Rickard Brännvall

Senior researcher in machine learning at RISE Research Institutes of Sweden. Background in physics and finance, previously in quantitative analysis and regulatory capital modeling at Citigroup in Tokyo and London. Current research focus:

Quantitative methods for AI privacy risk assessment and regulatory compliance
(GDPR, EU AI Act)
Optimization of privacy–utility trade-offs using privacy-enhancing technologies (PETs)
Statistical robustness of membership inference attacks against AI systems
Co-design of Transformer neural networks efficient under homomorphic encryption;
a quantum-resilient technology for end-to-end secure data processing
Decentralised secure processing environments for health data
Scalable heterogeneous federated learning

News & Updates

April 2026

Paper accepted: FLICS 2026 (Valencia)

“Client-Conditional Federated Learning via Local Training Data Statistics” accepted for presentation at the 2nd International Conference on Federated Learning and Intelligent Computing Systems, Valencia, June 9–12. Published in IEEE proceedings. Extended version with full experimental appendices on arXiv. arXiv (extended) poster

March 2026

New preprint: Exponential-Family Membership Inference

Unifies LiRA, RMIA, and BASE attacks in an exponential-family framework; introduces BaVarIA, a Bayesian approach that improves privacy auditing across shadow-model budgets. arXiv

February 2026

New preprint: Client-Conditional Federated Learning

Conditions a shared global model on locally computed data statistics to handle client heterogeneity — without extra communication cost or additional private information disclosure. arXiv

November 2025

LeakPro Phase II granted

Vinnova funds Phase II of LeakPro, expanding from privacy risk assessment to generative models, regulatory compliance, and PET optimisation. Vinnova poster

All news & updates →

Recent Projects

LeakPro Phase II (Vinnova, 2026–) — Phase II of the LeakPro project, focusing on generative language models, privacy risk audit, model validation for regulatory compliance, and optimisation of privacy-enhancing technologies. Poster
LeakPro (Vinnova) — Development of the LeakPro framework for AI privacy risk assessment, supporting Data Protection Impact Assessments and regulatory compliance under the GDPR and the EU AI Act. arXiv
SARDIN (Vinnova) — A decentralised sandbox platform for secondary use of health data, built on ColonyOS, in preparation for the European Health Data Space (EHDS) in Sweden. The platform sends analysis code to regional edge nodes rather than centralising personal data. Builds on differential privacy and cryptographic aggregation protocols. DiVA report
CryptoBuildingBlocks — Co-design of neural network architectures efficient under fully homomorphic encryption (FHE).
HEIDA (Vinnova) — A research project on secure sharing of healthcare data using homomorphic encryption. Developed software tools and demonstrated privacy-preserving analysis and machine learning with health data. DOI
VINTER (Vinnova Innovation Competition, 2022) — Technical lead for RISE’s winning team in Vinnova’s innovation competition, on an encryption-based solution for secure computation with health data. DOI
NGI Explorer Fellowship (2021) — Research fellowship for a collaboration on homomorphic encryption with Binghamton University, New York.

Papers & Supplementary Material →