Research &
Education
Advancing the frontiers of Byzantine-robust federated learning and autonomous systems through rigorous academic research
Research Output
Peer-reviewed publications in leading conferences and journals
A3: Adaptive Attack-Aware Aggregation for Byzantine-Robust Federated Learning
Springer Information Systems Frontiers
Novel Byzantine-resilient aggregation technique for federated learning achieving 94.99% average accuracy with only 0.71% variance across 5 attack types - significantly outperforming state-of-the-art methods (TrimmedMean: 2.62% variance, Multikrum: 54% variance). Features triple-weighted aggregation mechanism (diversity, confidence, trust) with adaptive strategy selection. Validated on CIC-IDS2017 dataset (2.8M network flows) with comprehensive statistical analysis.
The Practitioner's Dilemma: A Critical Review of Trade-offs between Privacy, Efficiency, and Model Utility in Federated Learning Systems
Springer Computing
Comprehensive PRISMA-style systematic review quantifying the fundamental trilemma in federated learning: privacy, efficiency, and utility cannot be simultaneously maximized. Meta-analysis of 100+ papers and production systems (2022-2025). Key findings: Strong differential privacy (ε<1) causes 5-15% accuracy loss; Cryptographic security has 2-5× overhead; Communication optimization achieves 90%+ bandwidth reduction. Introduces five-dimensional assessment framework for Pareto-optimal FL designs.
Education
Graduate Research Assistant
Federated Learning & Privacy-Preserving ML
Highlights
- 🔬 Designed A3 aggregation algorithm achieving 94.99% accuracy with only 0.71% variance across 5 Byzantine attack types
- 📊 Outperformed state-of-the-art methods: TrimmedMean (2.62% variance), Multikrum (54% variance)
- 📄 Produced two research papers submitted to peer-reviewed venues (Springer journals, Under Review)
- 📚 Conducted systematic review analyzing 100+ federated learning papers (2022-2025)
- 🎯 Quantified privacy-efficiency-utility trilemma in production FL systems
- 💻 Validated on CIC-IDS2017 dataset with 2.8M network flow records
Master of Science
Robotics & Artificial Intelligence
Thesis
A3: Adaptive Attack-Aware Aggregation for Byzantine-Robust Federated Learning
Highlights
- Developed novel Byzantine-resilient federated learning algorithm achieving state-of-the-art performance (94.99% accuracy, 0.71% variance across 5 attack types)
- Two research papers submitted to peer-reviewed venues (Under Review, 2025)
- Conducted comprehensive systematic review analyzing 100+ federated learning papers
- Research directly applied to production security systems at Cybersilo
- Supervisor: Dr. Shahbaz Khan
Bachelor of Science
Computer Science
Highlights
- First Class with Distinction
- Gold Medal for achieving highest CGPA in graduating class
- Comprehensive foundation in software engineering, algorithms, and cybersecurity
- Consistent academic excellence throughout the program with hands-on project experience
Research Interests
My research focuses on making distributed machine learning systems robust, secure, and efficient. I'm particularly interested in Byzantine-robust algorithms, federated learning, and their applications in autonomous systems and edge computing environments.