Research Focus
Federated & Privacy-Preserving ML
Research spanning federated learning, decentralized optimization, privacy-aware training, and robust learning under client data heterogeneity.
Privacy-Preserving Machine Learning
Dr Victor Obarafor(PhD)
Building robust and privacy-preserving machine learning systems for decentralized environments.
I work at the intersection of federated learning, privacy-preserving machine learning, robust optimization, and reproducible research systems. My research focuses on understanding instability under data heterogeneity, designing principled aggregation strategies, and building publication-standard experimental pipelines for reliable machine learning research.
Federated Learning
Privacy-Preserving ML
Robust Optimization
Research Systems
Dr Victor Obarafor
Privacy-Preserving Machine Learning Researcher
PhD Researcher
Federated Learning
Privacy-Preserving ML
Robust Aggregation
Reproducible Research Systems
Method Design
Aggregation & Optimization
Designing adaptive aggregation mechanisms that use update structure, drift, and geometry signals to improve robustness in challenging federated settings.
Engineering Strength
Reproducible Research Systems
Building clean, modular, and publication-grade machine learning infrastructure for rigorous experimentation, benchmarking, and technical delivery.
Building reliable machine learning under heterogeneity and privacy constraints
My work sits at the intersection of federated learning, privacy-preserving machine learning, robust optimization, and reproducible research engineering. I’m particularly interested in designing methods and systems that remain stable, interpretable, and technically rigorous in real-world decentralized settings.
Selected Work
Research Projects
Geometry-Aware Federated LoRA
Investigating low-rank adaptation geometry and instability signals in heterogeneous federated learning.
View projectDrift-Aware Adaptive Aggregation
Studying aggregation strategies that respond dynamically to client drift and instability.
View projectPersonalization Depth in Federated Learning
Analyzing client-specific personalization depth and oracle routing headroom in heterogeneous federated environments.
View projectReproducible ML Infrastructure
Engineering publication-grade ML systems for scalable experimentation and research reproducibility.
View projectPublications & Output
Research Direction
Early Dynamics and Stability in Federated Learning under Non-IID Data
Investigating whether early-round training dynamics can predict final convergence behavior and instability in heterogeneous federated learning environments.
Client-Specific Personalization Depth in Federated Learning
Studying personalization depth as a client-level decision problem and evaluating oracle routing headroom under heterogeneous federated distributions.
Ongoing Output
Active work includes aggregation geometry, federated LoRA, adaptive optimization, personalization strategies, and publication-grade ML experimentation systems.