This section highlights my peer-reviewed research publications in AI and cybersecurity, focusing on practical security challenges informed by real-world systems and data. The work emphasizes problem formulation, empirical analysis, and validated insights, without exposing proprietary implementations.
Each publication distills generalizable insights and lessons learned from applied research, contributing to both academic understanding and practical security decision-making.
Browse the publications below or use tags to explore specific research themes such as Federated Learning, Data Quality, Secure ML, and User Behavior Analytics
An empirical study of impulsive clicking behavior using real-world browser history data to enable adaptive, behavior-aware cybersecurity defenses.
A systematic study of how label noise degrades annotation error detection systems and how simple noise-robust training strategies can restore scalability and reliability.
An empirical investigation into annotation errors in cybersecurity benchmarks, revealing how noisy test labels distort model evaluation and mislead progress in threat intelligence.
An empirical study showing how parameter-efficient finetuning enables scalable, privacy-preserving federated learning for large language models with drastically reduced communication costs.
Introducing FedFAME, a data augmentation–free framework for federated semi-supervised learning that remains robust under extreme non-IID data distributions.
A practical utility to extract enriched context sentences for highlighted annotations in PDFs, augmenting standard annotation export tools with full-sentence context for better comprehension and reuse.
An empirical study of impulsive clicking behavior using real-world browser history data to enable adaptive, behavior-aware cybersecurity defenses.
A systematic study of how label noise degrades annotation error detection systems and how simple noise-robust training strategies can restore scalability and reliability.
An empirical investigation into annotation errors in cybersecurity benchmarks, revealing how noisy test labels distort model evaluation and mislead progress in threat intelligence.
An empirical study showing how parameter-efficient finetuning enables scalable, privacy-preserving federated learning for large language models with drastically reduced communication costs.
Introducing FedFAME, a data augmentation–free framework for federated semi-supervised learning that remains robust under extreme non-IID data distributions.
A practical utility to extract enriched context sentences for highlighted annotations in PDFs, augmenting standard annotation export tools with full-sentence context for better comprehension and reuse.