Archivio della ricerca - Fondazione Bruno Kessler
Not a member yet
21227 research outputs found
Sort by
Experimental verification of threshold quantum state tomography on a fully-reconfigurable photonic integrated circuit
No full textAbstract
Reconstructing the state of a quantum system represents a pivotal task for quantum information applications. The standard approach based on quantum state tomography requires a number of measurements that scales exponentially with the number of qubits. Other methods have been proposed and tested to reduce the number of measurements, or to focus on specific properties of the output state rather than on its complete reconstruction. Here, we show experimentally the application of an approach, called threshold quantum state tomography, in an advanced hybrid photonic platform with states up to
n
= 4 qubits. This method does not require prior knowledge and selects only the informative projectors starting from the measurement of the density matrix diagonal. We demonstrate its effectiveness by showing that a consistent reduction in the number of measurements is obtained for relevant states, with only very limited loss of information. These results open perspective for its application in larger systems
The IDRE Dataset in Practice: Training and Evaluation of Small-to-Medium-Sized LLMs for Empathetic Rephrasing
No full textBack-side-illuminated Silicon photomultipliers for improved radiation hardness
No full textSilicon Photomultipliers (SiPMs) are single-photon sensitive detectors that continue to attract increasing interest
in several industrial and scientific applications that require fast detection speed, high sensitivity, compactness,
insensitivity to magnetic fields and low bias voltages. In particular, the SiPMs are used in high-energy
physics (HEP) experiments, and for the readout of scintillators in gamma-ray detectors for space experiments.
In such applications they receive a significant dose of radiation (e.g. protons, electrons, neutrons, ⋯) which
degrades their performance.
During the last years, at FBK (Trento, Italy) we have been developing many different technologies for SiPMs
and SPADs, optimized for different applications. We also studied extensively the effect of ionizing energy
loss effects and non-ionizing energy loss effects (i.e. bulk displacement damage) on many different SPAD and
SiPM technologies, highlighting the most interesting effects on noise and detection efficiency.
Based on such results, we started specific technological improvements aimed to improve the radiation hardness
of novel SiPMs technologies. We are currently working on several directions. Among the most promising:
i) we are exploiting the reduction of the high-field active area, with a novel SiPM structure based on chargefocusing
mechanisms and back-side illumination, to mitigate the noise increment due to back damage, and
ii) we are working on active control and draining of radiation-induced charge in the dielectrics, to mitigate
the electric field modification effects of ionizing-energy loss. We performed TCAD simulations of the microcell
(i.e. SPAD) structure, and we estimated the noise generation (including field-enhancement effects), to
verify and quantify the beneficial effects of charge-focusing on the mitigation of the irradiation-induced bulkdamage
effects, showing a reduction of the primary dark count rate (also after irradiation) and a reduction of
the activation-energy lowering after irradiation
Leveraging Multi-agent Systems for Domain-Pertinence Query Classification in Informative Chatbots
No full textLarge Language Models generate answers to any questions provided by users. Even though this is a positive characteristic, when they are integrated into a real-world domain-specific solution, the generation of an answer to a question that is not related to the domain is a weakness. This work addresses the challenge of classifying users’ questions as in-topic or out-of-topic to limit the capabilities of a Large Language Model. We propose a multi-agent approach, called “Pool of Experts”, which leverages a structured hierarchy of specialized agents to synthesize expert contributions into a final decision. To evaluate the effectiveness of the proposed approach, we tested our methodology by integrating two description-based frameworks for agent profile creation: User Design Persona and Cognitive Load Theory. We compared our approach against traditional Transformer-based Natural Language Inference models as a baseline. Experimental results, observed in a real-world scenario concerning a question-answering system supporting pregnant women, demonstrate the superiority of the proposed methodology
Distributed Estimation of Sensor Statistics Using Wireless Networks of Single-Transistor Chaotic Oscillators
No full textEfficient aggregation of readings from numerous nodes is essential for large-scale applications of wireless sensor networks. This study provides the first systematic numerical investigation of a network paradigm in which sensing and distributed computation are embedded directly into the physical layer through the collective dynamics of coupled chaotic oscillators. A minimalist, single-transistor chaotic circuit characterized by rich nonlinear dynamics and high sensitivity to supply voltage variations is considered. This sensitivity enables each node to translate local physical signals from a transducer into perturbations of its oscillatory state. Networks of these oscillators arranged in irregular two-dimensional geometries are simulated with coupling implemented via near-field inductive links. The influence of coupling strength and node-to-node supply voltage variability on collective dynamics is analyzed. While increased coupling elevated the level of partial synchronization, resulting in sharper spectral signatures, variability had a blurring effect; this was accompanied by a gradual shift in spectral shape determined by the average voltage. The statistical properties of a distributed physical variable, namely its mean and variability, could be reliably recovered from these signals using listeners operating either in near-field or far-field antenna configurations, via a neural network-based approach. We further investigated frequency-division and time-division multiplexing techniques as scalable strategies for practically realizing long-range couplings. This study demonstrates the feasibility of embedding distributed sensing based on minimalist chaotic circuits
Innovating Space Operations with AI: The AISHGO Project
No full textArtificial Intelligence (AI) is increasingly permeating and improving various aspects of our daily lives; but can it also revolutionize how spacecraft are operated? This is the central question driving the project "AI for Automation of Satellite Health Monitoring and Ground Operations (AISHGO)." The project aims to integrate AI into mission control environments, such as the European Space Agency's European Space Operations Centre (ESOC) DLR's GSOC, to automate and optimize satellite health monitoring and ground operations.
AISHGO is structured around four pivotal use cases, each addressing distinct aspects of satellite operations. These include machine learning-based incident classification and root-cause analysis assistance, AI-based predictive
maintenance, intelligent telemetry data anomaly detection, and AI-based long-term satellite health monitoring. The research leverages both structured and unstructured data to develop data-driven AI solutions that surpass classical models in efficiency and effectiveness. By utilizing advanced AI techniques and models such as transformers, large language models (LLM), and long-short-term memory (LSTM) networks, the project demonstrates the capability of AI to predict system behaviors, flag potential issues, and enable operators to take preventative measures.
The solutions proposed have been validated through both simulations and real-world scenarios, confirming their practical value and highlighting the feasibility of adopting AI technologies in modern space operations
Novel class discovery meets foundation models for 3D semantic segmentation
No full textThe task of Novel Class Discovery (NCD) in semantic segmentation involves training a model to accurately segment unlabelled (novel) classes, using the supervision available from annotated (base) classes. The NCD task within the 3D point cloud domain is novel, and it is characterised by assumptions and challenges absent in its 2D counterpart. This paper advances the analysis of point cloud data in four directions. Firstly, it introduces the novel task of NCD for point cloud semantic segmentation. Secondly, it demonstrates that directly applying an existing NCD method for 2D image semantic segmentation to 3D data yields limited results. Thirdly, it presents a new NCD approach based on online clustering, uncertainty estimation, and semantic distillation. Lastly, it proposes a novel evaluation protocol to rigorously assess the performance of NCD in point cloud semantic segmentation. Through comprehensive evaluations on the SemanticKITTI, SemanticPOSS, and S3DIS datasets, our approach show superior performance compared to the considered baselines
Formal methods in industrial critical systems
No full textFormal methods are well-established and extensively used to ensure the correctness of core hardware and software components in safety-critical systems across industries such as railways, avionics, nuclear plants, and automotive. Their ability to provide mathematically rigorous guarantees makes them invaluable for verifying critical properties like safety, reliability, and security. However, the increasing complexity of modern systems requires enhanced support for applying these techniques effectively in industrial context. Addressing these challenges demands that formal methods evolve to become more scalable, interoperable with industrial development workflows, and better supported by automation and user-friendly tools for developers and engineers. These advancements are essential to enable the broader and more systematic adoption of formal methods in the engineering of complex, real-world systems. In this introduction to the special issue, we highlight several recent advances in the application of formal methods for specifying and verifying safety-critical systems in various industrial domains. These advances are showcased through four thoroughly revised and extended papers originally presented at the 28th International Conference on Formal Methods for Industrial Critical Systems (FMICS 2023)