Archivio della ricerca - Fondazione Bruno Kessler
Not a member yet
21227 research outputs found
Sort by
The impact of wearing a heart rate monitoring wristband on museum visitors memory and emotions: a randomized controlled trial
No full textAdvances in technology have enabled museum curators to employ equipment that can measure visitors' physiological responses, offering a means to monitor these responses, while, at the same time, potentially engaging visitors. However, it is unclear whether these devices genuinely promote a positive experience or, conversely, are perceived as intrusive monitoring tools. Following traditional theories linking physiological responses, emotions and memory, we tested whether wearing a heart rate monitoring wristband during a temporary art exhibition could affect visitors' emotions and if emotional changes due to this manipulation could, in turn, affect the long-term memory of the artworks. Our findings show that using such a device heightened pleasant emotions experienced by visitors and improved their memory of the exhibit. These effects were still present even after six days from the visit. Moreover, we found that providing fake feedback concerning the emotions experienced in a specific room increased visitors' memory of artworks within that room. Our results are encouraging regarding the use of these technologies in museum exhibitions and bring evidence that they can enhance visitors' experiences, regardless of their accuracy
Common issues and human intervention in object detection from handcrafted features to deep learning: discussion
No full textWhile in traditional methods object detection is based on the handcrafted definition of relevant visual features and rules, in machine/deep learning methods this task is achieved by learning both features and rules from a training set. The traditional and machine/deep learning object detection workflows are often described as opposite because in the traditional framework, the visual features and rules to detect the object of interest are provided as input, while in the machine/deep learning-based framework they are automatically learned from the data depending on the task considered and constitute the final trained model. In this work, we analyze the object detection recipe, and we show that these two approaches actually present three common issues that require human supervision and ad hoc procedures to be addressed: the design of an object model suitable for the context, devices, and task at hand; the achievement of detection robustness against several factors like noise, image quality, changes in geometry, and light variations; and the definition of an appropriate matching function. We also briefly review some common metrics for evaluating object detection performance, proving that human intervention is crucial in this task as well. Our analysis aims at fostering a more aware use of the object detection approaches and stimulating new research for automating—where possible—the tasks that humans are still in charge of
From Edge to Cloud: Securing Distributed Containerized Applications
No full textThe adoption of containers in complex software systems is rapidly increasing, due to their flexibility that facilitates integration, scalability, and dynamic deployment. However, assessing the security of container-based applications remains challenging in distributed and heterogeneous environments: the scale and diversity of deployment scenarios call for sophisticated security evaluation and verification techniques. In this paper, we present Project SecCo (Securing Containers), whose aim is to develop an innovative framework for the systematic integration of security assessment services into the Continuous Integration and Continuous Delivery (CI/CD) DevOps pipeline. The framework orchestrates automatic services to prevent and reduce vulnerabilities in the design, implementation, and deployment phases, and to mitigate runtime attacks. This allows developers and IT operators to focus on integration and delivery, reducing security management tasks. Finally, the paper highlights the main research challenges for realizing this vision
LGAD technology for heavy ion detection and radiation damage diagnostics in diamond
No full textWe present the application of LGAD technology for time-of-flight measurements of heavy ions and for precise diagnostics of radiation damage in diamond sensors. The polycrystalline CVD (pcCVD) diamond sensor used in this work was irradiated with heavy ion beams at GSI, Darmstadt, Germany during several experimental campaigns and subsequently investigated at MedAustron, Wiener Neustadt, Austria. To mitigate radiation-induced performance degradation, we propose a dedicated amplification system originally developed for LGAD sensors, which significantly extends the operational lifetime of diamond detectors. For precise sensor diagnostics, we employed strip LGAD sensors — commonly used for minimum ionizing particle (MIP) detection — and demonstrated their excellent performance for heavy ion (He/C) detection, achieving timing resolutions below 40 ps
Dobbiamo credere nell'intelligenza artificiale? Un dialogo per comprendere il presente con Michela Milano e Paolo Traverso
No full textThis is a transcript of a round table discussion on the topic of trustworthy Artificial Intelligence (AI) organized and conducted by three FBK-ISR researchers involved in the project "Resilient Beliefs: Religion and Beyond." The experts consulted are Michela Milano and Paolo Traverso, two leading scholars in the field of AI. Starting from the notion of resilient belief, topics such as the difference between human and artificial intelligence, the danger of replacing people with machines even in jobs traditionally considered creative, the geopolitical dimension of the AI revolution, the legal regulation of AI hazards and the goal of consistent human oversight, how to create an epistemic and social environment conducive to the human-centric development of digital technologies, are discussed
Modelling practices, data provisioning, sharing and dissemination needs for pandemic decision-making: a European survey-based modellers’ perspective, 2020 to 2022
No full text: BACKGROUNDAdvanced outbreak analytics were instrumental in informing governmental decision-making during the COVID-19 pandemic. However, systematic evaluations of how modelling practices, data use and science-policy interactions evolved during this and previous emergencies remain scarce.AIMThis study assessed the evolution of modelling practices, data usage, gaps, and engagement between modellers and decision-makers to inform future global epidemic intelligence.METHODSWe conducted a two-stage semiquantitative survey among modellers in a large European epidemic intelligence consortium. Responses were analysed descriptively across early, mid- and late-pandemic phases. We used policy citations in Overton to assess policy impact.RESULTSOur sample included 66 modelling contributions from 11 institutions in four European countries. COVID-19 modelling initially prioritised understanding epidemic dynamics; evaluating non-pharmaceutical interventions and vaccination impacts later became equally important. Traditional surveillance data (e.g. case line lists) were widely available in near-real time. Conversely, real-time non-traditional data (notably social contact and behavioural surveys) and serological data were frequently reported as lacking. Gaps included poor stratification and incomplete geographical coverage. Frequent bidirectional engagement with decision-makers shaped modelling scope and recommendations. However, fewer than half of the studies shared open-access code.CONCLUSIONSWe highlight the evolving use and needs of modelling during public health crises. Persistent gaps in the availability of non-traditional data underscore the need to rethink sustainable data collection and sharing practices, including from for-profit providers. Future preparedness should focus on strengthening collaborative platforms, research consortia and modelling networks to foster data and code sharing and effective collaboration between academia, decision-makers and data providers
Single-photon counting pixel detector for soft X-rays
No full textSoft X-ray experiments at synchrotron light sources are essential for a wide range of research fields. However, commercially available detectors for this energy range often cannot deliver the necessary combination of quantum efficiency, signal-to-noise ratio, dynamic range, speed, and radiation hardness within a single system. While hybrid detectors have addressed these challenges effectively in the hard X-ray regime, specifically with single photon counting pixel detectors extensively used in high-performance synchrotron applications, similar solutions are desired for energies below 2 keV. In this work, we introduce a single photon counting hybrid pixel detector capable of detecting X-ray energies as low as 550 eV, utilizing the internal amplification of Low Gain Avalanche Diode (LGAD) sensors. This detector is thoroughly characterized in terms of Signal-to-Noise Ratio and Detective Quantum Efficiency. We demonstrate its capabilities through ptychographic imaging at MAX IV 4th-generation synchrotron light source at the Fe L3-edge (707 eV), showcasing the enhanced detection performance of the system. This development sets a benchmark for soft X-ray applications at synchrotrons, paving the way for significant advancements in imaging and analysis at lower photon energies
A Hardware-Friendly Data Compression Algorithm for SPAD-Based d-ToF Systems
No full textThis study introduces a data compression algorithm for Single Photon Avalanche Diode (SPAD)-based direct Time of Flight (dToF) systems, to reduce their memory requirements. The algorithm compresses most of the data at the pixel level, easing the readout process and providing higher frame rates. According to numerical simulations, with a maximum targeting range of 50 meters (m), the data compression algorithm can provide distance measurements up to 30 m with a relative error of 0.15 m under a challenging background illumination of 50 kilolux. The results show that the algorithm can compress 84.65 % of the data at the pixel level
A Digital Imager Architecture with On-Chip Compression for High-Speed Imaging Applications
No full textIn response to the demand for high-resolution, high dynamic range, and high frame rate image sensors with low power consumption, the early 2000s saw a surge in research on on-chip compression techniques. These aimed to reduce data transmission bandwidth with minimal image quality degradation, balancing compression performance with CMOS planar technology limitations. Now, thanks to 3D integration and recent advancements in SPAD arrays and digital pixel sensors, these techniques are becoming interesting again. This work presents a novel sensor architecture suitable for digital imagers featuring on-chip compression. The proposed solution employing an in-pixel shot noise-based differential modulation scheme (DPCM) combined with an intra-pixel suppression scheme to exploits both temporal and spatial redundancies. Preliminary evaluations of compression performance through numerical simulations using synthesized videos of real scenes have demonstrated the advantages of the proposed solution achieving simultaneous high-fidelity image reconstruction (PSNR ∼28 dB) and a high compression ratio at the same time (CR\~{}27). This reveals significant potential for further enhancements, paving the way for high speed and dynamic range imaging in various applications
A Practical Tutorial on Explainable AI Techniques
No full textThe past years have been characterized by an upsurge in opaque automatic decision support systems, such as Deep Neural Networks (DNNs). Although DNNs have great generalization and prediction abilities, it is difficult to obtain detailed explanations for their behavior. As opaque Machine Learning models are increasingly being employed to make important predictions in critical domains, there is a danger of creating and using decisions that are not justifiable or legitimate. Therefore, there is a general agreement on the importance of endowing DNNs with explainability. EXplainable Artificial Intelligence (XAI) techniques can serve to verify and certify model outputs and enhance them with desirable notions such as trustworthiness, accountability, transparency, and fairness. This guide is intended to be the go-to handbook for anyone with a computer science background aiming to obtain an intuitive insight from Machine Learning models accompanied by explanations out-of-the-box. The article aims to rectify the lack of a practical XAI guide by applying XAI techniques, in particular, day-to-day models, datasets and use-cases. In each chapter, the reader will find a description of the proposed method as well as one or several examples of use with Python notebooks. These can be easily modified to be applied to specific applications. We also explain what the prerequisites are for using each technique, what the user will learn about them, and which tasks they are aimed at