Archivio della ricerca - Fondazione Bruno Kessler
Not a member yet
21227 research outputs found
Sort by
RUSH: A Novel Fully AI-driven Framework for Seamless Integration of Observations and Global AI Forecasts in Short-term Weather Prediction
No full textLarge Language Models are Zero-Shot Next Location Predictors
No full textPredicting the locations an individual will visit in the future is crucial for solving many societal issues like disease diffusion and pollution reduction. However, next-location predictors often require a significant amount of individual-level information that may be scarce or unavailable (e.g., in cold-start scenarios). Large Language Models (LLMs) have demonstrated strong generalization and reasoning capabilities while being rich in geographical knowledge, suggesting that they can operate as zero-shot next-location predictors. In our study, we evaluate over 15 LLMs on three real-world mobility datasets and find that they achieve accuracies up to 36.2%, representing a relative improvement of almost 640% compared to traditional models designed for human mobility. We further assess data contamination risks and explore the potential for using LLMs as text-based explainers for next-location predictions. Our results indicate that, irrespective of model size, LLMs can both predict and justify their decisions effectively
3D-Bioprinted Light-Sensitive Cell Scaffold Based on Alginate-Conjugated Polymer Nanoparticles for Biophotonics Applications
No full textThe field of biotechnology has rapidly grown in recent years leading to unprecedent achievements in regenerative medicine and tissue engineering. Among latest high-tech available technologies, 3D bioprinting can be surely considered the most promising to develop complex structures mimicking organs and tissues, as well as functional 3D cell scaffolds. If supplemented with organic functional materials, engineered cell scaffolds can be used as bioelectronic interfaces and biomedical sensors. In this work, a novel 3D-bioprinted cell scaffold enhanced with light-responsive organic semiconducting polymer nanoparticles (100-nm hydrodynamic diameter, absorbance spectra peak at λ = 496 nm, and emission spectra peak at λ = 645 nm) is presented. The light-sensitive cell scaffold offers excellent biocompatibility and support of cell growth. The 3D-bioprinted biocompatible light-sensitive cell scaffolds can be used for light control and modulation of cellular activities in a 3D and real-mimetic tissue/organ-like environment, paving the way to new applications in neural engineering and regenerative medicine
Enhanced compressive threshold quantum state tomography for qudit systems
No full textWe propose an efficient quantum state tomography method inspired by compressed sensing and threshold quantum state tomography that can drastically reduce the number of measurement settings to reconstruct the density matrix of an N-qudit system. We validate our algorithm with demonstrations on IBMQ and show the efficient and accurate reconstruction of N≤7 qubit systems, reproducing GHZ, W, and random states with O(1), O(N2), and O(N) settings
Non-uniformly lighted image enhancement exploiting the Atangana–Baleanu fractional integral and the Sobel filter
No full textImproving the quality of non-uniformly lighted images is particularly hard because these images contain regions with different brightness, thus needing a different level of enhancement. Recently, this problem has been addressed by combining the input image with a new version where the input brightness is linearly up-scaled by a parameter, which is estimated without supervision by segmenting the image into dark and bright regions and comparing their brightness. This estimate represents a pitfall because it makes the algorithm performance dependent on the segmentation accuracy and cannot be applied to images whose bright regions are almost white. To overcome these issues, the present work introduces a new image-aware estimate of the brightness up-scaling parameter, which exploits edge information extracted by the Sobel filter and by the first derivative of the Atangana–Baleanu fractional integral. The joint use of integer- and fractional-order calculus proposed here represents the main contribution of this work and, as proved by the experiments, enables us to reach a good level of enhancement surpassing other cutting-edge techniques, particularly in terms of reduced artifact production
Integrating Planning and Learning for Agents Acting in Unknown Environments
No full textAn Artificial Intelligence (AI) agent can perceive an environment through sensors and act in the environment through actuators. When performing tasks in a known environment, an agent knows what actions it can execute and how they affect the environment state, but when the environment is unknown, the agent needs to learn how the environment works to make good decisions for accomplishing tasks. In a real-world situation, an agent may have only low-level perceptions of the environment rather than the high-level representations required to make decisions by means of symbolic planning.
This book, Integrating Planning and Learning for Agents Acting in Unknown Environments proposes an architecture that integrates learning, planning, and acting. The author, Leonardo Lamanna, won the 2023 Marco Cadoli award, an annual award from the Italian Association for Artificial Intelligence (AIxIA) for the best doctoral thesis in the field of artificial intelligence, for this work. The approach combines data-driven learning methods for building an environment model with symbolic planning techniques for reasoning on the learned model, focusing on learning the model, either from continuous or symbolic observations. The problem of online learning the mapping between continuous perceptions and symbolic states is tackled, and symbolic planning techniques are exploited to enable an agent to autonomously gather relevant information online, which is required by learning methods to overcome some of the simplifying assumptions of symbolic planning. The effectiveness of the approach in simulated complex environments is shown experimentally and the applicability of the approach in real environments is demonstrated by conducting experiments on a real robot.
Outperforming state-of-the-art methods, the approach described in this book will be of interest to all those working in the field of AI and autonomous agents
Functional tests of the detector assembly demonstration model of the eXTP wide field monitor: system description and results
No full textThe Wide Field Monitor is one of the 4 state-of-the-art instruments onboard the enhanced X-ray Timing and Polarimetry mission, which will search for and observe neutron stars, magnetars and black holes to study matter under extreme conditions of density, gravity and magnetism. The wide-field cameras of the Wide Field Monitor are based on large-area silicon drift detectors. Such position-sensitive spectroscopic detectors, together with their front-end electronics, must pass a series of functional tests before proceeding with the mass production of the detector assemblies. This paper describes the test system developed to carry out such measurements, as well as reporting the results of the functional tests together with a preliminary characterisation of the performance of the detector assemblies
Miscellaneous applications of deep learning based multi-sensor Earth observation
No full textMulti-sensor Earth observation offers significant potential across several fields by combining data from diverse sources, providing an understanding of the Earth's dynamics. From environmental monitoring to disaster management, agriculture to urban planning, it enables precise analysis and monitoring, tracking changes in land use, vegetation health, and natural resources with unprecedented accuracy. Advances in satellite and aerial imaging technology have expanded Earth observation's applications, enhanced further by deep learning techniques. In urban settings, multi-sensor data coupled with deep learning algorithms excel in characterizing man-made surfaces and classifying urban areas. It is also useful in mining sector, especially for mineral exploration. At the same time, in marine applications, it helps in ship detection and oil spill monitoring. Environmental endeavors like wetland monitoring and tree species classification benefit from multi-sensor observation and deep learning. Forest fire monitoring, essential for ecosystem preservation, also benefits from multi-sensor Earth observation. This chapter explores the applications of deep learning-based multi-sensor Earth observation across various application domains. The chapter demonstrates a few case studies as well
