43569 research outputs found
Sort by
Review of Kaveh Yazdani, India, modernity and the Great Divergence: Mysore and Gujarat, 17th-19th c.
No full textImagine, discover, inspire: Proceedings of the 4th international conference of the Trisomy 21 research society
No full textDown syndrome (DS) or trisomy 21 (T21) is present in a significant number of children and adults around the world and is associated with cognitive and medical challenges. Through research, the T21 Research Society (T21RS), established in 2014, unites a worldwide community dedicated to understanding the impact of T21 on biological systems and improving the quality of life of people with DS across the lifespan. T21RS hosts an international conference every two years to support collaboration, dissemination, and information sharing for this goal. In 2022, T21RS hosted an international conference in Long Beach, California, from June 9 to 12. The conference, attended by 483 people including scientists, families, self-advocates, and industry representatives from 17 countries, was a dynamic and interactive meeting that shared discoveries from international research teams. This summary highlights the scientific discoveries shared at the 4th T21RS meeting with the Imagine, Discover, Inspire theme
Self-supervised and in-context learning techniques for automated optical inspection
No full textTreball fi de màster de: Erasmus Mundus joint Master in Artificial Intelligence (EMAI)Supervisora: Lejla Batina
Co-Supervisor: Faysal BoughorbelAutomated Optical Inspection (AOI) is a family of techniques used to find defects and anomalies in electronic devices from high-quality photographs of different regions of an integrated component and its packaging. Current methods use computer vision models and image preprocessing pipelines specific to each chip design and manufacturer. As a result, the current deep learning approach for AOI requires a long retraining process whenever new devices are introduced or significant covariate shifts occur in the input image distribution. In this work, we adapt and evaluate different pre-training techniques (DINO, iBOT, and MAE) for small vision transformers (ViT and FasterViT) to streamline the design process of AOI semantic segmentation models and shorten the training time needed to adapt the models to new input conditions. We use a custom, relatively small dataset for model pre-training with only 7000 unlabeled images, showing how the pre-training strategies perform well in small data regimes. Furthermore, we introduce a set of retrieval-based scene understanding techniques to solve the task of semantic segmentation of wire-bonded devices with virtually no training time in labeled data. Our results demonstrate how our custom pre-trained encoders and retrieval strategies outperform comparable convolutional architectures pre-trained using full supervision in semantic segmentation, both in speed and quality, when training time is constrained. Moreover, we show how our proposed image retrieval strategies generalize to existing ViT models pretrained on different datasets, and how the techniques can be used to predict images of a single device and produce high-quality segmentation masks using a relatively small number of labeled training images. Finally, we show how the retrieval strategies outperform fine-tuned, convolutional encoder-decoder models in the context of out-of-distribution, unseen images
Supporting teachers' value-sensitive reflections on the cost–benefit dynamics of technology in educational practices
No full textThis paper explores the application of a benefits versus costs reflection approach within non-university teaching environments, grounded in the principles of Value-Sensitive Design. Aimed at integrating human values systematically into the adoption of digital educational tools, this study involved 136 in-service school teachers across various workshops in Catalonia. Through the use of a structured customisable worksheet, educators critically self-evaluated their feelings about both the benefits and costs associated with the use of digital technologies in their teaching practices. The study found that the approach was meaningful to the teachers, who were able to adapt the use of the workshop to their cases. The positive reception by teachers suggests not only a satisfactory level of usability and utility of the approach but also their agreement with the need to integrate related strategies in their training, learning design and community debate processes.This work is co-funded by MICIU/AEI/10.13039/501100011033, Agencia Estatal de Investigación, Ministerio de Ciencia, Innovación y Universidades, Gobierno de España (PID2020-112584RB-C33, PID2023-146692OB-C33, CEX2021-001195-M) and the Government of Catalonia (through SGR 00930 and NextGeneration DICOLED EDU128/23). DHL (Serra Húnter) also acknowledges the support by ICREA under the ICREA Academia programme
MusGO: a community-driven framework for assessing openness in music-generative AI
No full textComunicació presentada al 26th International Society for Music Information Retrieval Conference (ISMIR 2025), celebrada a Daejeon (Korea) del 21 al 25 de setembre del 2025Since 2023, generative AI has rapidly advanced in the music domain. Despite significant technological advancements, music-generative models raise critical ethical challenges, including a lack of transparency and accountability, along with risks such as the replication of artists' works, which highlights the importance of fostering openness. With upcoming regulations such as the EU AI Act encouraging open models, many generative models are being released labelled as 'open'. However, the definition of an open model remains widely debated. In this article, we adapt a recently proposed evidence-based framework for assessing openness in LLMs to the music domain. Using feedback from a survey of 110 participants from the Music Information Retrieval (MIR) community, we refine the framework into MusGO (Music-Generative Open AI), which comprises 13 openness categories: 8 essential and 5 desirable. We evaluate 16 state-of-the-art generative models and provide an openness leaderboard that is fully open to public scrutiny and community contributions. Through this work, we aim to clarify the concept of openness in music-generative AI and promote its transparent and responsible development.This work has been supported by IA y Música: Cátedra en Inteligencia Artificial y Música (TSI-100929-2023-1), funded by the Secretaría de Estado de Digitalización e Inteligencia Artificial and the European Union-Next Generation EU, and IMPA: Multimodal AI for Audio Processing (PID2023-152250OB-I00), funded by the Ministry of Science, Innovation and Universities of the Spanish Government, the Agencia Estatal de Investigación (AEI) and cofinanced by the European Union. We thank our colleagues at the Music Technology Group at Universitat Pompeu Fabra for their thoughtful insights, constructive discussions and active engagement throughout the development of this wor
Dels auzels qui perteno ad ornament del ayre’: Remarques autour du livre XII de l’‘Elucidari’, traduction occitane du ‘De proprietatibus rerum’
No full textDatant de la deuxième moitié du 14e siècle, le ms. 1029 de la Bibliothèque Sainte-Geneviève de Paris est le seul manuscrit conservé de la version occitane du De proprietatibus rerum de Barthélemy l'Anglais (DPR). Ce somptueux codex a fait partie de la bibliothèque de Gaston III Fébus, comte de Foix-Béarn entre 1343 et 1391, destinataire et sans doute commanditaire de cette traduction. La place réservée aux animaux et en particulier aux oiseaux dans cette encyclopédie médiévale est fort importante. Intégré à une recherche sur la langue et le vocabulaire de l'Elucidari, cet article offre quelques réflexions autour du texte et du lexique ornithologique du livre XII (De las naturas et proprietatz dels auzels) de la version occitane du De proprietatibus rerum
Differentially private fine-tuning of self-supervised learning models for human activity recognition on wearables
No full textTreball fi de màster de: Master in Intelligent Interactive SystemsSupervisor: Mario Ceresa
Co-Supervisor: Vicenç GómezThe widespread use of wearables in health applications has advanced personalized Human Activity Recognition (HAR), but it also introduces privacy challenges under regulations such as the European Health Data Space (EHDS). This thesis explores fine-tuning strategies for self-supervised learning models on wearable sensor data to balance user privacy and model utility within the stringent EU regulatory landscape. We employ Differentially Private Stochastic Gradient Descent (DP-SGD) to fine-tune the pre-trained HarNet10 model on the PAMAP2 dataset, evaluating two distinct strategies: classifier head fine-tuning and full model fine-tuning. Our two sequential experimental design, first investigates the privacy-utility trade off between the two strategies, revealing that classifier head fine-tuning consistently outperforms the full model approach by maintaining higher accuracy and F1-scores. This strategy better preserves the rich, pre-trained representations in the feature extractor, mitigating the impact of DP-SGD’s noise. Second, an empirical privacy evaluation using a membership inference attack confirms these findings. The differentially private classifier head model demonstrates robust protection, reducing the attack’s success to near random guessing (AUC score of 0.532) compared to the vulnerable non-private baseline (AUC of 0.690), thus aligning theoretical guarantees with practical resilience. These results confirm that classifier head fine-tuning with DP-SGD offers an optimal privacy-utility balance for HAR tasks compared to its full model variant. Our study contributes a validated framework for developing trustworthy AI in wearables,
demonstrating that effective privacy can be achieved by fine-tuning a small fraction (4.83%) of a foundational model’s parameters. This research provides a practical reference for building secure, privacy preserving solutions that align with EU regulations
¿Cómo movilizar las audiencias en medios de comunicación? Modelos de Activación de la Relevancia y el Engagement de Audiencias en Medios de Comunicación (MAREA)
No full textLA MAREA ofrece una metodología práctica para que los medios de comunicación evalúen y refuercen su capacidad de relevancia y engagement, elementos cruciales para su viabilidad conómica a largo plazo. Este marco metodológico operacionaliza la sostenibilidad mediática a través de tres dimensiones fundamentales: reputación, visibilidad y fidelización de audiencias. MAREA se basa en un conjunto de recomendaciones de expertos — 337 acciones y 226 métricas — sintetizadas en siete parámetros de acción: Escala, Orientación al Usuario, Desarrollo Organizacional, Abandono, Valor Diferencial, Transparencia y Reconocimiento. Estos parámetros, junto con métricas claras, permiten a los medios de comunicación —especialmente a los de tamaño reducido y recursos escasos— identificar áreas de mejora y elaborar planes de acción concretos. El capítulo detalla cada parámetro, ilustra su aplicación con un caso hipotético y subraya la importancia de reinterpretar las métricas más allá de la mera visibilidad, integrando la reputación y la fidelización de la audiencia, lo que facilita una gestión proactiva y una mayor resiliencia en un entorno mediático cambiante.Este trabajo forma parte del proyecto “Parámetros y estrategias para incrementar la relevancia de los medios y la comunicación digital en la sociedad: curación, visualización y visibilidad (CUVICOM)”. Ayuda PID2021-123579OB-I00 financiada por MICIU/AEI/10.13039/501100011033 y por FEDER, UE
Adherence to a Mediterranean diet and leisure-time physical activity are associated with reduced initiation of antidepressant, anxiolytic, antipsychotic and antiseizure drug use in older adults: a cohort study
No full textBackground: We explored how adherence to the Mediterranean diet (MedDiet) and leisure-time physical activity (LTPA) impact psychoactive medication use in older adults. Methods: We assessed the cumulative MedDiet adherence and LTPA's impact on mental health medication initiation in older individuals at high risk of chronic disease. Associations between the cumulative average of MedDiet adherence (per one-point increase in the adherence score) and LTPA (per increase in 20 metabolic equivalents of task-minute/day [METs-min/day]) with drug initiation were assessed by multivariable Cox regressions. We explored non-linear exposure-outcome associations using smoothed cubic splines and the multiplicative interaction between MedDiet and LTPA. Results: A total of 5940-6896 participants (mean age 67, 58% women) over 4.2-4.7 years, each point increase in MedDiet adherence decreased the initiation of antidepressants by 23-28% (HR 0.72, 95% CI 0.67-0.77), anxiolytics (HR 0.75, 0.70-0.81), antipsychotics (HR 0.77, 0.65-0.91), and antiseizures (HR 0.77, 0.69-0.85). Associations for anxiolytics and antiseizures were strong at low MedDiet adherence levels. Relationships between LTPA and initiation of antidepressants and anxiolytics were linear in the lowest LTPA values (0-150 METs-min/day); every 20 METs-min/day increases were associated with 20% lower risk of initiating antidepressants (HR 0.80, 0.75-0.86) and 15% less risk in anxiolytics (HR 0.85, 0.79-0.90). Association with antiseizures was linear (+20 METs-min/day: HR 0.96, 0.94-0.99), and no associations were found for antipsychotics. High MedDiet adherence (≥10) and LTPA (≥150 METs-min/day) reduced psychoactive drug initiation by 42%-59%. Combination was additive for antidepressants, antipsychotics and antiseizures and synergistic for anxiolytics. Conclusions: MedDiet and LTPA adherence reduced psychoactive drugs initiation in older adults