Politecnio die Bari - Catalogo di prodotti della Ricerca
Not a member yet
36616 research outputs found
Sort by
Do LLMs Memorize Recommendation Datasets? A Preliminary Study on MovieLens-1M
No full textLarge Language Models (LLMs) have become increasingly central to recommendation scenarios due to their remarkable natural language understanding and generation capabilities. Although significant research has explored the use of LLMs for various recommendation tasks, little effort has been dedicated to verifying whether they have memorized public recommendation dataset as part of their training data. This is undesirable because memorization reduces the generalizability of research findings, as benchmarking on memorized datasets does not guarantee generalization to unseen datasets. Furthermore, memorization can amplify biases, for example, some popular items may be recommended more frequently than others
Micro-tapered Long Period Grating for Mid-Infrared Wavelengths
No full textTraditionally, long period gratings (LPGs) are fabricated by exposing the optical fiber core to ultraviolet (UV) lasers to induce periodic refractive index changes [1]. Alternative techniques that bypass UV laser exposure, useful in case of non-photosensitive materials, include the use of femtosecond lasers, CO2 lasers, electric arc discharges, and resistive filament heating under tension [2,3]. The capability to fabricate intricate structures with high precision is crucial for advancing the practical applications of fluoride glass, particularly in diverse mid-infrared technologies, including sensing, filtering, and laser systems [4,5]
Computer vision-based seismic assessment of RC simply supported bridges characterized by corroded circular piers
No full textThis study proposes a framework for the rapid assessment of seismic fragility and risk of reinforced concrete (RC) circular bridge piers affected by corrosion. The methodology integrates a novel computer vision (CV) algorithm to enhance visual inspections for corrosion level identification, combined with a probabilistic approach to seismic fragility analysis. The aim of the methodology is to quantify the impact of corrosion-induced deterioration on structural performance, expressed as an increment in terms of seismic risk. The first part of the framework consists of defining a custom convolutional neural network able to automatically predict the corrosion severity class starting from a metric-photographic survey. The proposed network incorporates attention mechanisms and color space transformations to ensure robust performance under varying image conditions. The output is used within a probabilistic-based structural modelling and analysis framework, which allows to derive seismic performance of the considered bridge pier typology. On the modelling side, a specific fiber-based approach was employed, in order to account for non-uniform cross-sectional corrosion and current deterioration condition. The results are returned in terms of seismic fragility and risk metrics for quantifying the reduction of seismic performance with respect to the initial conditions. The framework was tested on a real-life case-study exhibiting non-uniform cross-sectional base corrosion, and subsequently, additional scenarios considering full-section base corrosion at varying severity levels were investigated. The outcomes of this study demonstrate the potentialities of artificial intelligence in improving the current practices in the field of seismic assessment of aging RC infrastructures
An Indoor Experimental Testbed for 5G-Based UAV Control and Communication
No full textThe integration of Unmanned Aerial Vehicles (UAVs) into next-generation
mobile networks is widely recognized as a key enabler of disruptive
applications, where aerial platforms may function either as network nodes
or as advanced network users supporting a variety of services.
Unfortunately, experimental testbeds in which UAVs perform tasks while
communicating with ground infrastructure over Fifth-Generation (5G)
networks remain scarce, primarily due to the challenges posed by legal
restrictions on Beyond Line-of-Sight (BLoS) and autonomous operations.
Motivated by this need, this work presents the design, implementation, and
evaluation of a novel indoor experimental testbed for assessing the
performance of UAV-based systems operating over 5G networks.
The testbed features an autonomously controlled quadcopter equipped with a
5G modem, connected to a private 5G network implemented using
Software-Defined Radio (SDR) technology and the OpenAirInterface (OAI)
framework.
To ensure a controlled environment, a motion capture system is used to
provide absolute indoor positioning data, emulating Global Navigation
Satellite System (GNSS) coordinates without relying on external satellites.
A preliminary experimental campaign is conducted to evaluate the proposed
system in terms of 5G network performance, radio link characteristics, and
UAV platform energy consumption
Highly reliable personalized noninvasive hemoglobin estimation by using Vision Transformers and dual fine-tuning
No full textArtificial intelligence is revolutionizing health care, particularly in precision medicine and noninvasive diagnostics. Anemia, which is a widespread condition that affects billions of people worldwide, compromises oxygen transport due to low hemoglobin levels, which leads to severe complications if left undetected. Early and frequent monitoring is essential, yet traditional blood tests can be invasive, costly, and impractical for continuous assessment. This study presents the first patient-specific system for noninvasive hemoglobin estimation from palpebral conjunctiva images. Unlike previous approaches, our model integrates the vision transformer (ViT) architecture with dual fine-tuning, which enables personalized adaptation to each patient's unique physiological characteristics. The dataset consists of conjunctival images captured over multiple days from the same patients, which allows for an individualized calibration process that enhances predictive accuracy. Our model achieved an R2 of 0.94, an accuracy of 98%, and a mean absolute error (MAE) of 0.25 g/dL, thus demonstrating a performance comparable to that of laboratory tests. Additionally, the model's 100% sensitivity ensured that all anemic cases were detected, thereby minimizing the risk of false-negatives. By providing highly precise, rapid, and accessible anemia screening, this approach has the potential to redefine long-term hematological monitoring, thereby reducing reliance on frequent blood tests and improving clinical decision-making in resource-limited settings
Assessment of the energy consumption of the glass production process by compressed air flow measurements
No full textIn this work the authors propose a methodology to manage the operation of a compressed air plant to guarantee its best energy efficiency, so controlling the energy consumption of the industry where such a plant is present. In particular, the aim is to set a strategy for controlling the optimal operation of the compressors in the plant, in the perspective of Industry 4.0, to ensure the minimization of energy consumption found by their use and their automated management. The methodology bases its principle on the digitization of the operational maps of the machines, able to provide in real-time their operating conditions found by the measurements of the characteristic parameters, and thus intervene, through feedback chains, on control systems to restore optimal conditions, when possible, or intervene with routine maintenance to prevent irreversible damage. It is intended to test the procedure on a compressed air plant of a glass industry located in southern Italy
Slot Waveguide Ring Resonators for Label-Free Detection of SARS-CoV-2 and RSV in the Near-Infrared Region
No full textOptimized Blood Pressure Prediction With a Hybrid CNN-SVR Model Using Multiwavelength PPG
No full textNoninvasive vital sign monitoring, especially blood pressure (BP), is crucial for evaluating overall health and identifying early indicators of medical issues. Photoplethysmography (PPG) is a noninvasive technique increasingly adopted in vital sign monitoring, as it allows for the continuous and precise measurement of cardiac-related signals. This technology is sensitive to fluctuations in blood volume within blood vessels, enabling real-time tracking of each heartbeat and identification of any irregularities. This study investigates the effectiveness of machine learning (ML) models for predicting arterial pressure based solely on PPG signals, utilizing a novel dataset and multiwavelength sensor technology that captures data from four distinct wavelengths: infrared (IR), red, green, and blue. The participant pool consisted of 88 individuals, with each measurement lasting 30 s at a sampling rate of 100 Hz. After a dedicated phase of signal preprocessing, three methodologies were assessed: a multilayer perceptron (MLP) utilizing 84 features per subject, a dimensionality reduction strategy through principal component analysis (PCA), and a convolutional neural network (CNN) architecture. The CNN approach showcased impressive performance metrics, with results aligning with the standards set by the Association for the Advancement of Medical Instrumentation (AAMI) and the British Hypertension Society (BHS). Additionally, to further enhance accuracy, a support vector regression (SVR) correction algorithm was applied, as a postprocessing phase, addressing discrepancies in measurements related to age, thereby improving the reliability of the arterial pressure predictions and reaching errors of 1.36 ± 2.98 and 1.52 ± 2.67 mmHg for systolic BP (SBP) and diastolic BP (DBP), respectively. Compared with recent cuffless BP estimators that rely on single-wavelength PPG or pulse transit time (PTT) features and typically report mean absolute error (MAE) values of 3-5 mmHg, the proposed multiwavelength hybrid CNN-SVR shrinks the error by more than 60% (MAE =1.84/1.98 mmHg for SBP/DBP), while still meeting AAMI and BHS grade A/B requirements. This establishes, to the best of authors' knowledge, the first sub-2 mmHg, AAMI-compliant result obtained with a commodity optical front end and no calibration