Privacy-Preserving Federated Learning for Science: Building Sustainable and Trustworthy Foundation Models

This repository is part of a U.S. Department of Energy (DOE) Advanced Scientific Computing Research (ASCR) project focused on developing privacy-preserving federated learning approaches for scientific foundation models. The project aims to advance secure, trustworthy, and sustainable machine learning methodologies for large-scale scientific applications.

ORNL Principal Investigator: Olivera Kotevska, PhD.

Project Duration: 2024-2027

---

Awards

PRESTO: Privacy Recommendation and Security Optimization - R&D 100 Award Winner 2025
PRESTO is an innovative framework designed to provide privacy recommendations and security optimization for machine learning systems. This recognition highlights the project's significant contribution to advancing privacy-preserving technologies in scientific computing.
Learn more about PRESTO

---

Publications

Under review

1. Automated Membership Inference Attacks (MIA): Discovering MIA Signal Computations using Large Language Model (LLM) Agents
   Link

2. SelfGrader: Stable Jailbreak Detection for Large Language Models using Token-Level Logits
   Link

Accepted

1. XMark: Reliable Multi-Bit Watermarking for LLM-Generated Texts The 64th Annual Meeting of the Association for Computational Linguistics
   Link | Code

2. Scalable Federated Learning for Scientific Foundation Models on Leadership-Class Systems
   The 6th Workshop on Machine Learning and Systems (EuroMLSys) co-located with EuroSys '26

3. Traceable Black-box Watermarks for Federated Learning
   The Fourteenth International Conference on Learning Representations (ICLR) 2026
   Link | Code

4. Energy-Efficiency Metrics for Privacy-Preserving Federated Learning with SmartNIC Server Acceleration
   The Sixteenth International Workshop on Accelerators and Hybrid Emerging Systems co-located with 40th IEEE International Parallel and Distributed Processing Symposium

5. Selective Amnesia using Contrastive Subnet Erasure for Class Level Unlearning in Vision Models
   The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026
   Code

6. DP-TwoLevel: Entropy-weighted multi-layer attention for token-level attribution in autoregressive language models
   SPIE Conference on Assurance and Security for AI-enabled Systems 2026

7. Entropy-weighted Multi-layer Attention for Token-level Attribution in Autoregressive Language Models
   SPIE Conference on Assurance and Security for AI-enabled Systems 2026

Published

1. Engineering Privacy at the Edge: A Practical Guide to Differential Privacy in System Architectures
   The 43rd IEEE International Conference on Computer Design (ICCD 2025)
   Link | Code

2. Privacy-Preserving Federated Learning for Science: Challenges and Research Directions
   The 13th IEEE International Conference on Big Data (IEEE BigData 2025)
   Link

3. Balancing Trade-offs: Adaptive Differential Privacy in Interpretable Machine Learning Models
   22nd Annual International Conference on Privacy, Security, and Trust (PST2025)
   Link

4. Optimal Client Sampling in Federated Learning with Client-level Heterogeneous Differential Privacy
   IEEE Internet of Things Journal
   Link | Code

5. MIC-DP: A Scalable Correlation-Aware Differential Privacy Framework for High-Dimensional Data
   IEEE Transactions on Privacy Journal
   Link | Code

6. Privacy Preservation from High-Performance Computing to Autonomous Science [Industrial and Governmental Activities]
   IEEE Computational Intelligence Magazine
   Link

7. OmniFed: A Modular Framework for Configurable Federated Learning from Edge to HPC
   2025 International Conference for High Performance Computing, Networking, Storage and Analysis (SC'25), ExHedtAI: The Workshop on Extreme Heterogeneity and AI Convergence in HPC
   Link | Code