Politecnio die Bari - Catalogo di prodotti della Ricerca
Not a member yet
36616 research outputs found
Sort by
Gyroscope Real-Time Denoising by an Adaptive Threshold Wavelet Algorithm: Achieving over 12 dB SNR Improvement
No full textGyroscopes play a pivotal role in applications ranging from navigation and robotics to aerospace and consumer electronics, where denoising is often critical to improve overall system performance. Traditional Kalman-based filters are often regarded as the gold standard for inertial sensor denoising, yet they require assumptions on the system's dynamics that may not always hold, particularly in the presence of abrupt or unpredictable maneuvers. Several alternative approaches avoid such assumptions, but typically exhibit inferior performance compared to Kalman filters (KFs). Here we report on a novel wavelet-based denoising algorithm that operates in real time without relying on prior knowledge of the sensor's dynamic conditions. Our technique adaptively calibrates the threshold by modeling noise with a generalized Gaussian distribution (GGD) and adjusts it according to the ongoing signal variance. This strategy offers two core advantages: it preserves relevant signal discontinuities and handles broad noise distributions effectively, including non-Gaussian noise. We validate the algorithm on two distinct gyroscope platforms: a state-of-the-art fiber optic gyroscope, characterized by low noise and non-Gaussian behavior, and a commercial MEMS gyroscope with primarily Gaussian noise. Standard test signals - such as blocks, step, heavisine, and Doppler - reveal that our approach surpasses the KF by up to 1 dB and outperforms alternative wavelet-based techniques by at least 4 dB in signal-to-noise ratio (SNR) enhancement. Furthermore, the algorithm exhibits minimal overshoot at signal discontinuities, ensuring accurate angular rate reconstruction. These results establish our method as a high-performance and robust solution for gyroscope denoising especially in high-end inertial sensing. The algorithm operates without any prior knowledge of the host platform's motion model; it relies only on weak, sensor-level statistical assumptions that are satisfied by practically all gyroscopes
Multi-Agentic Recommender Systems: Foundations, Design Patterns, and E-Commerce Applications — An Industrial Tutorial
No full textThe goal of this tutorial is to provide our perspective on the most recent advances in LLM-powered agents for recommender systems. Building on our extensive experience deploying agentic tools in large-scale environments, this tutorial hopes to deepen the understanding of participants with diverse backgrounds on the alphabets that underpin multi-agentic frameworks. Organized by the founders of leading agentic tools, the tutorial will highlight how these frameworks are being applied to create next-generation recommender systems in diverse applications. The examples include context-aware recommendation, dynamic multi-step orchestration, and personalized recommendation systems. To provide a solid foundation, we begin with a brief background on the evolution of recommender systems and how recent breakthroughs in large language models (LLMs) have shifted the paradigm toward more interactive, adaptive, and autonomous systems. The hands-on session will allow participants to directly engage with state-of-the-art techniques, bridging the gap between theoretical concepts and practical implementations
Inertial Sensors Miniaturization through Silicon Photonics: Emerging Trends and Future Prospects
No full textSilicon photonics has proven transformative in data links and telecommunications; however, its potential impact on inertial sensing is still in an early stage of research and development. Recent progress in photonic accelerometers and gyroscopes signals a growing maturity that may enable a new generation of miniaturized inertial measurement units (mini-IMUs). This paper reviews current silicon photonic inertial sensor technology and discusses how key advancements are bringing navigation-grade and tactical-grade mini-IMUs within reach. A particular focus is placed on chip-scale interferometric optical gyroscopes and a newly developed stochastic model critical for predicting their performance. This model is designed to accelerate the transition from laboratory prototypes to practical, deployable devices. Future perspectives are also examined, emphasizing pivotal challenges and emerging strategies that may further drive the miniaturization and performance of silicon photonic inertial sensors in a variety of high-impact applications
Plasmonic metasensors comprising gold nanodisks and prisms on SiO2 substrate in the near infrared regime
No full textThis study presents a theoretical comparative analysis of localized surface plasmon resonance (LSPR) sensors based on gold nanodisks and triangular prisms patterned on a silicon dioxide (SiO2) substrate, with the aim of advancing biosensing technologies operating in the near-infrared regime. While conventional surface plasmon resonance (SPR) techniques offer high sensitivity, they often suffer from complex fabrication processes and reduced performance in detecting large biomolecules. In this work, we investigate how geometric parameters - such as height, nanodisk diameter/triangle side, and periodicity - influence the optical response and sensitivity of the two nanostructured configurations. Simulation results reveal that prisms outperform nanodisks in terms of bulk refractive index sensitivity, reaching values up to 572 nm/RIU compared to approximately 289 nm/RIU for nanodisks. This enhanced performance is attributed to the sharper edges and corners of the prism geometry, which promote stronger electric field confinement and the formation of localized electromagnetic hot spots. Furthermore, the use of gold ensures chemical stability and biocompatibility in both cases, making the designs suitable for real-world biosensing applications. Overall, this work not only emphasizes the potential of triangular prism-based metasurfaces in simplifying sensor architectures and reducing costs but also provides a solid theoretical basis for future experimental validation and integration into practical biosensing platforms
Are the Hidden States Hiding Something? Testing the Limits of Factuality-Encoding Capabilities in LLMs
No full textUnderstanding Metal–Organic Framework Densification: Solvent Effects and the Growth of Colloidal Primary Nanoparticles in Monolithic ZIF‐8
No full textTo commercialize metal–organic frameworks (MOFs), it is vital they are made easier to handle. There have been many attempts to synthesize them as pellets, tablets, or granules, though they come with innate drawbacks. Only recently have these been overcome, through the advent of self-shaping densified or monolithic MOFs (monoMOFs), which require minimal post-synthetic modification and avoid poor structural integrity, intractability, and pore collapse or blockage. ZIF-8 (zeolitic imidazolate framework-8) has emerged as a prototypical monoMOF in pure and in situ doped forms. Now its formation in solvent mixtures is studied to better understand the early stages of monolith formation and improve the scope of monoliths for hosting solvent-sensitive guests. Solvent-, temperature- and coagulant-dependent control over reaction kinetics induces variations in morphology that are explained by relating the nucleation and growth rates of primary nanocrystallites to the stability of colloidal dispersions during reaction. This yields mesoporous monoZIF-8 with mean pore size 16 nm, SBET >1400 m2g−1, bulk density 0.76 g cm−3, and resistance to permanent deformation exceeding previous reports. While the study highlights the powerful manipulation of monoMOF characteristics, a new understanding of the growth and stability of primary nanocrystallites has consequences for colloid synthesis generally
Modulating ultralong room-temperature phosphorescence through mechanical confinement of tailored polymer/MOF hybrid interfaces
No full textAchieving ultra-long room temperature phosphorescence (RTP) remains a significant challenge due to the inherent trade-off between excited-state lifetime and photoluminescence quantum yield (QY). Herein, we report the synthesis of polymer/metal-organic framework (MOF) hybrids via a bottom-up approach, enabling the formation of direct covalent integration between polymer matrices and MOF structures. By systematically optimizing the incorporation of long-chain alkyl amines during the synthesis process, the resultant hybrids demonstrate green RTP performance, achieving a lifetime of 359 milliseconds (ms) at room temperature (592 ms at 77 K), and a phosphorescence QY of ∼28%, whereas no detectable phosphorescence is observed in the parent MOFs and polymer components. Notably, nanoindentation-based mechanical analysis reveals, for the first time, a clear relationship between increased matrix rigidity and enhanced RTP performance. Additionally, a detailed investigation highlights the pivotal roles of extensive intramolecular hydrogen bonding and covalent linking between the polymer and extended frameworks in stabilizing triplet states, enabling efficient RTP. The hybrid materials also demonstrate good processability, allowing the creation of flexible RTP-emissive fibers and films (lifetimes of 408 ms at RT and 613 ms at 77 K) that leverage their inherent flexibility. Furthermore, these hybrids exhibit good selectivity in detecting water over other alcohols, underscoring their potential for smart sensor applications
Explainable Machine Learning for Environment-Aware Channel State Prediction in UAV-Based 6G Networks
No full textGraph neural networks in the nephropathological diagnosis of antibody-mediated rejection
No full textAntibody-mediated rejection (AMR) is a leading cause of kidney transplant failure, requiring accurate histopathological assessment for diagnosis. This study evaluates graph-based deep learning models for AMR classification using periodic acid–Schiff (PAS)-stained whole slide images (WSIs), with the aim of improving diagnostic accuracy and reproducibility. A multi-institutional dataset of 1193 WSIs from 348 patients was used, where glomeruli, arteries, and cortical tubulointerstitial regions were segmented via deep learning and represented as nodes in graph-structured data. Feature extraction was performed using both supervised and self-supervised methods, and classification was conducted with four graph neural network (GNN) architectures: Graph-Transformer, and the novel SimpleGCN, DenseGCN and SimpleGAT. Patch-wise convolutional and transformer-based classifiers served as baselines. All models were evaluated at both the WSI and biopsy levels using stratified five-fold cross-validation. GNN-based models consistently outperformed patch-wise baselines, with the best glomeruli-only GNN achieving a 5.34 % improvement in WSI-level accuracy (71.00 %) over the strongest baseline. Incorporating additional compartments (arteries and cortex) further improved accuracy to 86.97 % at the WSI level and 89.53 % at the biopsy level, with statistically significant gains confirming the additive value of multi-compartment modeling. Performance varied across feature extractors and graph configurations, underscoring the complexity of optimizing computational pipelines for AMR diagnosis. Overall, graph-based modeling substantially enhances AMR diagnostic performance over conventional approaches, enabling scalable, low-cost and reproducible workflows with minimal expert input. These findings demonstrate the potential of GNNs to support nephropathologists in delivering more consistent and reliable diagnoses, with future work needed to refine feature representations and integrate multimodal data for broader clinical utility