Archivio della ricerca - Fondazione Bruno Kessler
Not a member yet
21227 research outputs found
Sort by
Le frontiere digitali della ricerca storica: digitalizzazione delle fonti, open access, intelligenza artificiale
No full textThe digitisation of historical sources has opened up new possibilities for researchers and this panel discussion examines the new situation. Two librarians specialising in historical subjects and a digital humanities researcher discuss the consequences of digitised resources spreading in open access and the role of artificial intelligence applied to the humanities. The panel discussion shows the enormous potential that digital tools offer for research, but also the need to adapt the training of researchers, librarians and archivists to this ongoing transformation
A Privacy-Preserving Framework Enhancing University Student Engagement Using Machine Learning and Gamification
No full text
Architectural Modeling and Experimental Characterization of SPAD-based Imager developed for Fast-Quantum Ghost Imaging Applications
No full textQuantum Ghost Imaging (QGI) uses quantum light properties to investigate biological samples. It involves a single quantum source generating two beams: a visible signal beam detected by an image sensor and an idler beam with sample information detected by a single-channel bucket detector. Temporal correlations between these two detectors are exploited to create a ghost image of the sample at the target wavelength without requiring an expensive custom detector. This study presents a Monte Carlo simulation model of a Single Photon Avalanche Diode (SPAD) based array for QGI microscopy highlighting the key features and major limitations. A 100X100-pixel array prototype is presented and characterized showing a less than 3% false-event rate and an average correlation window ranging between 3 and 7.8 ns with 0.4 and 0.6 ns standard deviation respectively. A smart zero-suppression readout allows fast QGI up to 80 kframe/s
The 2023 Dengue Outbreak in Lombardy, Italy: A One-Health Perspective
No full textIntroduction: Here we reported the virological, entomological and epidemiological characteristics of the large autochthonous outbreak of dengue (DENV) occurred in a small village of the Lombardy region (Northern Italy) during summer 2023. Methods: After the diagnosis of the first autochthonous case on 18 August 2023, public health measures, including epidemiological investigation and vector control measures, were carried out. A serological screening for DENV antibodies detection was offered to the population. In the case of positive DENV IgM, a second sample was collected to detect DENV RNA and verify seroconversion. Entomological and epidemiological investigations were also performed. A modeling analysis was conducted to estimate the dengue generation time, transmission potential, distance of transmission, and assess diagnostic delays. Results: Overall, 416 subjects participated to the screening program and 20 were identified as DENV-1 cases (15 confirmed and 5 probable). In addition, DENV-1 infection was diagnosed in 24 symptomatic subjects referred to the local Emergency Room Department for suggestive symptoms and 1 case was identified through blood donation screening. The average generation time was estimated to be 18.3 days (95% CI: 13.1-23.5 days). R0 was estimated at 1.31 (95% CI: 0.76-1.98); 90% of transmission occurred within 500m. Entomological investigations performed in 46 pools of mosquitoes revealed the presence of only one positive pool for DENV-1. Discussion: This report highlights the importance of synergic surveillance, including virological, entomological and public health measures to control the spread of arboviral infections
First results of Back-side Illuminated SiPM for VUV/NUV light detection fabricated at Fondazione Bruno Kessler
No full textAdvancements in 3D interconnecting technologies have significantly contributed to the emergence of a new generation of Silicon Photomultipliers (SiPM), which we can refer to as hybrid devices. These devices integrate the functionalities of digital SiPMs with the exceptional performance characteristics of specialized custom technologies. In recent years, the Fondazione Bruno Kessler (FBK) has been working on the technological development of Backside Illuminated (BSI) SiPMs for Vacuum Ultraviolet (VUV) and Near Ultraviolet (NUV) light detection, particularly in applications in particle physics experiments, such as detection of scintillation from liquefied noble gases.
For this wavelength range, a BSI detection technology faces critical challenges due to silicon's low photon interaction depth (less than 100 nm for λ = 400 nm). This necessitates the complete removal of the substrate, a process that has already been successfully demonstrated at FBK and the creation of a thin active “entrance window”. Additionally, for VUV-sensitive devices, the glass carrier wafer must be removed from the entrance window, as it typically absorbs light for wavelengths shorter than 350 nm. We will present recent technological advancements in this area, including validation of the feasibility of full substrate removal and preliminary electrical performance data of the devices
Controstoria dell'alpinismo
No full textÈ una recensione lunga (11.500 caratteri) del libro di Andrea Zannini, Controstoria dell’alpinismo (Laterza 2024), ospitata in un forum insieme a un’altra recensione (in tedesco) redatta da Andrea Pojer. Nel testo il libro viene prima inquadrato nel contesto dei cambiamenti recenti della ricerca storica, esaminato nei suoi principali contenuti e la sua tesi principale viene infine discussa criticamente: fino a che punto si può parlare sensatamente di un “alpinismo” prima dell’alpinismo
Large Language Models are Strong Audio-Visual Speech Recognition Learners
Get PDFMultimodal large language models (MLLMs) have recently become a focal point of research due to their formidable multimodal understanding capabilities. For example, in the audio and speech domains, an LLM can be equipped with (automatic) speech recognition (ASR) abilities by just concatenating the audio tokens, computed with an audio encoder, and the text tokens to achieve state-of-the-art results. On the contrary, tasks like visual and audio-visual speech recognition (VSR/AVSR), which also exploit noise-invariant lip movement information, have received little or no attention. To bridge this gap, we propose Llama-AVSR, a new MLLM with strong audio-visual speech recognition capabilities. It leverages pre-trained audio and video encoders to produce modality-specific tokens which, together with the text tokens, are processed by a pre-trained LLM (e.g., Llama3.1-8B) to yield the resulting response in an auto-regressive fashion. Llama-AVSR requires a small number of trainable parameters as only modality-specific projectors and LoRA modules are trained whereas the multi-modal encoders and LLM are kept frozen. We evaluate our proposed approach on LRS3, the largest public AVSR benchmark, and we achieve new state-of-the-art results for the tasks of ASR and AVSR with a WER of 0.79% and 0.77%, respectively. To bolster our results, we investigate the key factors that underpin the effectiveness of Llama-AVSR: the choice of the pre-trained encoders and LLM, the efficient integration of LoRA modules, and the optimal performance-efficiency trade-off obtained via modality-aware compression rates
Quantifying infectious disease epidemic risks: A practical approach for seasonal pathogens
No full textFor many infectious diseases, the risk of outbreaks varies seasonally. If a pathogen is usually absent from a host population, a key public health policy question is whether the pathogen's arrival will initiate local transmission, which depends on the season in which arrival occurs. This question can be addressed by estimating the "probability of a major outbreak" (the probability that introduced cases will initiate sustained local transmission). A standard approach for inferring this probability exists for seasonal pathogens (involving calculating the Case Epidemic Risk; CER) based on the mathematical theory of branching processes. Under that theory, the probability of pathogen extinction is estimated, neglecting depletion of susceptible individuals. The CER is then one minus the extinction probability. However, as we show, if transmission cannot occur for long periods of the year (e.g., over winter or over summer), the pathogen will most likely go extinct, leading to a CER that is equal (or very close) to zero even if seasonal outbreaks can occur. This renders the CER uninformative in those scenarios. We therefore devise an alternative approach for inferring outbreak risks for seasonal pathogens (involving calculating the Threshold Epidemic Risk; TER). Estimation of the TER involves calculating the probability that introduced cases will initiate a local outbreak in which a threshold number of cumulative infections is exceeded before outbreak extinction. For simple seasonal epidemic models, such as the stochastic Susceptible-Infectious-Removed model, the TER can be calculated numerically (without model simulations). For more complex models, such as stochastic host-vector models, the TER can be estimated using model simulations. We demonstrate the application of our approach by considering chikungunya virus in northern Italy as a case study. In that context, transmission is most likely in summer, when environmental conditions promote vector abundance. We show that the TER provides more useful assessments of outbreak risks than the CER, enabling practically relevant risk quantification for seasonal pathogens