Archivio della ricerca - Fondazione Bruno Kessler
Not a member yet
21227 research outputs found
Sort by
Search for Magnetic Monopole Pair Production in Ultraperipheral Pb+Pb Collisions at √sNN=5.36 TeV with the ATLAS Detector at the LHC
No full text:This Letter presents a search for highly ionizing magnetic monopoles in 262 μb^{-1} of ultraperipheral Pb+Pb collision data at sqrt[s_{NN}]=5.36 TeV collected by the ATLAS detector at the LHC. A new methodology that exploits the properties of clusters of hits reconstructed in the innermost silicon detector layers is introduced to study highly ionizing particles in heavy-ion data. No significant excess above the background, which is estimated using a data-driven technique, is observed. Using a nonperturbative semiclassical model, upper limits at 95% confidence level are set on the cross section for pair production of monopoles with a single Dirac magnetic charge in the mass range of 20-150 GeV. Depending on the model, monopoles with a single Dirac magnetic charge and mass below 80-120 GeV are excluded
EFL-PEFT: A communication Efficient Federated Learning framework using PEFT sparsification for ASR
Get PDFFederated Learning (FL) has garnered substantial interest in training different speech-based tasks (e.g. automatic speech recognition (ASR), and other speech classification tasks): recently, fine-tuning pre-trained self-supervised models for different speech-based tasks has shown promising performance and been successfully applied in FL settings. Nevertheless, fine-tuning these architectures is computationally burdensome and not affordable in several real-time settings. Moreover, the communication costs of transferring all the model parameters for the aggregation stage is critically high. As an alternative approach, parameter-efficient fine-tuning (PEFT) approaches provide promising performance without changing the backbone of the pre-trained model. PEFT has been fruitfully applied, in a variety of flavours, for ASR in central training configurations while only few works investigate its use in FL settings. In this paper, we consolidate the use of PEFT for ASR with pre-trained models, demonstrating that it enables efficient FL reducing the amount of parameters to share with respect to full fine-tuning. We also explore combining PEFT with sparsification methods to further reduce communication cost by transmitting only a fraction of the adapter parameters. Additionally, we show that agglomerating adapters using "FedAvg" is compatible with differential privacy, aligning with trends observed in other domains. Our proposed approach is supported by experimental analysis on ASR using two public datasets, as well as on intent classification tasks
Impact of routine prophylaxis with monoclonal antibodies and maternal immunisation to prevent respiratory syncytial virus hospitalisations, Lombardy region, Italy, 2024/25 season
No full textBackgroundRespiratory syncytial virus (RSV) is a leading cause of hospitalisation in children worldwide. Recent regulatory approval of monoclonal antibody (mAb) nirsevimab for infants and the RSVpreF vaccine for pregnant women offers promising approaches to mitigate RSV-associated morbidity.AimTo evaluate potential impacts of routine prophylactic campaigns (mAbs targeting infants or maternal vaccination) introduced in the 2024/25 season on hospitalisations from RSV lower respiratory tract infections in Lombardy, Italy.MethodsWe used a catalytic model informed by data from pre-COVID-19 pandemic (before 2020) and post-pandemic periods (until 2022) to quantify the number of cases and hospitalisations that could be averted by seasonal nirsevimab administration to infants and RSVpreF maternal vaccination, considering changes in susceptibility caused by reduced RSV circulation during the pandemic.ResultsAs a marked proportion of RSV hospitalisations occurs in infants aged ≤ 1 year, seasonal mAb administration to 80% of newborns (uptake levels observed in Spain) was estimated to avert 50.2% (95% CI: 43.5-55.8) of hospitalisations in the total population. Coverage levels close to those observed for childhood vaccines (95%) could result in an additional average 18% reduction in hospitalisations. Vaccination of 65% of pregnant women, resembling the diphtheria-tetanus-pertussis vaccine coverage in Lombardy for this population, was estimated to avert 30.5% (95% CI: 19.6-39.7) of hospitalisations. At influenza vaccine coverage (12%), less than 8% of hospitalisations could be averted by maternal immunisation.ConclusionRoutine nirsevimab administration to infants demonstrates clear potential to reduce RSV-associated hospitalisations. Maternal immunisation can help in achieving high protection in at-risk populations
Collaborative forecasting of influenza-like illness in Italy: The Influcast experience
No full textCollaborative hubs that integrate multiple teams to generate ensemble projections and forecasts for shared targets are now regarded as state-of-the-art in epidemic predictive modeling. In this paper, we introduce Influcast, Italy’s first epidemic forecasting hub for influenza-like illness. During the 2023/2024 winter season, Influcast provided 20 rounds of forecasts, involving five teams and eight models to predict influenza-like illness incidence up to four weeks in advance at the national and regional administrative level. The individual forecasts were synthesized into an ensemble and benchmarked against a baseline model. Across all models, the ensemble most frequently ranks among the top performers at the national level considering different metrics and forecasting rounds. Additionally, the ensemble outperforms the baseline and most individual models across all regions. Despite a decline in absolute performance over longer horizons, the ensemble model outperformed the baseline in all considered horizons. These findings show the importance of multimodel forecasting hubs in producing reliable short-term influenza-like illnesses forecasts that can inform public health preparedness and mitigation strategies
Forecasting norovirus cases on cruise ships to support outbreak management on board
No full textBackground: Norovirus outbreaks on cruise ships are a significant threat to the cruising industry. Mathematical models have the potential to leverage routinely collected syndromic surveillance data on board to provide insight into outbreak evolution. Methods: We used historical data from seven norovirus outbreaks occurred in 2011-2013, totalling 359 diagnosed cases, to assess the performance of automated forecasts in real-time. We compared the performance of a set of alternative models on three endpoints (the number of cases by symptom onset time, by diagnosis date, and the total number of cases until the end of the cruise), using the logarithmic score (logS), the ranked probability score (RPS), and the 95% coverage. Results: We found that the best forecasting performance was given by a model that includes both superspreading and the effect of case isolation. This model had in most cases a better score than that of a baseline model assuming constant incidence; this happened in 59-70% of data points when assessed using the logS and 53-57% with the RPS (depending on the considered endpoint). The best model also had the highest coverage over all endpoints. Its added value was especially evident for longer forecasting horizons, with an improvement in performance for up to 78% of data points, both according to the logS and the RPS. Conclusions: Simple mathematical models integrating key mechanisms of norovirus transmission can help predict the number of cases on board. This knowledge can be automatized in syndromic surveillance systems to support decision making for the management of outbreaks
Toward Proactive Dialogic AI Agents
No full textThis paper introduces an ongoing research on the development of a proactive dialogic AI agent, focusing on enhancing an LLM’s pragmatic competence in goal-oriented dialogues. We investigate proactivity as a collaborative behaviour that enables to provide relevant and useful information that has not been explicitly requested, thereby improving interaction efficiency and dialogue naturalness. Our approach is grounded in a corpus-based analysis of proactive behaviours in human-human dialogues across five goal-oriented dialogue corpora, leading to the creation of the D-Pro Corpus, a manually annotated resource for studying proactivity. Its analysis provides information on qualitative and distributional features of proactivity in human dialogues, as well as clues on recurrent linguistics structures that co-occur with the display of proactive behaviours. We then leverage the D-Pro Corpus to evaluate the performance of a GPT-4o model in proactivity annotation, addressing the task by providing a 4-turns context size and by targeting the last utterance for proactivity prediction. By experimenting with parameter setting and prompt configurations, we assess the model’s performance across multiple dialogue corpora, obtaining encouraging results toward human-like performance, particularly with the NESPOLE! corpus. We propose to advance our research by exploring the potential of open-source models for cost-effective, large-scale automatic annotation of unlabelled dialogic data. As a final step we plan to use the large-scale annotated corpus to instruction-tune an open model, expanding its pragmatic competence for the development of more proactive and contextually aware dialogic AI system and more natural human-machine conversation
Pilot Line "From lab to fab" per sviluppi di sensori e dispositivi
No full textLa fabbricazione di sensori avviene in laboratori detti cleanroom, ambienti dove
temperatura, umidità e pressione sono mantenuti in intervalli controllati per non
interferire coi processi di lavorazione, e dove un sistema di filtraggio continuo
dell’aria limita la presenza di particelle di sporco (anche di dimensioni inferiori al
millesimo di millimetro) che, se depositate sulla superficie delle fette di semiconduttore,
possono compromettere la qualità dei dispositivi.
Il termine pilot line si focalizza in questo contesto su cleanroom pubbliche in
grado di sviluppare dispositivi micro-nano elettronici completi per permettere
una validazione di mercato, curandone il design in linea con le necessità dell’utilizzatore
finale, lo sviluppo, l’ingegnerizzazione e il controllo di qualità
Stochastic nature of large‐scale contact printed ZnO nanowires based transistors
Get PDFPrinting technology holds great potential for resource-efficient development of electronic devices and circuits. However, even after decades of research, achieving uniformly responding nanowires (NWs) based printed devices is still a challenge. To date, there is no design rule that clearly guides the fabrication of NW ensemble-based field-effect transistors (FETs) and the variables that influence device-level uniformity remain unclear. The lack of fundamental understanding severely limits the large-scale and very large-scale integration (LSI and VLSI). Herein this longstanding issue is addressed with a holistic approach that starts with optimization of the synthesis of ZnO NWs, their printing, and further processing to fabricate transistors with uniform responses (e.g., on-state current, threshold voltage). Monte Carlo simulation based on statistical analysis of printed ZnO NWs is carried out to develop a probabilistic framework that can predict the large-scale performance of FETs. As a proof of concept, inverter circuits have been developed using printed ZnO NWs based FETs. This work provides a valuable toolkit to handle the stochastic nature of FETs based on printed ZnO NW ensemble, which can be used for neuromorphic integrated circuit in the future
Towards a Fine-Grained Threat Model for Video-Based Remote Identity Proofing
No full textThe attack landscape against video-based face verification is rapidly evolving, thus leading to increased opportunities of impersonation and identity theft within remote identity proofing processes. To support reliable security and risk analyses, we provide an extended threat model composed of threats and security controls for the face acquisition phase. Special emphasis is devoted to recent advancements in video synthesis and physical rendering, as well as diversified approaches for liveness detection
What is more likely in orthorexia nervosa: perfectionism or OC symptoms? A bayesian method in clinical and non-clinical samples
No full textPurpose
Orthorexia Nervosa (ON) exhibits specific features that may overlap with Obsessive-Compulsive Disorder (OCD), Perfectionism Striving (PS), and Perfectionism Concern (PC). While previous literature has shown predictiveness in different characteristics of ON, this study aimed to determine if PS, PC and OCD symptoms could predict ON dimensions in at-risk populations using Bayesian models.
Method
The study enrolled 622 individuals from three different at-risk populations: people who were following treatment for an Eating Disorder (Patients), people who were following a diet (Dieters) and University students with a degree in medicine or nursing (Students).
Results
Using Bayesian probabilistic methods and considering group factors, the model was enhanced highlighting that ON characterised Patients, Dieters, and Students. The predictiveness of OC features, PS, and PC in different ON characteristics was confirmed in each group, and different patterns were observed in the three groups. Regarding problems related to ON, predictors were higher in Patients, followed by Dieters and Students. Posterior Predictive Distribution (PPD) showed that almost 50% of Patients incur ON-related problems. In ON knowledge, Patients and Dieters are very similar. When feelings related to ON were considered, Patients and Dieters showed different interactions.
Conclusion
Clinicians should consider that one in two patients suffering from EDs might develop ON-related problems. People on a diet could show similar patterns of symptoms to patients in ON knowledge and feelings. Finally, our results confirm that perfectionism represents a risk factor for ON in each group considered