检索结果-内蒙古大学图书馆

Unveiling Group-Specific Distributed Concept Drift: A Fairness Imperative in Federated Learning

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Salazar, Teresa Gama, João Araujo, Helder Abreu, Pedro Henriques The Centre for Informatics and Systems Department of Informatics Engineering University of Coimbra Coimbra3030-790 Portugal INESC TEC The Faculty of Economy University of Porto Porto4200-465 Portugal The Institute of Systems and Robotics Department of Electrical and Computer Engineering University of Coimbra Coimbra3004-531 Portugal

In the evolving field of machine learning, ensuring fairness has become a critical concern, prompting the development of algorithms designed to mitigate bias in decision-making processes. However, achieving fairness in the presence of group-specific concept drift remains an unexplored frontier, and our research represents pioneering efforts in this regard. Group-specific concept drift refers to situations where one group experiences concept drift over time while another does not, leading to a decrease in fairness even if accuracy remains fairly stable. Within the framework of Federated Learning, where clients collaboratively train models, its distributed nature further amplifies these challenges since each client can experience group-specific concept drift independently while still sharing the same underlying concept, creating a complex and dynamic environment for maintaining fairness. The most significant contribution of our research is the formalization and introduction of the problem of group-specific concept drift and its distributed counterpart, shedding light on its critical importance in the realm of fairness. Additionally, leveraging insights from prior research, we adapt an existing distributed concept drift adaptation algorithm to tackle group-specific distributed concept drift which uses a multi-model approach, a local group-specific drift detection mechanism, and continuous clustering of models over time. The findings from our experiments highlight the importance of addressing group-specific concept drift and its distributed counterpart to advance fairness in machine *** Codes 68T01 © 2024, CC BY.

关键词： Machine learning

HO-FMN: Hyperparameter Optimization for Fast Minimum-Norm Attacks

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Mura, Raffaele Floris, Giuseppe Scionis, Luca Piras, Giorgio Pintor, Maura Demontis, Ambra Giacinto, Giorgio Biggio, Battista Roli, Fabio University of Cagliari Dept. of Electrical and Electronic Engineering Cagliari Italy Sapienza University of Rome Department of Computer Engineering Rome Italy University of Genoa Department of Computer Science Bioengineering Robotics and Systems Engineering Genoa Italy

Gradient-based attacks are a primary tool to evaluate robustness of machine-learning models. However, many attacks tend to provide overly-optimistic evaluations as they use fixed loss functions, optimizers, step-size schedulers, and default hyperparameters. In this work, we tackle these limitations by proposing a parametric variation of the well-known fast minimum-norm attack algorithm, whose loss, optimizer, step-size scheduler, and hyperparameters can be dynamically adjusted. We re-evaluate 12 robust models, showing that our attack finds smaller adversarial perturbations without requiring any additional tuning. This also enables reporting adversarial robustness as a function of the perturbation budget, providing a more complete evaluation than that offered by fixed-budget attacks, while remaining efficient. © 2024, CC BY.

关键词： Machine learning

A Wide Database for a Multicenter Study on Pneumocystis jirovecii Pneumonia in Intensive Care Units 32

学校读者我要写书评

暂无评论

A Wide Database for a Multicenter Study on Pneumocystis jiro...

32nd Medical informatics Europe Conference, MIE 2022

作者： Di Meco, Gabriele Mora, Sara Giacobbe, Daniele Roberto Dettori, Silvia Karaiskos, Ilias Bassetti, Matteo Giacomini, Mauro Department of Informatics Bioengineering Robotics and System Engineering Department of Health Sciences University of Genoa Genoa Italy San Martino Polyclinic Hospital-IRCCS Genoa Italy 1st Department of Internal Medicine-Infectious Diseases Hygeia General Hospital Athens Greece

ISBN: (纸本)9781643682846

Pneumocystis jirovecii pneumonia (PJP) is an opportunistic fungal infection that may affect patients with immunosuppression. In order to improve the diagnosis accuracy for PJP, facilitating the collection of data across Europe to reliably assess the performance of diagnostic tests for PJP is essential to improve the care of critically ill patients developing this severe condition. Such large data can be collected thanks to the contribution of several European hospitals in the compilation of a dedicated electronic Case Report Form (eCRF). The main focus of this work is to create an interface with high ergonomics both in the compilation and in the subsequent validation of the records. © 2022 European Federation for Medical informatics (EFMI) and IOS Press.

关键词： Intensive care units

Biodiversity at risk in the SPAMI Pelagos Sanctuary: The impact of marine litter on biota

学校读者我要写书评

暂无评论

Science of the Total Environment 2025年 964卷 178527页

作者： Fossi, Maria Cristina Baini, Matteo Galli, Matteo Tepsich, Paola Grossi, Francesca Concato, Margherita Giani, Dario Rosso, Massimiliano Borroni, Anna Romeo, Teresa Panti, Cristina Department of Physical Earth and Environmental Sciences University of Siena Italy NBFC National Biodiversity Future Center Palermo Italy CIMA Research Foundation Savona17100 Italy Department of Informatics Bioengineering Robotics and Systems Engineering University of Genoa Genova Italy Department of Biology and Evolution of Marine Organisms Stazione Zoologica Anton Dohrn Sicily Marine Centre Messina Italy

The Mediterranean basin is the second largest global biodiversity hotspot in the world, which coexists with a plethora of anthropogenic stress. This study examines the risks that marine litter poses to Mediterranean biodiversity, using the Special Protection Area of Mediterranean Importance (SPAMI) "Pelagos Sanctuary" as a case study. In this paper, a new survey method and data analysis strategies to assess the impact of marine litter, including microplastics, on Mediterranean organisms, is proposed. A total of 23 species, from invertebrates to cetaceans, were analysed using two main monitoring approaches. In the first approach, the core density distribution areas of 11 megafauna species (elasmobranchs, sea turtles, seabirds, and cetaceans) and 4 invertebrate species were evaluated through an extended field survey. Simultaneously, monitoring of floating macro- and micro-litter was performed to be overlapped with the species distribution and abundance. The second monitoring approach assessed the ingestion of marine litter, microplastics and the levels of plastic additives in 10 stranded megafauna species and in 7 invertebrate and fish species. The final data processing, merging the data on biodiversity and marine litter abundance and distribution in the environment, with the data of ingested marine litter, microplastics, and plastic additives in sentinel species, allowed to: a) create risk maps for the species inhabiting the Pelagos Sanctuary, identifying critical areas for biodiversity conservation;b) identify the most exposed species to the risk of marine litter by an "exposure score";c) to select sentinel species for pelagic environment. © 2025 The Authors

关键词： Invertebrates

Persuasive and polite sentences to drive human-robot interaction in smart homes for elderly care 7

学校读者我要写书评

暂无评论

Persuasive and polite sentences to drive human-robot interac...

7th Italian Workshop on Artificial Intelligence and robotics, AIRO 2020

作者： Buoncompagni, Luca Capitanelli, Alessio Cristofanini, Marta Giuni, Antonella Mastrogiovanni, Fulvio Motolese, Carola Nisticò, Andrea Sperindé, Alessandro Zaccaria, Renato Teseo srl Piazza Montano 2a/1 Genoa16126 Italy Department of Informatics Bioengineering Robotics and Systems Engineering University of Genoa Via Opera Pia 13 Genoa16145 Italy

This paper addresses a scenario where a smart environment drives a proactive robot to motivate the elderly in maintaining a healthy lifestyle. The robot is in charge to trigger a vocal interaction with the person by proposing a suggestion that is polite and persuasive. We focus on the representation of sentences based on some attributes and, through spatial and temporal reasoning, we classify them over time. In an ontology, attributes are used to classify sentences and decide if the robot should start an interaction, and with which sentence. We present a preliminary implementation of the system, and we show that it is suitable for undertaking an iterative development process guided by domain experts toward a polite and persuasive virtual coaching. © 2020 Copyright for this paper by its authors.

关键词： Ontology

Learnable Data Augmentation for One-Shot Unsupervised Domain Adaptation

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Carrazco, Julio Ivan Davila Morerio, Pietro Del Bue, Alessio Murino, Vittorio Italian Institute of Technology Genoa Italy Department of Computer Science University of Verona Verona Italy Department of Computer Science and Technology Bioengineering Robotics and Systems Engineering University of Genova Genoa Italy Department of Marine Electrical Electronic and Telecommunications Engineering Robotics and Systems Engineering University of Genova Genoa Italy

This paper presents a classification framework based on learnable data augmentation to tackle the One-Shot Unsupervised Domain Adaptation (OS-UDA) problem. OS-UDA is the most challenging setting in Domain Adaptation, as only one single unlabeled target sample is assumed to be available for model adaptation. Driven by such single sample, our method LearnAug-UDA learns how to augment source data, making it perceptually similar to the target. As a result, a classifier trained on such augmented data will generalize well for the target domain. To achieve this, we designed an encoder-decoder architecture that exploits a perceptual loss and style transfer strategies to augment the source data. Our method achieves state-of-the-art performance on two well-known Domain Adaptation benchmarks, DomainNet and VisDA. The project code is available at https://***/IIT-PAVIS/LearnAug-UDA. © 2023, CC BY.

关键词： Benchmarking

Optimal Design of Agile Jumping Maneuvers for a Single Leg System

学校读者我要写书评

暂无评论

Optimal Design of Agile Jumping Maneuvers for a Single Leg S...

IEEE-RAS International Conference on Humanoid Robots

作者： Andrea Patrizi Francesco Ruscelli Arturo Laurenzi Nikos G. Tsagarakis Humanoids and Human-Centred Mechatronics (HHCM) Istituto Ital-iano di Tecnologia (IIT) Genova Department of Informatics Bioengineering Robotics and Systems En-gineering Università di Genova Genova

In this work we showcase a comprehensive study, developed around our heavy-duty electrically-actuated quadruped leg prototype, focusing on the task of jumping. Our study employs a two-stage pipeline for the design and execution of agile and efficient jumping maneuvers. We combine offline non-linear whole-body Trajectory Optimization (TO) to formulate and solve two distinct non-linear problems, each tackling different phases and performance aspects of the jumping maneuver. The first one jointly addresses the thrust and flight phases and aims at maximizing the achieved jump height, while accounting for several hardware limitations. The second problem deals with the impact and braking phases minimizing the kinetic energy dissipated at the impact, while maximizing the energy regenerated by the actuators during braking. We validate the accuracy of the employed models and the results of the optimizations both in simulation and on the real leg hardware, showing the effectiveness of our approach in maximizing hardware performance and the benefits of exploiting combined impact-energy considerations during the execution of jumping maneuvers.

关键词：

An Ergonomic Role Allocation Framework for Dynamic Human-Robot Collaborative Tasks

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Merlo, Elena Lamon, Edoardo Fusaro, Fabio Lorenzini, Marta Carfì, Alessandro Mastrogiovanni, Fulvio Ajoudani, Arash Human-Robot Interfaces and Interaction Istituto Italiano di Tecnologia Genoa Italy Dept. of Informatics Bioengineering Robotics and Systems Engineering University of Genoa Genoa Italy Dept. of Electronics Information and Bioengineering Politecnico di Milano Milan Italy

By incorporating ergonomics principles into the task allocation processes, human-robot collaboration (HRC) frameworks can favour the prevention of work-related musculoskeletal disorders (WMSDs). In this context, existing offline methodologies do not account for the variability of human actions and states;therefore, planning and dynamically assigning roles in human-robot teams remains an unaddressed challenge. This study aims to create an ergonomic role allocation framework that optimises the HRC, taking into account task features and human state measurements. The presented framework consists of two main modules: the first provides the HRC task model, exploiting AND/OR Graphs (AOG)s, which we adapted to solve the allocation problem;the second module describes the ergonomic risk assessment during task execution through a risk indicator and updates the AOG-related variables to influence future task allocation. The proposed framework can be combined with any time-varying ergonomic risk indicator that evaluates human cognitive and physical burden. In this work, we tested our framework in an assembly scenario, introducing a risk index named Kinematic Wear. The overall framework has been tested with a multi-subject experiment. The task allocation results and subjective evaluations, measured with questionnaires, show that high-risk actions are correctly recognised and not assigned to humans, reducing fatigue and frustration in collaborative tasks. © 2023, CC BY-NC-ND.

关键词： Ergonomics

A Hybrid Training-time and Run-time Defense Against Adversarial Attacks in Modulation Classification

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zhang, Lu Lambotharan, Sangarapillai Zheng, Gan Liao, Guisheng Demontis, Ambra Roli, Fabio School of Electronic Engineering Xidian University Xi’an710071 China Wolfson School of Mechanical Electrical and Manufacturing Engineering Loughborough University Loughborough United Kingdom Dept. of Electrical and Electronic Engineering University of Cagliari Piazza d’Armi Cagliari09123 Italy Department of Informatics Bioengineering Robotics and Systems Engineering University of Genova Genoa16145 Italy

Motivated by the superior performance of deep learning in many applications including computer vision and natural language processing, several recent studies have focused on applying deep neural network for devising future generations of wireless networks. However, several recent works have pointed out that imperceptible and carefully designed adversarial examples (attacks) can significantly deteriorate the classification accuracy. In this paper, we investigate a defense mechanism based on both training-time and run-time defense techniques for protecting machine learning-based radio signal (modulation) classification against adversarial attacks. The training-time defense consists of adversarial training and label smoothing, while the run-time defense employs a support vector machine-based neural rejection (NR). Considering a white-box scenario and real datasets, we demonstrate that our proposed techniques outperform existing state-of-the-art technologies. Copyright © 2024, The Authors. All rights reserved.

关键词： Support vector machines