for DRE, aiming to improve patients’ quality of life.
Saturation in two-timescale RIS beamforming
(Institute of Electrical and Electronics Engineers (IEEE), 2025) Sadeghian, Masoud; Lozano Solsona, Angel; Fodor, Gabor
This paper investigates reconfigurable intelligent surface (RIS)-assisted wireless communications under a two-timescale architecture, in which the RIS phase shifts are optimized from long-term channel statistics, eliminating per-element training and thereby slashing channel estimation overhead. It is shown that, while the power captured by the RIS scales linearly with the number of its elements, the two-timescale beamforming gain upon re-radiation towards the receiver saturates rapidly as the number of RIS elements increases, for a broad class of power angular spectra (PAS). The saturation ceiling depends on the PAS angular decay, which governs how quickly inter-element spatial correlation vanishes. Steeper decays yield stronger correlations and, hence, a higher ceiling. The implications of this saturation on the effectiveness of two-timescale RIS-assisted communications are discussed.
Variable-rate incremental-redundancy HARQ for finite blocklengths
(Institute of Electrical and Electronics Engineers (IEEE), 2025) Fu, Yu; Lozano Solsona, Angel; Yang, Hongwen
Incremental redundancy (IR) hybrid automatic repeat request (HARQ) is a staple component of modern wireless systems, instrumental for efficient and reliable low-latency communication. To further improve the performance, the blocklengths - and therefore the incremental rates - of the various transmissions can be released from being fixed. This paper optimizes these variable blocklengths in truncated IR-HARQ, with the goal of maximizing the overall throughput at any desired operating point (meaning any combination of signal-to-noise ratio and target error rate). The optimization relies on a finite-blocklength information-theoretical analysis, whereby the block error rate emerges as a function of the channel capacity and the channel dispersion. Numerical results confirm that the optimized variable-rate IR-HARQ significantly outperforms its fixed-rate counterpart, both with ideal codes and with a 5G commercial code. Additionally, a heuristic scheme that mimics the behavior of the optimized solution, but is simpler to implement, is set forth and shown to essentially attain the same performance.
Salsa, a dataset for beat estimation in salsa music
(Ubiquity Press, 2024) Gómez-Marín, Daniel; Paz, Jesús; Ospina-Caicedo, Rafael; Jordà Puig, Sergi; Díaz-Cely, Javier; Herrera, Perfecto
This paper introduces a dataset of salsa music, a genre deeply rooted in Latin American culture that is known for its intricate yet captivating rhythms, which pose challenges even for seasoned dancers. Our work involved creating a comprehensive track selection and meticulously annotating beat occurrences. The dataset comprises 124 expert-analyzed salsa songs, offering a rich resource for further beat estimation and related studies within the salsa music domain. We detail the dataset, outline the methodology carried out for compiling and validating beat annotations, and finally test two contemporary beat prediction models on the dataset. Our contributions include the establishment of a labeled dataset for beat estimation research in salsa music and a robust methodology for identifying beat occurrences. Through this work, we aspire to enrich contemporary and future studies on Latin American culture, particularly the integral aspect of salsa music, fostering rhythm analysis and other musical properties that can derive from it.
A multiRater multiorgan abdominal CT dataset for calibration analysis and uncertainty modeling in segmentation
(Nature Research, 2026) Riera-Marin, Meritxell; Kleiss, Joy-Marie; Aubanell, Anton; Antolin, Andreu; Moreno-Vedia, Juan; Rodriguez-Comas, Júlia; Okkath Krishnanunni, Sikha; May, Matthias; Garcia-Lopez, Javier; Galdran, Adrian; González Ballester, Miguel Ángel, 1973-
In medical imaging, deep learning (DL) models often struggle to delineate ambiguous structures such as tumors or organ boundaries, leading to uncertainty in defining precise contours. This challenge is amplified by inter-rater variability, where experts may disagree on boundary delineations, resulting in inconsistent segmentation outcomes. Addressing these issues requires robust algorithms capable of quantifying uncertainty, standardizing annotation practices, and improving calibration to ensure reliable predictions, particularly in multi-class and multi-rater scenarios. When models are miscalibrated and overconfident, their outputs can mislead clinical decision-making, potentially influencing radiologists to over- or under-estimate malignancy risks. The CURVAS challenge (Calibration and Uncertainty for multiRater Volume Assessment in multiorgan Segmentation) was established to address these challenges by jointly assessing uncertainty, calibration, and segmentation quality, as well as promoting clinical relevance by evaluating organ volumes while accounting for annotation variability. To support this, a dataset of 90 contrast-enhanced CT scans from University Hospital Erlangen was curated, containing pancreas, liver, and kidney segmentations annotated by three experts. This resource provides a foundation for developing and benchmarking algorithms that balance segmentation accuracy, calibration, and reliability. A quantitative analysis of the annotations shows that kidney and liver segmentations exhibit strong consistency, whereas the pancreas remains challenging, emphasizing the need for refined labeling protocols and improved training strategies.
Advances in automated fetal brain MRI segmentation and biometry: insights from the FeTA 2024 challenge
(Elsevier, 2026) Zalevskyi, Vladyslav; Marti-Juan, Gerard; Gonzalez Ballester, Miguel Ángel; Bach Cuadra, Meritxell
Accurate fetal brain tissue segmentation and biometric measurement are essential for monitoring neurodevelopment and detecting abnormalities in utero. The Fetal Tissue Annotation (FeTA) Challenges have established robust multi-center benchmarks for evaluating state-of-the-art segmentation methods. This paper presents the results of the 2024 challenge edition, which introduced three key innovations. First, we introduced a topology-aware metric based on the Euler characteristic difference (ED) to overcome the performance plateau observed with traditional metrics like Dice or Hausdorff distance (HD), as the performance of the best models in segmentation surpassed the inter-rater variability. While the best teams reached similar scores in Dice (0.81-0.82) and HD95 (2.1-2.3 mm), ED provided greater discriminative power: the winning method achieved an ED of 20.9, representing roughly a 50% improvement over the second- and third-ranked teams despite comparable Dice scores. Second, we introduced a new 0.55T low-field MRI test set, which, when paired with high-quality super-resolution reconstruction, achieved the highest segmentation performance across all test cohorts (Dice=0.86, HD95=1.69, ED=6.26). This provides the first quantitative evidence that low-cost, low-field MRI can match or surpass high-field systems in automated fetal brain segmentation. Third, the new biometry estimation task exposed a clear performance gap: although the best model reached a mean average percentage error (MAPE) of 7.72%, most submissions failed to outperform a simple gestational-age-based linear regression model (MAPE=9.56%), and all remained above inter-rater variability with a MAPE of 5.38%. Finally, by analyzing the top-performing models from FeTA 2024 alongside those from previous challenge editions, we identify ensembles of 3D nnU-Net trained on both real and synthetic data with both image- and anatomy-level augmentations as the most effective approaches for fetal brain segmentation. Our quantitative analysis reveals that acquisition site, super-resolution strategy, and image quality are the primary sources of domain shift, informing recommendations to enhance the robustness and generalizability of automated fetal brain analysis methods.
A simple and explainable model for park-and-ride car park occupancy prediction
(Springer, 2025) Kaltenbrunner, Andreas; Ferrer, Josep; Moreno, David; Gómez, Vicenç
In a scenario of growing usage of park-and-ride facilities, understanding and predicting car park occupancy is becoming increasingly important. This study presents a model that effectively captures the occupancy patterns of park-and-ride car parks for commuters using truncated normal distributions for vehicle arrival and departure times. The objective is to develop a predictive model with minimal parameters corresponding to commuter behaviour, enabling the estimation of parking saturation and unfulfilled demand. The proposed model successfully identifies the regular, periodic nature of commuter parking behaviour, where vehicles arrive in the morning and depart in the afternoon. It operates using aggregate data, eliminating the need for individual tracking of arrivals and departures. The model's predictive and nowcasting capabilities are demonstrated through real-world data from car parks in the Barcelona Metropolitan Area. A simple model extension furthermore enables the prediction of when a car park will reach its occupancy limit and estimates the additional spaces required to accommodate such excess demand. Thus, beyond forecasting, the model serves as a valuable tool for evaluating interventions, such as expanding parking capacity, to optimise park-and-ride facilities.
Analyzing news engagement on Facebook: tracking ideological segregation and news quality in the Facebook URL dataset
(Springer, 2025) Fraxanet, Emma; Kaltenbrunner, Andreas; Germano, Fabrizio; Gómez, Vicenç
The Facebook Privacy-Protected Full URLs Dataset was released to enable independent, academic research on the impact of Facebook's platform on society while ensuring user privacy. The dataset has been used in several studies to analyze the relationship between social media engagement and societal issues such as misinformation, polarization, and the quality of consumed news. In this paper, we conduct a comprehensive analysis of the engagement with popular news domains, covering four years from January 2017 to December 2020, with a focus on user engagement metrics related to news URLs in the U.S. By incorporating the ideological alignment and composite score of quality and reliability of news sources, along with users' political preferences, we construct weighted averages of ideology and quality of news consumption for liberal, conservative, and moderate audiences. This allows us to track the evolution of (i) the ideological gap in news consumption between liberals and conservatives and (ii) the average quality of each group's news consumption. We identify two major shifts in trends, each tied to engagement changes. In both, the ideological gap widens and news quality declines. However, engagement rises in the first shift but falls in the second. Finally, we contextualize these trends by linking them to two major Facebook News Feed updates. Our findings provide empirical evidence to better understand user behavior and engagement with news and their leaning and reliability during the period covered by the dataset.
Ranking for engagement: how social media algorithms fuel misinformation and polarization
(Elsevier, 2026) Germano, Fabrizio; Gomez, Vicenç; Sobbrio, Francesco
Social media are at the center of countless debates on polarization, misinformation, and even the state of democracy in various parts of the world. An essential feature of social media is their recommendation algorithm that determines the ranking of content presented to the users. This paper investigates the dynamic feedback loop between recommendation algorithms and user behavior, and develops a theoretical framework to assess the impact of popularity-based parameters on platform engagement, misinformation, and polarization. The model uncovers a fundamental trade-off: assigning greater weight to online social interactions—such as likes and shares—increases user engagement but also increases misinformation (crowding-out the truth) and polarization. Building on this insight, the analysis considers how a simple “engagement tax” on social interactions can mitigate these negative externalities by altering platform incentives in the design of profit-maximizing algorithms. The framework is extended to include personalized rankings, demonstrating that personalization further amplifies polarization. Finally, empirical evidence from survey data in Italy and the United States indicates that Facebook’s 2018 “Meaningful Social Interactions” update—which increased the emphasis on certain engagement metrics—contributed to increased ideological extremism and affective polarization.
Personalisation and profiling using algorithms and not-so-popular Colombian music: goal-directed mechanisms in music emotion recognition
(Springer, 2025) Gómez-Cañon, Juan Sebastián; Lennie, Thomas Magnus; Eerola, Tuomas; Aragón, Pablo; Cano, Estefanía; Herrera, Perfecto; Gómez, Emilia
This work investigates how personalised Music Emotion Recognition (MER) systems may lead to sensitive profiling when applied to musically induced emotions in politically charged contexts. We focus on traditional Colombian music with explicit political content, including (1) vallenatos and social songs aligned with the left-wing guerrilla Fuerzas Armadas Revolucionarias de Colombia (FARC), and (2) corridos linked to sympathisers of the right-wing paramilitary group Autodefensas Unidas de Colombia (AUC). Using data from 49 participants with diverse political leanings, we train personalised machine learning models to predict induced emotional responses -particularly negative emotions. Our findings reveal that political identity plays a significant role in shaping emotional experiences of music with explicit political content, and that emotion recognition models can capture this variation to a certain extent. These results raise critical concerns about the potential misuse of emotion recognition technologies. What is often framed as a tool for wellbeing and emotional regulation could, in politically sensitive contexts, be repurposed for user profiling. This work highlights the ethical risks of deploying AI-driven emotion analysis without safeguards, particularly among populations that are politically or socially vulnerable. We argue that subjective emotional responses may constitute sensitive personal data, and that failing to account for their sociopolitical context could amplify harm and exclusion.
Data-driven cellular mobility management via Bayesian optimization and reinforcement learning
(Institute of Electrical and Electronics Engineers (IEEE), 2026) Benzaghta, Mohamed; Ammar, Sahar; López Pérez, David; Shihada, Basem; Geraci, Giovanni
Mobility management in cellular networks faces increasing complexity due to network densification and heterogeneous user mobility characteristics. Traditional handover (HO) mechanisms, which rely on predefined parameters such as A3-offset and time-to-trigger (TTT), often fail to optimize mobility performance across varying speeds and deployment conditions. Fixed A3-offset and TTT configurations either delay HOs, increasing radio link failures (RLFs), or accelerate them, leading to excessive ping-pong effects. To address these challenges, we propose two distinct data-driven mobility management approaches leveraging high-dimensional Bayesian optimization (HD-BO) and deep reinforcement learning (DRL). While HD-BO optimizes predefined HO parameters such as A3-offset and TTT, DRL provides a parameter-free alternative by allowing an agent to select serving cells based on real-time network conditions. We systematically compare these two approaches in real-world site-specific deployment scenarios (employing Sionna ray tracing for site-specific channel propagation modeling), highlighting their complementary strengths. Results show that both HD-BO and DRL outperform 3GPP set-1 (TTT of 480 ms and A3-offset of 3 dB) and set-5 (TTT of 40 ms and A3-offset of -1 dB) benchmarks. We augment HD-BO with transfer learning so it can generalize across a range of user speeds. Applying the same transfer-learning strategy to the DRL method reduces its training time by a factor of 2.5 while preserving optimal HO performance, showing that it adapts efficiently to the mobility of aerial users such as UAVs. Simulations further reveal that HD-BO remains more sample-efficient than DRL, making it more suitable for scenarios with limited training data.

Navigate

Browse

Recent Submissions