Privacy in collaborative learning
Inference attacks, decentralized recommendation, and confidential learning.
About Me
I am a tenured research scientist at Inria, in the Privatics team in Lyon. I study how data and AI systems can reveal sensitive information about people. My goal is to limit those risks while keeping systems useful.
My research ranges from privacy attacks on decentralized learning to anonymization of speech, mobility data, and medical images. More recently, I have been exploring privacy in agentic AI.
Anonymization & evaluation
Measuring disclosure risks in speech, mobility data, and medical imaging.
Background & education
Before joining Privatics, I worked on privacy-preserving federated learning at ENTPE (2021–2022), within the DSVD chair in partnership with Renault Group, and on private machine learning for speech processing at Inria (2019–2021), in the Magnet team.
PhD in Computer Science · 
Computer Science Engineering degree · News
Yacine Belal wins the GDR RSD thesis award
Congratulations to Yacine Belal, whom I co-supervised with Sonia Ben Mokhtar, on receiving the 2026 GDR RSD thesis award for his work on trustworthy collaborative learning.
INESIA funding for privacy in agentic AI
Our project on evaluating and enforcing privacy in agentic AI has received INESIA funding. I am co-PI with Sonia Ben Mokhtar and Raouf Kerkouche. The project will support the PhD of Sami Kouzeiha, joining Privatics in 2026.
Publications
Research papers and preprints.
Beyond the Mask: The Illusion of Privacy in Defaced Brain MRI
Imen Bajar, Mohamed Maouche, Carole Frindel.
Differentially Private and Decentralized Randomized Power Method.
J. Nicolas, C. Sabater, M. Maouche, S. Ben Mokhtar, M. Coates.
Exposing the Vulnerability of Decentralized Learning to Membership Inference Attacks Through the Lens of Graph Mixing
Ousmane Touat, Jezekael Brunon, Yacine Belal, Julien Nicolas, César Sabater, et al.
GRANITE: a Byzantine-Resilient Dynamic Gossip Learning Framework.
Y. Belal, S. Ben Mokhtar, M. Maouche, A. Simonet-Boulogne.
More publications & preprints
Inferring Communities of Interest in Collaborative Learning-based Recommender Systems
Yacine Belal, Mohamed Maouche, Sonia Ben Mokhtar, Anthony Simonet-Boulogne.
PriviRec: Confidential and Decentralized Graph Filtering for Recommender Systems
Julien Nicolas, César Sabater, Mohamed Maouche, Mark Coates, Sonia Ben Mokhtar.
Secure Federated Graph-Filtering for Recommender Systems.
J Nicolas, C Sabater, M Maouche, S Ben Mokhtar, M Coates.
Synthetic Data: Generate Avatar Data on Demand.
T Lebrun, L Béziaud, T Allard, A Boutet, S Gambs, M Maouche.
Differentially Private Speaker Anonymization.
A. Shamsabadi, B. Srivastava, A. Bellet, N. Vauquier, E. Vincent, M. Maouche, M. Tommasi, N. Papernot.
Towards an evolution in the characterization of the risk of re-identification of medical images.
A Boutet, C Frindel, M Maouche.
Enhancing Speech Privacy with Slicing.
M. Maouche, B. Srivastava, N. Vauquier, A. Bellet, M. Tommasi, E. Vincent.
Privacy and utility of x-vector based speaker anonymization.
B. Srivastava, M. Maouche, Md. Sahidullah, E. Vincent, A. Bellet, M. Tommasi, N. Tomashenko, X. Wang, E. Vincent, J. Yamagishi.
The VoicePrivacy 2020 Challenge: Results and findings.
N. Tomashenko, X. Wang, E. Vincent, J. Patino, B. Srivastava, PG. Noé, A. Nautsch, N. Evans, J. Yamagishi, B. O'Brien, A. Chanclu, JF. Bonastre, M. Todisco, M. Maouche.
A comparative study of speech anonymization metrics.
M. Maouche, B. Srivastava, N. Vauquier, A. Bellet, M. Tommasi, E. Vincent.
Design Choices for X-vector Based Speaker Anonymization.
B. Srivastava, N. Tomashenko, X. Wang, E. Vincent, J. Yamagishi, M. Maouche, A. Bellet, M. Tommasi.
MooD: MObility Data Privacy as Orphan Disease -Experimentation and Deployment Paper.
B. Khalfoun, M. Maouche, S. Ben Mokhtar and S. Bouchenak.
ACCIO: How to Make Location Privacy Experimentation Open and Easy.
V. Primault, M. Maouche, A. Boutet, S. Ben Mokhtar, S. Bouchenak, L. Brunie.
HMC: Robust Privacy Protection of Mobility Data against Multiple Re-Identification Attacks.
M. Maouche, S. Ben Mokhtar and S. Bouchenak.
AP-Attack: A novel Re-identification Attack On Mobility Datasets.
M. Maouche, S. Ben Mokhtar and S. Bouchenak.
Projects
Research projects I lead or participate in.
Current & upcoming projects
2026–2029 · Upcoming
Evaluating and enforcing privacy in agentic AI
Program: INESIA · AI Evaluation
Role: Co-PI with Sonia Ben Mokhtar and Raouf Kerkouche.
Focus: Assessing privacy risks and developing protections for personal data across AI agents, tools, and APIs.
2025–2029
RAIDAC+ · Responsible AI: Design, Regulation, and Conformity
Program: MIAI Cluster
Role: Member.
Focus: Responsible AI at the intersection of privacy, fairness, transparency, and regulation.
REDEEM · Resilient, decentralized, and privacy-preserving machine learning
Program: PEPR IA
Role: Non-funded member.
Focus: Privacy, robustness, and resilience in decentralized machine learning.
IPoP · Interdisciplinary Project on Privacy
Program: PEPR Cybersécurité
Role: Member.
Focus: Understanding privacy threats and developing personal-data protection methods for emerging digital services.
2023–2026
Secure, private, and multimodal authentication
Program: CIFRE · OpenSezam
Role: PI with Vincent Roca.
Focus: Privacy-preserving biometric authentication and continuous detection of anomalies and intrusions.
Past projects
2024–2025
Riviera · CITI–CREATIS
Program: Inter-laboratory project · Lyon Computer Science Federation (FIL)
Role: PI with Carole Frindel.
Focus: Research collaboration between CITI and CREATIS.
Student supervision
PhD co-supervision in Privatics
Sami Kouzeiha
Incoming · 2026Evaluating and enforcing privacy in agentic AI: assessing privacy risks and developing protections for personal data across agents, tools, and APIs.
Inria / Privatics · INESIA project · With Sonia Ben Mokhtar and Raouf Kerkouche
Imen Bajar
Since 2025Privacy and re-identification risks in brain MRI.
Inria / Privatics · Joint PhD with ÉTS Montréal · MIAI Cluster · With Carole Frindel, Héber Hwang Arcolezi, and Antoine Boutet
Alix Ntoutoume Nzame
Since 2023Secure, private, and multimodal authentication.
CIFRE PhD with OpenSezam · With Vincent Roca and Pierre-Guillaume Gourio-Jewell
PhD co-supervision outside Privatics
I also co-supervise PhD students in the DRIM team at LIRIS.
Ousmane Touat
Since 2024Byzantine resilience and robustness to external adversaries in decentralized learning.
Julien Nicolas
Since 2023Subspace methods for privacy, decentralization, and noise reduction in machine learning.
LIRIS / DRIM · Joint PhD, INSA Lyon / McGill University · REDEEM · With Sonia Ben Mokhtar and Mark Coates
Former PhD students
Yacine Belal
2022–2025 · Defended in 2025Trustworthy collaborative learning: personalization, privacy, and robustness at the edge.
Academic service
Journal reviewing
Conference & workshop organization
- Co-General Chair, Atelier pour la Protection de la Vie Privée (APVP 2024)
- Local organizing committee, SRDS 2019
- Organization of the Lyon Computer Science Federation (FIL) 10th anniversary
- Organization of the FIL federated learning workshop
Research evaluation
- Committee member, ANR TSIA 2025 call
- Reviewer, ANR AAPG 2023 (CE23)
Institutional responsibilities
- CITI laboratory council (since 2025)
- Communication officer, Lyon Computer Science Federation (since 2024)
Teaching
Current teaching
CSC — Cryptography & security
SERE — Security & networks
Previous teaching
Dimensionality reduction
Université de Lille · 2020–2021
Course lead for first-year Master’s students (M1) in machine learning. Topics: PCA, LDA, and autoencoders.
Teaching during my PhD · INSA Lyon · 2016–2019
- Security · 4th year
SQL injection, XSS, cryptanalysis, and firewalls.
- Computer science · 1st and 2nd year
Algorithms, object-oriented programming, dynamic data structures, and graphical interfaces in Java.
- Systems · 3rd year
C, scheduling, concurrency, and memory management.
- Web data · 3rd year
XML, XPath, XSL, XQuery, JSON, and MongoDB.
- Semantic web · 4th year
RDF and SPARQL.
- Human–computer interaction · 4th year
Android project.
Software
Research tools
Anonymization Metrics
Evaluate anonymization quality and re-identification risk. Integrated into the VoicePrivacy 2020 Challenge.
HMC
Heat Map Confusion: protect mobility traces against re-identification while preserving utility.
SFERA
A toolkit for experiments with re-identification attacks on mobility traces.
Software from collaborative publications
Code accompanying publications I co-authored.
Brain MRI privacy evaluation
Contributor: Imen Bajar
Feature extraction, patient linkage, and attribute-inference experiments on brain MRI representations.
Paper · 2026 · Anonymous repository
Giskard
Contributor: Ousmane Touat
Experiments and a Rust proof of concept for confidential, Byzantine-robust decentralized aggregation.
Paper · 2026 · Anonymous repository
GRANITE
Contributor: Yacine Belal
A GossiPy-based framework for Byzantine-resilient gossip learning over dynamic graphs.
Paper · 2025 · Anonymous repository
Decentralized learning privacy evaluation
Contributor: Ousmane Touat
Reproduce membership-inference experiments across communication graphs and model-mixing strategies.
Paper · 2025 · Anonymous repository
ACCIO
Contributor: Vincent Primault
A scientific workflow framework for reproducible location-privacy experiments.
Paper · 2018 · Archived repository