Machine learning to augment shared knowledge in federated privacy-preserving scenarios

Informazioni relative al progetto

MUSKETEER

ID dell’accordo di sovvenzione: 824988

DOI

10.3030/824988

Progetto chiuso

Data della firma CE 23 Novembre 2018

Data di avvio 1 Dicembre 2018

Data di completamento 30 Novembre 2021

Finanziato da

• INDUSTRIAL LEADERSHIP - Leadership in enabling and industrial technologies - Information and Communication Technologies (ICT)

Costo totale

€ 4 380 346,25

Contributo UE

€ 4 380 335,00

Coordinato da

IBM IRELAND LIMITED
Ireland

CORDIS fornisce collegamenti ai risultati finali pubblici e alle pubblicazioni dei progetti ORIZZONTE.

I link ai risultati e alle pubblicazioni dei progetti del 7° PQ, così come i link ad alcuni tipi di risultati specifici come dataset e software, sono recuperati dinamicamente da .OpenAIRE .

Risultati finali

Documents, reports (19)

Project communication and engagement activities

Details for communication and engagement activities and materials along with the time line and success indicators. It includes a record of communication activities that have been undertaken during the first half of the project, and those planned for the second period.

Architecture design – Initial version

A document describing the first version of the platform architecture, how the architecture meets the (initial) requirements of the federated and privacy-preserving machine learning services designed in WP 4.1, how it addresses the (initial) user stories developed in WP7, and how it aligns with existing Industrial Data Platform standards.

Industrial and technical requirements

Report containing an exhaustive list of domain-specific and business requirements coming from the two scenarios and from other complementary domains, and assessment of available datasets. Report containing a complete specification of technical requirements to drive technical developments in WPs 3,4,5,6 and WP7 integration.

Use case execution and KPI evaluation in the Health domain

This document reports the description of the Health pilot setup and execution together with the evaluation of the KPIs in order to assess the usage of the platform

Key performance indicators selection and definition

Detailed description of the technical and domain business-specific KPIs that will be used for validating the MUSKETEER data platform.

Scientific dissemination activities

Details for scientific dissemination activities and materials along with the time line and success indicators It includes a record of activities related to scientific dissemination that have been undertaken during the he project

Use case execution and KPI evaluation in the Smart Manufacturing domain

This document reports the description of the Smart Manufacturing pilot setup and execution together with the evaluation of the KPIs in order to assess the usage of the platform

Client connectors’ architecture design – Initial version

A document describing the main functionalities of the client connector. It will contain the design of the component and how it interact with services at server side. This is the first version of the report.

Architecture design – Final version

A document describing the final version of the platform architecture, how it meets the final requirements of the federated and privacy-preserving machine learning services designed in WP 4.1, how it addresses the final user stories developed in WP7, how it supports incorporating active security measures against adversarial attacks (data poisoning, evasion), and how it aligns with existing Industrial Data Platform standards.

Investigative overview of targeted architecture and algorithms

A technical report containing the final structure of the privacy operation modes describing how every algorithm will operate over these modes and the details about the SW architecture and design patterns that will facilitate future extensions of the SW library.

Threat analysis for federated machine learning algorithms

A report describing the main threats and vulnerabilities that may be present in federated machine learning algorithms considering both, attacks at training and test time and defining requirements for the design, deployment and testing of federated machine learning algorithms. This report would also form a strong basis from which governance and or legislative input could be drawn.

Dissemination and communication plan

his task aims to detail the strategy for dissemination of project results through appropriate channels and during various stages of the project. The underlying goal of dissemination activities will to ensure public awareness of the project and promote interest. This task will ensure practical dissemination using the different instruments mentioned in WP8.

Client connectors architecture design – Final version

This report represent the updated and final version of D71 It capture the final status that might be adapted during the implementation and phase

Scalability of machine learning algorithms over every POMs

A report describing the scalability of every algorithm as a function of the POM the HW available the number of data providers and the volume of training data It will also contain recommendations to select the best algorithm and configuration for a given privacy restriction

Key performance indicators selection and definition - final version

An update version of D23

Assessment Framework design and specification

A document describing the main common evaluation framework. It will contain the design of the different tests and datasets that will be used in the evaluation, as well as the merit performance measurements to be obtained.

Community engagement and technology transfer activities

Details for the community engagement and technology transfer strategy for the project The intermediate report includes a record of activities related to community creation and engagement and technology transfer developed in the course of the project

Evaluation and impact assessment

A report which details the gains associated with the MUSKETEER solution using quantitative information and which identifies areas for further improvement and investment

Security of federated machine learning algorithms

A document describing how confident is the accuracy of a Machine Learning algorithm when we consider attacks and detection strategies

Other (1)

Project website and communication material

MUSKETEER public website, to be active and regularly updated during the whole project and maintained for 1 year following the project’s completion. The MUSKETEER factsheet will be an early leaflet for dissemination and communication purposes, including the most relevant information of the project in a nutshell, and will be available from the very beginning.

Demonstrators, pilots, prototypes (8)

Machine Learning Algorithms over Federated Operation Modes – Final version

This deliverable will complete the library with deep learning algorithms and unsupervised learning algorithms in the privacy operation models those predictive models where datasets never leave the installations of every client

First prototype of the MUSKETEER platform

A demonstration (and report) of a first prototype of the MUSKETEER platform, demonstrating the end-to-end execution of data sharing and federated machine learning for synthetic data and at least one use case, supporting privacy operating modes POM1-POM3. Includes demonstration of basic dashboard reporting.

First prototype of the MUSKETEER client connectors

This deliverable represents the first version of the implementation of the artefact designed in T21 together with the accompanying documentation

Pre-processing, normalization, data alignment and data value estimation algorithms – Final version

This deliverable is in the form of software will present Version 1 of the library covering a set of preprocessing steps needed before using the machine learning algorithms Models for the estimation of the data value will also be provided

Pre-processing, normalization, data alignment and data value estimation algorithms – Initial version

This deliverable is in the form of software will present Version 0 of the library covering a set of pre-processing steps needed before using the machine learning algorithms. Models for the estimation of the data value will also be provided.

Final prototype of the MUSKETEER platform

A demonstration and report of a complete instantiation of the MUSKETEER platform demonstrating the endtoend execution of data sharing and federated machine learning for all use case and all privacy operating modes Includes demonstration of complete dashboard reporting

Machine Learning Algorithms over Semi Honest Operation Modes – Final version

This deliverable will complete the library with more elaborate models unsupervised Deep Learning etc whenever computational overload is affordable in the semihonest scenarios where the operations are carried out using encrypted data as input every data provider never sees data from other users and the resulting model is private only known by the platform

Final prototype of the MUSKETEER client connectors

This deliverable represents the final version of the implementation of the artefact designed in T21 together with the accompanying documentation

Pubblicazioni

Peer reviewed articles (5)

Budget Distributed Support Vector Machine for Non-ID Federated Learning Scenarios

Autori: A. Navia-Vázquez, R. Díaz-Morales, M. Fernández-Díaz
Pubblicato in: ACM Transactions on Intelligent Systems and Technology. Special Numero on Federated Learning: Algorithms, Systems, and Applications (under review), 2022, ISSN 2157-6904
Editore: Association for Computing Machinery (ACM)

Recycling weak labels for multiclass classification

Autori: Miquel Perello-Nieto; Raul Santos-Rodriguez; Dario Garcia-Garcia; Jesús Cid-Sueiro
Pubblicato in: Journal of Neurocomputing, 2020, Pagina/e 206-215, ISSN 0925-2312
Editore: Elsevier BV
DOI: 10.1016/j.neucom.2020.03.002

"""A Priori"" Shapley Data Value Estimation for Data Monetization in Federated Learning"

Autori: A. Navia-Vázquez, J. Cid-Sueiro and M.A. Vázquez
Pubblicato in: Journal of Neurocomputing (under review), 2022, ISSN 0925-2312
Editore: Elsevier BV

Data protection by design in AI. The case of federated learning

Autori: S. Rossello, S., L. Muñoz-González, and R. Díaz Morales
Pubblicato in: Computerrecht, 2021, ISSN 0771-7784
Editore: Deventer : Kluwer Academic Publishers

First vs Second Order Doubly Confidential Distributed Learning for Logistic Regression

Autori: A. Navia-Vázquez, M.A. Vázquez,  and J. Cid-Sueiro
Pubblicato in: IEEE Transactions on Parallel and Distributed Systems (under review), 2022, ISSN 1045-9219
Editore: Institute of Electrical and Electronics Engineers

Monographic books (3)

Increasing trust within an ecosystem with Federated learning

Autori: S. Bonura, D. Dalle Carbonare, R. Díaz-Morales, A. Navia-Vázquez, M. Purcell and S. Rossello
Pubblicato in: BDVA Book Chapter, 2021
Editore: BDVA

Data protection in the era of artificial intelligence. Trends, existing solutions and recommendations for  privacy-preserving technologies

Autori: T. Timan, Z. Mann (Eds.), R. Araujo, A. Crespo-García, A. Farkash, A. Garnier, A. Vivian-Kiousi, P. Koster, A. Kung, G. Livraga, R. Díaz-Morales, M. Önen, A. Pa-lomares, A. Navia-Vázquez, A. Metzger (contributors).
Pubblicato in: BDVA position paper, 2019
Editore: BDVA

Big Data challenges in Smart Manufacturing Industry

Autori: Gusmeroli S., Dalle Carbonare D. (eds).
Pubblicato in: BDVA Whitepaper, 2020
Editore: BDVA

Book chapters (2)

Privacy Preserving Technologies for Trusted Data Spaces

Autori: S. Bonura, D. Dalle Carbonare, R. Díaz-Morales, M. Fernández-Díaz, L. Morabi-to, L. Muñoz-González,  C. Napione, A. Navia-Vázquez, M. Purcell
Pubblicato in: BDVA Book Chapter, 2021
Editore: BDVA

Security and Robustness in Federated Machine Learning

Autori: A. Rawat, G. Zizzo, M. Zaid Hameed, and L. Muñoz-González
Pubblicato in: Federated Learning: A Comprehensive Overview of Methods and Applications, 2021
Editore: Springer

Conference proceedings (4)

Privacy-Preserving Distributed Support Vector Machines

Autori: Bottoni, Simone, Stefano Braghin, Theodora Brisimi, and Alberto Trombetta
Pubblicato in: Heterogeneous Data Management, Polystores, and Analytics for Healthcare: VLDB Workshops, Poly 2021 and DMAH 2021, 2021, Pagina/e 85-102
Editore: Springer

FAT: Federated Adversarial Training

Autori: G. Zizzo, A. Rawat, M. Sinn, B. Buesser
Pubblicato in: NeurIPS 2020 Workshop on Scalability, Privacy, and Security in Federated Learning (SpicyFL), 2020
Editore: Neural Information Processing Systems

Defending against poisoning attacks in online learning settings

Autori: Greg Collinge, Emil C Lupu, Luis Muñoz-González
Pubblicato in: ESANN 2019 proceedings, European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, Numero Year., 2019, Pagina/e 43-48, ISBN 978-287-587-065-0
Editore: ESANN

Double Confidential Federated Machine Learning Logistic Regression for Industrial Data Platforms

Autori: A. Navia-Vazquez M. Vazquez-Lopez J. Cid-Sueiro
Pubblicato in: FL-ICML 2020 : International Workshop on Federated Learning for User Privacy and Data Confidentiality in Conjunction with ICML 2020, 2020
Editore: ICML 2020

Diritti di proprietà intellettuale

Other (2)

interproccesing messaging services.

Numero candidatura/pubblicazione: IBM/pycloudmessenger https://github.com/IBM/pycloudmessenger
Data: 2020-01-07
Candidato/i: IBM IRELAND LIMITED

Musketeer-Client prototype application.

Numero candidatura/pubblicazione: Musketeer-Client https://github.com/IBM/Musketeer-Client
Data: 2019-11-01
Candidato/i: IBM IRELAND LIMITED