Politecnio die Bari - Catalogo di prodotti della Ricerca
Not a member yet
36616 research outputs found
Sort by
Stealthy LLM-Driven Data Poisoning Attacks Against Embedding-Based Retrieval-Augmented Recommender Systems
No full textWe present a systematic study of provider-side data poisoning in retrieval-Augmented recommender systems (RAG-based). By modifying only a small fraction of tokens within item descriptions-for instance, adding emotional keywords or borrowing phrases from semantically related items-An attacker can significantly promote or demote targeted items. We formalize these attacks under token-edit and semantic-similarity constraints, and we examine their effectiveness in both promotion (long-Tail items) and demotion (short-head items) scenarios. Our experiments on MovieLens, using two large language model (LLM) retrieval modules, show that even subtle attacks shift final rankings and item exposures while eluding naive detection. The results underscore the vulnerability of RAG-based pipelines to small-scale metadata rewrites, and emphasize the need for robust textual consistency checks and provenance tracking to thwart stealthy provider-side poisoning
Machine Learning for Early Prediction of Cognitive Decline in Alzheimer's Disease
No full textEarly identification of cognitive decline is a crucial challenge in Alzheimer’s research, with significant implications for therapeutic intervention and disease
management. Currently, available treatments merely decelerate the progression of the disease, without achieving a complete halt, consequently researchers are concentrating their efforts on a key aspect that concerns the early prediction and prevention of Alzheimer’s disease (AD) in order to delay the onset and progression. However, initial clinical manifestations are not always decisive, and diagnosis often occurs at an advanced stage of cognitive impairment. Using data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) cohort, this study proposes an interpretable machine learning (ML) model for predicting mild cognitive impairment (MCI) from cognitively normal (CN) patients based on multimodal baseline biomarkers. After a thorough preprocessing phase and a features selection step, an optimized Random Forest (RF) classifier was implemented. The results obtained show an overall accuracy of the model of 76%, with a sensitivity of 64% and specificity of 84%, confirming the key role of cognitive, neurostructural and metabolic biomarkers in predicting the risk of progression to MCI. This study demonstrates the potential of ML in early prediction of cognitive decline, laying the foundation for more effective and personalized diagnostic tools in the context of neurodegenerative diseases
Skyrmions in Nanotechnology
No full textSkyrmions, topologically protected textures, have been observed in different fields of nanotechnology and have emerged as promising candidates for different applications due to their topological stability, low-power operation, and dynamic response to external stimuli. First introduced in particle physics, skyrmions have since been observed in different condensed matter fields, including magnetism, ferroelectricity, photonics, and acoustics. Their unique topological properties enable robust manipulation and detection, paving the way for innovative applications in room temperature sensing, storage, and computing. Recent advances in materials engineering and device integration have demonstrated several strategies for an efficient manipulation of skyrmions, addressing key challenges in their practical implementation. In this review, we summarize the state-of-the-art research on skyrmions across different platforms, highlighting their fundamental properties and characteristics, recent experimental breakthroughs, and technological potential. We present future perspectives and remaining challenges, emphasizing the interdisciplinary impact of skyrmions on nanotechnology
Interaction Between Buildings and UAM Infrastructures: Simulation of Wind Paths in the Urban Environment
No full textThe development of Urban Air Mobility as an emerging air transportation service requires a proper integration between air and ground infrastructures at the urban scale, whose investigation is still little explored in the literature. For instance, vertiports represent a crucial link between air and ground, requiring careful considerations for their placement. Among the factors most affecting the vertiports location, the wind paths in urban environments should be considered, as they could compromise both safety and security of aircrafts. However, the wind effect at the urban scale is challenging to predict, as it is closely related to the urban morphology, which can also generate turbulence flows. Among the existing methodologies to study urban wind, computational fluid dynamics simulations could be a good approach. The present study aims to explore the interaction between buildings and UAM infrastructure by exploring wind paths in a suburban area of the city of Bari (Southern Italy), where some unoccupied areas are potentially suitable for the location of vertiports. The CFD ENVI-met software has been adopted to perform simulations, showing the strong influence of urban morphology on wind flows, thus requiring accurate wind analyses to support the vertiport placement planning in the urban environment
WeBIUM 2025: 2nd Workshop on Wearable Devices and Brain-Computer Interfaces for User Modelling
No full textWearable Devices (WDs), such as smartwatches and fitness trackers, continuously produce extensive data streams that reveal valuable information about physiological states, activity patterns, and user interactions. These devices enable the construction of advanced user models, offering dynamic insights into personal routines, health trends, and behavioural tendencies. Meanwhile, Brain-Computer Interfaces (BCIs) emerge as a transformative technology by capturing neural activity to provide unprecedented access to cognitive and emotional states. However, BCIs are not conventionally classified as wearables; the latest technological advancements have reduced their size to resemble everyday accessories like earphones, suggesting their potential integration into wearable formats in the near future.Despite the promise of these technologies, the full exploitation of their data for user modelling and personalization-such as optimizing activities like media consumption or interaction design-remains underexplored. The convergence of WDs and BCIs opens up new avenues for understanding the complexity of human behaviour and preferences, and this potential is amplified by the integration of Large Language Models (LLMs). By synthesizing and interpreting multimodal datasets, LLMs can better understand the intricate interplay between physiological, cognitive, and behavioural signals, ultimately enriching user modelling processes.Following the success of the first edition, this workshop seeks to delve into the deep impact of combining data from wearable devices, neural interfaces, and advanced machine learning models. Participants will explore the opportunities and challenges that arise in this innovative context, examining how these technologies can be harnessed to enhance the granularity and accuracy of user models. The discussions will also address practical implications, such as ethical considerations and the necessity of privacy-Aware approaches when dealing with highly sensitive physiological and neural data.Through collaborative exchanges, the initiative aspires to chart new directions in the field, fostering novel research trajectories and interdisciplinary partnerships. The interplay of WDs, BCIs, and LLMs can redefine user modelling by creating systems that dynamically adapt to individual needs and behaviours, paving the way for transformative advancements in personalized experiences. By drawing on cutting-edge research and practical expertise, the workshop aims to inspire innovative solutions that capitalize on these emerging synergies, advancing the boundaries of what is possible in user modelling and adaptive systems. © 2025 Copyright held by the owner/author(s)
Entropic forces in rotaxane-based daisy chains: Toward tunable nanomechanical systems
No full textMechanically interlocked polymers and molecules exhibit unique topological, physical, and chemical properties, making them highly promising for applications in molecular machines, molecular switches, artificial muscles, nano-actuators, nano-sensors, and biomedical technologies. While significant progress has been made in their synthesis and practical implementation, theoretical studies remain underexplored. In this work, we examine the role of entropic forces in daisy chain structures incorporating rotaxanes, with the ultimate goal of characterizing entropic nano-springs for use in nanomechanics and nanotechnology. Potential applications include artificial cytoskeletons, synthetic cells, and nano-mechanical logic gates
Building with earth: between constructive memory and design Innovation / Costruire con la terra: tra memoria costruttiva e innovazione progettuale
No full textThe workshop “Cultura Material no Alentejo” highlighted how rammed earth, beyond its role as building material, can act as a critical device for rethinking architectural design. Its reversibility, local availability, and embedded memory position it as a tool for contextual, ethical, and transformative practices. Design, understood as an act of care and relational engagement, transcends form to become a cultural and environmental process—one capable of generating alternative imaginaries, inhabiting time, and re-signifying built heritage
Decentralized Control of Crop Growth Conditions in Vertical Farms under Dynamic Energy Markets
No full textThe growing global population and the increasing scarcity of arable land highlight the urgent need for reliable and efficient food production systems. With their controlled environments, vertical farms (VFs) offer a promising solution for sustainable food security. Nevertheless, their high energy demands call for innovative approaches to optimize energy consumption while maintaining optimal growing conditions. This paper introduces a novel control-oriented model for VFs, capturing the interactions between crop growth conditions and energy consumption. To address the high energy demand of VFs, the model is integrated into a dynamic energy market characterized by time-varying energy prices and a demand response scheme, which includes a discrete reward to encourage flexible energy consumption. Then, centralized and decentralized receding horizon control approaches are proposed to minimize the energy cost of the VF while ensuring optimal crop growth. Experimental evaluations on real systems of varying scales demonstrate the effectiveness of the proposed approaches in reducing costs and ensuring sustainable agricultural practices